Datasets:

CleverShovel

/

paper_reviews

like 1

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DEEP LEARNING METHODS FOR INVERSE PROBLEMS Review round: 1 Reviewer: 2

Basic reporting: The main innovation of this manuscript is to compare the robustness of the three categories, and report a statistical analysis of their differences, which is not novel enough. The method of this paper is not innovative enough. In fact, most of the work is done by combining other people’s methods. Authors need to highlight their innovative contributions. Although the article is clear on the whole, I still hope you pay attention to the clarity of the article. There are numerous places in the article that say the same thing over and over again. I hope you can check and delete the redundant parts. Pay attention to consistency to avoid confusion. Another obvious problem with this paper is the lack of sufficient experimentation to demonstrate the validity and applicability of the three methods. Experimental design: I recommend you to read this manuscript: " Quadratic Residual Networks: A New Class of Neural Networks for Solving Forward and Inverse Problems in Physics Involving PDEs"Perhaps the author can find inspiration by reading more literature in this field to further optimize this paper. Please check the normalization and accuracy of the figure in the paper carefully. Many figures are far away from the analysis paragraphs. I hope they can be adjusted. The literature on deep learning methods for solving inverse problems was classified into three categories, of which was evaluated on sample inverse problems of different types. Why not try more categories? You can make some experiments on more categories. Validity of the findings: One obvious problem with this paper is the lack of enough experimentation to demonstrate the validity and applicability of the proposed method. The author needs to do more experiments with more angles and show them in this paper. Additional comments: No comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DEEP LEARNING METHODS FOR INVERSE PROBLEMS Review round: 2 Reviewer: 1

Basic reporting: According to the author's response, the author solved related problems, revised the grammatical errors in the revised paper, and clearly expounded the contribution of this paper. Experimental design: No comment. Validity of the findings: No comment. Additional comments: According to the author's response, the author solved related problems, revised the grammatical errors in the revised paper, and clearly expounded the contribution of this paper.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DEEP LEARNING METHODS FOR INVERSE PROBLEMS Review round: 2 Reviewer: 2

Basic reporting: The article meet our standards. The article include sufficient introduction and background to demonstrate how the work fits into the broader field of knowledge. Most of the problems have been solved. Experimental design: In last version the article is the lack of sufficient experimentation to demonstrate the validity and applicability of the three methods. The author repeated the training procedures 5-10 times and reported the resulting statistics and aggregated the overall results in a separate “Discussion” section to provide a insights regarding issues facing all of the types of inverse problems in their experiments. Validity of the findings: The author maintain that the relative merits and the relative applicability of the various methods is not particularly understood, and it is precisely this relative performance which is the purpose and contribution of this paper. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: BUILDING A KOREAN MORPHOLOGICAL ANALYZER USING TWO KOREAN BERT MODELS Review round: 1 Reviewer: 1

Basic reporting: In this paper, a Korean morphological analyzer (KMA) is designed to output a sequence of a single morpheme and its Part-of-speech (POS) that follows the morpheme. As the lengths of the input and output are different, Korean morphological analysis can be defined as an encoder-decoder problem.In this work,the author adopt the Transformer architecture to implement a KMA. The author's contributions are as follows: 1. In this work, the author utilized two kinds of BERT models to initialize Transformer. The wBERT is pre-trained with Korean raw sentences and the mBERT is pre-trained with morphologically analyzed sentences consisting of morphemes and POS tags.And the author initialized an encoder of Transformer with wBERT and a decoder of Transformer with mBERT. 2. The author found the optimal number of layers for the encoder and decoder to achieve the best performance of KMA. 3. The author summarized the experimental results that evaluate the impact of initializing wBERT and mBERT. 4. The author measured changes in accuracy in the models as the size of training dataset decreased from 100% to 10%. Experimental design: Strengths and Weaknesses 1. In the experiment of finding the optimal number of layers for the decoder,the encoder is initialized with wBERT while the decoder is randomly initialized.But there seems to be no reason to do so, so I think the author should explain the specific reasons for it. 2. In the experiment of finding the optimal number of layers for the encoder, the author fixed the number of layers in the decoder at 4. So the conclusion of this experiment is based on the premise that decoder has 4 layers, and it doesn't mean that the optimal number of layers is that(maybe when the decoder has other layers). 3. In table 7, when wBERT and mBERT are used to initialize the model and the data set is 10%, the data in this cell should be 95.23% instead of 98.23%. 4. In table 8, the comparation of F1 values in different models is meaningless, because they used different test datasets. Validity of the findings: Innovation This paper is the first attempt to initialize both an encoder and decoder for Transformer with two kinds of Korean BERT. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: BUILDING A KOREAN MORPHOLOGICAL ANALYZER USING TWO KOREAN BERT MODELS Review round: 1 Reviewer: 2

Basic reporting: The structure of the overall article is very logical, but there are some problems in the details. （1）In the relevant work, the analysis of the relevant research on Korean at this stage is not very clear, and there is no clear analysis of the research deficiencies at this stage and the innovation of this research. It is necessary to explain the enlightenment or reference of the past literature research to this research. （2）The paper does not explain how many words the longest text and the shortest text of the experimental data are respectively. It is necessary to further explain whether the algorithm is suitable for long text or short text. According to the experimental data set in the paper, it seems that it is only suitable for the training of short text, so we need to focus on this aspect again. Experimental design: （1）The specific process description of the experiment is a little simple, such as pretreatment, model experiment process, etc., which should be connected with the subsequent evaluation. （2） In the introduction section, in the second paragraph, in order to verify the problem of "the output sequence of a KMA differs from a raw input sequence in both length and surface form.", I think that not all sentences are the same problem, please add which one kind of sentence has this problem. （3） In the section of SURVEY ON KOREAN MORPHOLOGICAL ANALYZER, it only proves the rationality of the transform structure, but does not prove the rationality of "initialize both an encoder and decoder for Transformer with two kinds of Korean BERT.", please add relevant explanations. Validity of the findings: （1） The main contribution of this article is "train the morphological analyzer faster and with less training data".In Table 7, the size of the data set is divided into 10%, 30%, 50%, 70%, 100%. The F1 score first drops and then rises, but the F1 value of the data set at 100% is significantly better than the F1 value of 10%, please Explain why the model proposed in this paper requires a smaller amount of data set. （2）For the main contribution 2 "find the most appropriate number of layers in the BERT models for a Korean morphological analyzer"&" we observed that the accuracy of the Korean morphological analyzer is highest when it has four layers in the decoder and 12 layers in the encoder."Compare the experiment, whether the data set in other fields is applicable to the results of this article. Additional comments: No comment.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: BUILDING A KOREAN MORPHOLOGICAL ANALYZER USING TWO KOREAN BERT MODELS Review round: 1 Reviewer: 3

Basic reporting: some details are missing in this version. the reviewer believes that the introduction is overlength. Experimental design: some mismatched data in the experiment section. Validity of the findings: the proposed model is worked in a certain dataset. more experiments are required to validate the proposed model. Additional comments: 1 the author should clarify how the encoder and decoder are initialized with the pre-trained BERT model. 2 it is really necessary to define the classification between the word separator token and morpheme tokens as a multi-task learning? The authors should provide sufficient rationale. 3 more experiments should be conducted to validate the multi-task learning in this work. 4 In section experiment, the data (line 180) is mismatched with that of in Tab.7. 5 more details should be provided to improve the readability of this work, such as the initialization.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: BUILDING A KOREAN MORPHOLOGICAL ANALYZER USING TWO KOREAN BERT MODELS Review round: 2 Reviewer: 1

Basic reporting: Compared with the previous version, the author has made some supplements and modifications to the experiment, but there are still some problems in the details. (1)When exploring the influence of training set size on experimental results, there should be experiments with full-layers for both encoder and decoder, as in the previous experiments. From my point of view,  I suggest a few modifications are still needed before publication. Experimental design: (1)In the experiment of finding the optimal number of encoder and decoder layers , I think the two experiments made by the author can be combined into one, that is, the optimal combination of layers for encoder and decoder can be determined by the first experiment alone. Validity of the findings: Compared with the previous version, the author has made some supplements and modifications to the experiment, but there are still some problems in the details. Additional comments: (2)In table 6, the fourth column heading should be '> 100 tokens' not '< 100 tokens'.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PREDICTING DEFECTS IN IMBALANCED DATA USING RESAMPLING METHODS: AN EMPIRICAL INVESTIGATION Review round: 1 Reviewer: 1

Basic reporting: The structure or the organization of the paper is not very clear to me. I understand that the authors of the paper are trying to investigate the impact of class reblancers on defect prediction models. However, I am not sure why the authors investigate the features that are selected consistently by a CFS method. They do not use this finding anywhere else. Furthermore, they assert that using all the features might lead to overfitting of the models. However, the studied defect datasets do not have more than 30 features! So I would recommend that the authors explain why they study a feature selection method along with class rebalancing if there is no connection between the findings. I recommend the authors to replace table 6, 7, 8,9 ,10, 11 with side-by-side boxplots for better readability Experimental design: several prior studies (e.g., https://arxiv.org/abs/1609.01759) find that tuning the machine learning classifiers is extremely important for the defect prediction models to perform well. In this case, it could particularly be an important confounder that affects the results. I would recommend the authors to tune the models that they use in their study I am not clear about how the statistical tests are used. For instance, the authors say they used a Friedman test followed by a Wilcoxon test to rank the performance of the classifiers. However, a friedman test to my knowledge doesn’t generate ranking and a Wilcoxon test is only a parwise comparison. So I am not sure how the authors arrive at the ranking. I would urge the authors to explain the statistical tests that they used to arrive at their results a little clearly Similarly, to rank the results of Friedman test, a post-hoc nemenyi test is the typical suggestion, rather than the Wilcoxon. Hence I would suggest the authors to use that (e.g., you can follow this paper: https://arxiv.org/abs/1707.09281, https://joss.theoj.org/papers/10.21105/joss.02173) If Wilcoxon is used to compare multiple pairs, you need to use a p-value correction like a Bonferroni correction for the results to be reliable. Validity of the findings: Major concern: Some of the key related works are missing. For instance, a recent TSE (https://arxiv.org/abs/1801.10269) and an ICSE (https://arxiv.org/pdf/1705.03697.pdf) study through a comprehensive study finds that SMOTE is indeed better than other class rebalancing methods. These studies are not discussed. I would recommend the authors to consider these studies and position their findings in the context of these studies. More along my previous point, the aforementioned papers assert that tuning the SMOTE is important for the benefits of SMOTE to shine through. Tantithamthavorn et. al. (https://arxiv.org/abs/1801.10269) in particular find that tuned SMOTE helps defect prediction models perform better than unbalanced classifier. They also compare several class rebalancing methods (though not as comprehensive). So I would like the authors to position their findings in the context of this paper. Therefore, I would suggest the authors to include the datasets used by Tantithamthavorn et. al. also in their studies to see if their findings agree with Tantithamthavorn et. al. or if they could refute them. Because the statistical issues outlined earlier I am not sure if the results are reliable. Additional comments: 1. The paper compares an impressive array of class rebalancing methods and defection prediction classifiers. 2. The paper comprehensively tries to estimate the impact of the class rebalancing method by experimenting with multiple configurations.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PREDICTING DEFECTS IN IMBALANCED DATA USING RESAMPLING METHODS: AN EMPIRICAL INVESTIGATION Review round: 1 Reviewer: 2

Basic reporting: - In general, the manuscript is clear and well-written, however it lacks a Discussion section which would be useful to present a general overview of the findings and discuss the results on a higher level. - The study can also benefit from a small description of the research questions under study to make it more clear to the reader. - While Figure 1 might be comprehensive, it would be better to include a small description which might make it easier for the reader to understand the full experimental process of the study. - The paper contains grammatical/spelling errors that can be fixed with a detailed proofread, some of them are listed below: Line 25: “…softwares…” --> “…software…” Line 37: “…with uncovering the probable…” --> “…with uncovering probable…” Line 44: “…becomes a difficult task.” --> “…become difficult tasks.” Line 52: “…trained on the similar…” --> “…trained on similar…” Line 65: “…it is widely…” --> “…it is a widely…” Line 66: “…and emerged…” --> “…and it has emerged…” Line 114: “…for existing ones.” --> “…from existing ones.” Line 303: “…by Wilcoxon signed-rank test…” --> “…by the Wilcoxon signed-rank test…” Line 341: “…IQA is scribed…” --> “…IQA is described…” Line 350: “…66.67%%...” --> “…66.67%...” Experimental design: The paper includes a good experimental design of a large empirical study by including a set of widely used ML techniques and datasets, a set of non-biased evaluation measures and statistical tests. However, statistical significance cannot be assessed alone without analysing the effect size (Arcuri and Briand, 2004 https://dl.acm.org/doi/10.1002/stvr.1486). The effect size is a quantitative measure of the magnitude of the experimental effect and it would give the reader a better understanding of the impact of the difference in results. Some of the datasets used in this study describe systems that consist of multiple versions, for example Log4j1.0 and Log4j1.1. Is there any reason 10-fold cross validation was favoured and applied over the more realistic cross-version defect prediction scenario whereby the model is trained on an older version and tested on the most recent one? Validity of the findings: The study includes a well-developed related work section but fails to position the work with respect to existing ones, for example, the work of Wang and Yao, 2013. This undermines the contribution and novelty of the study presented which should be better highlighted in the manuscript with respect to the large number of existing studies tackling the data imbalance problem and comparing over- and under-sampling techniques. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PREDICTING DEFECTS IN IMBALANCED DATA USING RESAMPLING METHODS: AN EMPIRICAL INVESTIGATION Review round: 2 Reviewer: 1

Basic reporting: The authors have addressed the previously raised comments Experimental design: The authors have addressed the previously raised comments Validity of the findings: The authors have mainly addressed the previously raised comments, thanks. I only still have a concern regarding the position of the paper with respect to related work. I believe it would be easier for the reader to understand the contribution of the work if the differences between the proposed study and previous work are highlighted in a better and clearer manner. I also believe that the newly added paragraph about previous similar work would be a better fit in Section 2 (Related Work) rather than in Section 5 (Discussion). Additional comments: no comment

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: NOVEL HYBRID FIREFLY ALGORITHM: AN APPLICATION TO ENHANCE XGBOOST TUNING FOR INTRUSION DETECTION CLASSIFICATION Review round: 1 Reviewer: 1

Basic reporting: 1. In the abstract of this paper, the authors consider that “One of the greatest issues from the domain of network intrusion detection systems are relatively high true positives and false negatives rates.” Nevertheless, a high true-positive rate is precisely what researchers in the IDS need and pursue. So why is it an issue to address? Please clarify whether the authors are a clerical error. If not, please fully explain. If it is indeed a clerical error, please check the article carefully to avoid ambiguity and misunderstanding caused by similar problems. 2. In the introduction of this paper, the authors need to verify their understanding of the basic concepts. In particular, the significance of false-positive rate and false-negative rate in IDS. If these two definitions are not proper, are subsequent performance metrics based on these defined correctly? Are subsequent experimental designs and data reliable? Experimental design: 1. The authors have carried on the sufficient experiment, the multi-level multi-stage comparison, and submitted the source code. However, could the authors consider presenting experimental data more diversely? Large and unlabeled forms can make reading difficult. In addition, the dataset is somewhat monolithic so that authors can try multiple datasets rather than just NSL-KDD. Validity of the findings: 1. Reducing false positive and false negative rates is an urgent problem in IDS. However, the authors need to explain more fully (1) why XGBoost is used as a classifier for IDS to solve this problem (especially when the model has been plagued by a high false-positive rate and false-negative rate) and (2) why the hyperparameters of XGBoost are optimized by a heuristic algorithm rather than other methods. It requires not only theoretical analysis but also experiments. 2. Please further describe the challenges met and problems solved in the research. And match them to the contribution of this work. 3. In the contribution of the paper, the authors state that “A novel enhanced FA metaheuristics has been developed by specifically targeting the well-known deficiencies of the original FA implementation;” Please specify the “well-known weaknesses” and how to overcome them? Given such shortcomings, why is it still necessary to base FA metaheuristics design solution? Why can’t other similar schemes replace FA metaheuristics? 4. The authors provide a clear and sufficient description of improvements of the firefly algorithm but lack the content of the overall scheme. Please describe the complete process of this scheme in an appropriate way. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: NOVEL HYBRID FIREFLY ALGORITHM: AN APPLICATION TO ENHANCE XGBOOST TUNING FOR INTRUSION DETECTION CLASSIFICATION Review round: 1 Reviewer: 2

Basic reporting: Basic reporting is average. Language and presentation should be improved. There are several typos that should be addressed. Authors need to make a careful revision of the document in this regard. Further IDS using ML and DL models is used widely discussed in the literature, including XGBoost algorithm and also with other ML based techniques. Further, there are techniques based on XGBoost to take care of encrypted environment. For eg. Intrusion detection model using fusion of chi-square feature selection and multi class SVM, 2017. Anomaly Detection Using XGBoost Ensemble of Deep Neural Network Models, 2021. Under this backdrop authors need to provide a justification for the new proposal. Is the focus is to improve FA or improve IDS accuracy? The statements in second page are confusing. Contributions listed do not indicate the IDS aspect. So the understanding goes is that the authors have worked on the optimization algorithm and in order to validate it they have used IDS datasets. In such case, the title of the paper is not justifiable. There are several computational aspects in the proposed algorithm which do not have specific way of handling. For e.g Perform search operations. What is the complexity? What is K in the algorithm? First of all,, what are the inputs to the algorithm and what are the output? It is better if the extensive literature reported are summarized in the background section. Experimental design: Results are fine. However there are other datasets such as UNSW, VIT Sparc, etc. Validity of the findings: Results are validated using the standard dataset. Hence they are validated. Additional comments: NIL

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: NOVEL HYBRID FIREFLY ALGORITHM: AN APPLICATION TO ENHANCE XGBOOST TUNING FOR INTRUSION DETECTION CLASSIFICATION Review round: 1 Reviewer: 3

Basic reporting: The English is clear. Experimental design: A design setup of the proposed architecture is required for analyzing the whole approach. Validity of the findings: What is the motivation for this work as there are many other models developed for IDS integrated with meta heuristic algorithms? The proposed model is tested on NSL-KDD dataset which is very old. The authors should select recent public benchmark intrusion detection datasets and show the superiority of the proposed model. Though ensemble models result in better accuracy, their execution time is greater in comparison to base classifiers. The authors should analyze the computational complexity of the proposed work. A comparative analysis of the proposed model with state-of-the-art intrusion detection models should be performed. What is the effect of optimizing the XGBoost parameters in the proposed approach. The authors should analyze in detail how these parameters had a positive impact in performance in discussion section. Additional comments: Certain intermediate results of CEC2013 can be separately added in a appendix section at the end of the paper.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: NOVEL HYBRID FIREFLY ALGORITHM: AN APPLICATION TO ENHANCE XGBOOST TUNING FOR INTRUSION DETECTION CLASSIFICATION Review round: 2 Reviewer: 1

Basic reporting: It is improved from the previous version. Experimental design: While the design looks good, experimental results are not presented clearly. Some of the tables are not visible. Validity of the findings: Yes. Additional comments: While answering the questions/suggestions some of the responses are overlapping towards the queries. The same answer is provided for different queries. This confusion should be clarified.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: NOVEL HYBRID FIREFLY ALGORITHM: AN APPLICATION TO ENHANCE XGBOOST TUNING FOR INTRUSION DETECTION CLASSIFICATION Review round: 2 Reviewer: 2

Basic reporting: Clear Experimental design: All comments are addressed by the authors Validity of the findings: All comments are addressed by the authors Additional comments: -

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: NOVEL HYBRID FIREFLY ALGORITHM: AN APPLICATION TO ENHANCE XGBOOST TUNING FOR INTRUSION DETECTION CLASSIFICATION Review round: 3 Reviewer: 1

Basic reporting: Basic reporting is fine. Experimental design: Experimental design is explained fine. Validity of the findings: It is experimentally validated by the authors. Additional comments: Article is improved from the current version.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: CGCNIMP: A CAUSAL GRAPH CONVOLUTIONAL NETWORK FOR MULTIVARIATE TIME SERIES IMPUTATION Review round: 1 Reviewer: 1

Basic reporting: Thanks for the submission and for sending me the manuscript (MS) by Liu et al. on Imputations of Geese titled “Bird-CGCNImp: a causal graph convolutional network for multivariate time series imputation” I find the approach is experimental and creative. I am not much against it, but the current MS falls short on a few relevant science points: -Whatever authors propose and defend, unless we see the data and the code with ISO compliant metadata, this work is not repeatable or transparent. It currently does not meet latest principles of science, or best professional practices. It can easily be fixed though as all data can now fully be made open access and with open source code models with ISO standard documentation (metadata), online, in GBIF and in Movebank even. I stress that specifically because work on these two species and the flyway is already full open access and available with 24 years of remote field work data. See GBIF and then Solovyeva et al. 2021. So why not here ? Not sharing data is not defendable whatsoever and it violates any best professional practices and collegiality. Before that is not clearly done I do not approve of the review or publication. -Imputation is just one of many methods. Personally, I would use the method of predicted ecological niches instead, or in parallel and to compare. Authors brought it down to a very narrow and self-fulfilling prophecy, just as the endless data filtering to get rid of so-called outliers (as often done in such telemetry works). From the MS and figures, authors seem to believe birds fly on a straight line connecting dots, which is highly dubious. How about krigging, Least Cost paths, or Circuitscape or Marxan approaches instead? One certainly needs a CERTAINTY and CONFIDENCE in the approach, outcome and maps; done how ? Anyways, if authors stick with imputations, please state citations on the issue, namely Jerome Friedman, classification trees and boosting, and many forest inventory and remote sensing references. Much literature and expertise sit there that was not used here. It should. -The way how to best approach the gap filling should be like a scenario, an hypothesis, or a re-analysis of the existing data. That should be made clear and pursued that way. -I like Figures 5, 6 and 7 for the concept but those are currently way too small and too selective to be useful or convincing. We need it for all the data, not just some examples. -let’s agree that the biggest topics - research design, representative sampling of the tagged species, technical fault patterns in the transmission, and impacts of anesthesia - are not mentioned. They must, as those do overrule the data and policy question one wants to purse, e.g. where are the ‘true’ dots, where are the flyways, habitats and do we have broadfront migration or any mixing ? Getting stuck with a line is just not appropriate for a cohort or population inference. The latter is presumably our all policy goal. -the title and concept of “Bird-CGCNImp” reads odd, and is odd. It’s not needed that way and should be improved. Same is true for ‘causal graph convolutional network’. That’s hardly English, not a term, and nobody speaks that way (in England, or the English speaking/reading world). -re ‘Identifying bird migration trajectories and discovering habitats is very important for conserving species diversity.’: That’s 100% untrue. Can you show me an example in China, or Malta or anywhere else ? We see flyway declines all over and for many years, see Jiao et al. 2016. The protected areas are way too tiny, paper parks, and do not help. So, this phrase above sets up a strawman; not needed but better to be honest. -the Literature references are poorly formatted, e.g. first names, abbreviations, and widely incomplete for topics -Finally, the problem we have in flyways and in China is a ruthless habitat conversion and loss scheme; as it is found globally now. Connecting the bird dots on a line for policy helps little. Happy to be shown any other. If authors can fix those items, specifically data shared and all data they have at hand, as well as conservation progress, then I would be happy to re-read their update perhaps. Thanks, best regards Falk Huettmann Experimental design: The article has no typical research design Validity of the findings: As stated in section 1, authors assume linear movements, which is not justified. Additional comments: Thanks. I am happy to support creative and experimental work, but the MS - as presented - fails on some basic science issues, namely transparency and repeatability.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: CGCNIMP: A CAUSAL GRAPH CONVOLUTIONAL NETWORK FOR MULTIVARIATE TIME SERIES IMPUTATION Review round: 1 Reviewer: 2

Basic reporting: In this paper, the authors propose a causal GCN for bird migration trajectories motivated by the existence of both the attribute correlation and the temporal auto-correlation dependencies. The proposed method is novel and the application of bird migration trajectories shows its effectiveness. Experimental design: In the proposed Bird-CGCNImp method, the authors establish an end-to-end multi-task model to capture both attribute correlation and temporal auto-correlation dependencies. Besides, the authors also notice the noise in the actual sampling process and design a total variation reconstruction regularization term to improve the imputation accuracy. The authors clearly present the proposed method's motivation and technology details. Validity of the findings: In the experiment, the authors conduct the experiment on one public time-series imputation benchmark and two real-world bird migration trajectory datasets. The experimental results are promising. The authors also demonstrate the effectiveness of each component in Bird-CGCNImp by ablation study. My suggestion of the experiment is that the authors should repeat the experiment several times and report the mean and standard deviation of the metric so that the experimental results would be more convincing. Additional comments: Missing citations: Suo Q, Yao L, Xun G, Sun J, Zhang A. Recurrent imputation for multivariate time series with missing values. In2019 IEEE International Conference on Healthcare Informatics (ICHI) 2019 Jun 10 (pp. 1-3). IEEE. minor: Some of the citations are incomplete. For example, the following citations miss the journal name: Nazabal, A., Olmos, P. M., Ghahramani, Z., and Valera, I. (2020). Handling incomplete heterogeneous 481 data using vaes. Yoon, J., Zame, W., and Schaar, M. (2017). Multi-directional recurrent neural networks : a novel method 498 for estimating missing data.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: FEATURE SELECTION BY INTEGRATING DOCUMENT FREQUENCY WITH GENETIC ALGORITHM FOR AMHARIC NEWS DOCUMENT CLASSIFICATION Review round: 1 Reviewer: 1

Basic reporting: This work focuses on the application of Feature Selection by Integrating Document Frequency with a Genetic Algorithm for Amharic News Document Classification. A hybrid of document frequency (DF) and genetic algorithm (GA)-based feature selection method is suitable for use in a variety of applications requiring Amharic document classification. Datasets from different domains with different categories are considered. First, the writing should be improved a lot. A lot of grammar issues, unsound statements, and incomplete sentences phrases are very common. there are also very very long sentences, please make it readable, write short and concise sentences. There are some issues to be fixed. Introduction: " Examine the performance of the proposed feature selection method in terms of accuracy, precision, recall, and F-measure." it should be not your main contribution. What was your starting Research Question or the gap you want to fill? Clearly define your research questions. Introduction: "2007 census [1]" This is a very old reference, do you have a near time reference for this? On related work: revise it again even you have not cited the latest paper that was done in Amharic news document classification like this https://doi.org/10.1371/journal.pone.0251902 Data processing -Normalization: Which is "canonical" normalization or standardization? What was your base to normalize homophone characters into a single representation? It would be also nice if your approach is compared with homophone normalization and without homophone normalization approaches. Results and discussion section (table 6): Is it the different hybrid results are from the previously conducted paper or from your data? if it is from the previous paper, the data you have used is different, set clearly your experiment results. Experimental design: - Even the paper has minor, easily fixable, technical flaws that do not impact the validity of the main results, the paper contributes some new ideas and applicable resources. - Extensive experiments with multiple feature selection techniques. - Set clearly the gap that you have filled at the end of your experiments. - Link your results to the research questions that should be provided. Validity of the findings: The paper contributes some new findings regarding feature selection for Amharic news document classification. The dataset that has been provided is also valid and highly applicable for the next research. Additional comments: Go ahead through the suggestions that are mentioned in the above sections. Indicate your significant contribution very briefly.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: FEATURE SELECTION BY INTEGRATING DOCUMENT FREQUENCY WITH GENETIC ALGORITHM FOR AMHARIC NEWS DOCUMENT CLASSIFICATION Review round: 1 Reviewer: 2

Basic reporting: No comment Experimental design: no comment Validity of the findings: There is no comparative study of results with other authors [10] on the same dataset. The dataset used in this study does not match the dataset referenced [10]. Additional comments: Line 124, It looks like a regular pre-processing step for any other language. How is the pre-processing different in context to Amharic? Line 144, Gasser's HornMorpho stemmer: what is the accuracy of the stemmer used? How is it evaluated? is it a Rule-Based or Statistical?? Table 2, what basis are the documents classified into major categories? Who did this classification? Can a document not belong to multiple types, e.g. education and health? Table 2, what is the length of the documents in each category? Line 248, Any observations on varying the splitting ratio versus the proposed model's performance? Can the proposed methodology, etc. used be adapted/used for other similar languages? Conclusion of future work may discuss this. The dataset used here is small in size, and I recommend that the authors test your proposed algorithm on a larger dataset in a future studies.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: FEATURE SELECTION BY INTEGRATING DOCUMENT FREQUENCY WITH GENETIC ALGORITHM FOR AMHARIC NEWS DOCUMENT CLASSIFICATION Review round: 2 Reviewer: 1

Basic reporting: No comment. Experimental design: No comment. Validity of the findings: No comment. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: FEATURE SELECTION BY INTEGRATING DOCUMENT FREQUENCY WITH GENETIC ALGORITHM FOR AMHARIC NEWS DOCUMENT CLASSIFICATION Review round: 2 Reviewer: 2

Basic reporting: No Comment Experimental design: No Comment Validity of the findings: No Comment Additional comments: Kindly cite the following papers. https://doi.org/10.4218/etrij.2019-0458 doi: 10.1109/ICDIM.2018.8847044

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MULTI-LABEL EMOTION CLASSIFICATION OF URDU TWEETS Review round: 1 Reviewer: 1

Basic reporting: please see my review below Experimental design: please see my review below Validity of the findings: please see my review below Additional comments: The manuscript is centered on a very interesting and timely topic, which is also quite relevant to the themes of PeerJ Computer Science. Organization of the paper is good and the proposed method is quite novel. The length of the manuscript is about right but keyword list is missing. The paper, moreover, does not link well with recent literature on sentiment analysis appeared in relevant top-tier journals, e.g., the IEEE Intelligent Systems department on "Affective Computing and Sentiment Analysis". Also, latest trends in multilingual sentiment analysis are missing, e.g., see Lo et al.’s recent survey on multilingual sentiment analysis (from formal to informal and scarce resource languages). Finally, check recent resources for multilingual sentiment analysis, e.g., BabelSenticNet. Authors seem to handle sentiment analysis simply as a binary classification problem (positive versus negative). What about the issue of neutrality or ambivalence? Check relevant literature on detecting and filtering neutrality in sentiment analysis and recent works on sentiment sensing with ambivalence handling. Finally, the manuscript only cites a few papers from 2020 and 2021: check latest works on attention-based deep models for sentiment analysis and recent efforts on predicting sentiment intensity using stacked ensemble. Some parts of the manuscript may result unclear for some readers of this journal. A short excursus on emotion categorization models and algorithms could resolve this lack of clarity (as the journal does not really feature many papers on this topic) and improve the overall readability of the paper. On a related note, the manuscript presents some bad English constructions, grammar mistakes, and misuse of articles: a professional language editing service is strongly recommended (e.g., the ones offered by IEEE, Elsevier, and Springer) to sufficiently improve the paper's presentation quality for meeting the high standards of PeerJ Computer Science. Finally, double-check both definition and usage of acronyms: every acronym should be defined only once (at the first occurrence) and always used afterwards (except for abstract and section titles). Also, it is not recommendable to generate acronyms for multiword expressions that are shorter than 3 words (unless they are universally recognized, e.g., AI).

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MULTI-LABEL EMOTION CLASSIFICATION OF URDU TWEETS Review round: 1 Reviewer: 2

Basic reporting: Well written, but needs revisions. Experimental design: NA Validity of the findings: NA Additional comments: Abstract and Introduction should be revised. Problem statement must be clearly defined in the Introduction. Use simple present tense throughout the paper. Related work should have one paragraph of motivation due to limitations of existing approaches. Also, it should have references to the recent similar works. Authors can consider referring the following articles: A Consolidated Decision Tree-based Intrusion Detection System for binary and multiclass imbalanced datasets Performance Assessment of Supervised Classifiers for Designing Intrusion Detection Systems: A Comprehensive Review and Recommendations for Future Research Applications in Security and Evasions in Machine Learning: A Survey Comparison of the work with an recent existing approach is necessary to show the performance improvement. Result analysis must be thorough. Conclusion should include limitations of the existing work.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MULTI-LABEL EMOTION CLASSIFICATION OF URDU TWEETS Review round: 1 Reviewer: 3

Basic reporting: - The use of the verb "talks" refers to Section 2 in line 64 of page 2 is not suitable. - Spelling errors and punctuation errors are found in the second paragraph of page 9 (line 286-295). - Section 4.3: I suggest use variables represent "number of c(e)" or "number of p(e)" in equations on page 10. The variable c(e) should be written in mathematical form (italic). This change should apply to all variables in this article. - In Section 4.2, page 9: The use of capital letters in "We used the following BERT 301 parameters: MAXSEQLENGTH= 64, BATCH SIZE = 32, LEARNING RATE =2e5, and NUM TRAIN EPOCHS= 2.0. Experimental design: - The paper should explain the importance of all evaluation metrics in the context of this task. Explain how the metrics can describe the performance of the algorithms. - Coding for the project is publicly available. - In paragraphs 3 and 4 of section 5: The deep learning algorithms performed poorly in this study, why do authors claim that the algorithms perform well and the results are promising? - In section 4.2, why the values of parameters were selected in this project. Validity of the findings: - The description of the challenges faced in collecting information for this dataset and the annotation process should not be considered a contribution. - No prior work on multilabel emotion classification exists for the Urdu language, and the generation of the dataset can be considered a contribution of this research. - However, the performance of deep learning and machine learning algorithms for this task are very poor with low accuracy. Would you please explain why this is the case? - I would suggest more evaluation be conducted to identify algorithms with better accuracy. Please justify the relevance of your findings even though the accuracy results are mostly low for all algorithms in this study. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MULTI-LABEL EMOTION CLASSIFICATION OF URDU TWEETS Review round: 2 Reviewer: 1

Basic reporting: Many of the claims made by the authors are not backed by the revision, e.g., author claim to have added literature related to neutrality and ambivalence for sentiment analysis but there is no mention whatsoever of neither of them. Also, authors claim to have fixed acronyms but that is simply not true. Finally, presentation is still not up to PeerJ standards and important relevant literature on multilingual sentiment analysis is still missing. Experimental design: Please refer to my previous review Validity of the findings: Please refer to my previous review Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MULTI-LABEL EMOTION CLASSIFICATION OF URDU TWEETS Review round: 3 Reviewer: 1

Basic reporting: The authors have addressed all of my concerns and their revisions have substantially improved the manuscript. Experimental design: Good. Validity of the findings: Good. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: BMDD: A NOVEL APPROACH FOR IOT PLATFORM (BROKER-LESS AND MICROSERVICE ARCHITECTURE, DECENTRALIZED IDENTITY, AND DYNAMIC TRANSMISSION MESSAGES) Review round: 1 Reviewer: 1

Basic reporting: 1. Abstract has not describe completely the outstanding results of this research. 2. Why used Raspberry pi, not the other microcontroller, need to justified in this paper? Experimental design: The area of work seems to be interested. However, authors just made an experiment in lab (Amazon EC2 virtual machines (VMs) or prototype)), has no real experiment yet. We suggest authors to do more. Validity of the findings: The results in tables 4,5,6 and 7 can be simplified again, making it easier for readers to see the comparison results. Additional comments: no comment

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: BMDD: A NOVEL APPROACH FOR IOT PLATFORM (BROKER-LESS AND MICROSERVICE ARCHITECTURE, DECENTRALIZED IDENTITY, AND DYNAMIC TRANSMISSION MESSAGES) Review round: 1 Reviewer: 2

Basic reporting: The authors propose an IoT platform called BMDD (Broker-less and Microservice architecture, Decentralized identity, and Dynamic transmission messages) built on broker-less and microservice architecture using gRPC protocol as the primary protocol for data collection and device control. Besides, a decentralized authentication model based on blockchain technology to enhance the security of the IoT Platform will be introduced. Furthermore, BMDD also provides the function to manage user devices and channels that reduce security issues from user behavior. In addition, providing dynamic message exchange, as the purpose of this proposal, creates uniformity in IoT architecture and eliminates the distinction between different applications, data types, and IoT device manufacturers. The final recommendation is to add a message queue system to enhance platform reliability. The paper is well structured and readable. The paper has a good potential for being appreciated and cited, but it requires some improvements and also extension. The section Introduction should clarify better and provide concise information with regard to the problem definition and scope of the paper. About the related work section, each paper should clearly specify what is the proposed methodology, novelty, and results from experimentation. At the end of related works, highlight better in some lines what overall technical gaps are observed in existing works, that led to the design of the proposed approach. Innovative and self-organizing methodologies, as https://ieeexplore.ieee.org/abstract/document/9409962 should be reported. Experimental design: What is the time complexity for the proposed algorithm? The authors should highlight in what %age and in what parameters, the proposed methodology was found better as compared to existing ones Analysis about scalability features of the approach could be added to further improve the strength of the paper. Validity of the findings: The future scope of the methodology should be extended/highlighted. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: WRITER VERIFICATION OF PARTIALLY DAMAGED HANDWRITTEN ARABIC DOCUMENTS BASED ON INDIVIDUAL CHARACTER SHAPES Review round: 1 Reviewer: 1

Basic reporting: [ABSTRACT] The abstract should reflect the contributions of the manuscript. I suggest rewriting it. [INTRODUCTION] The authors should provide a clear problem definition and contributions in the introduction section. [RELATED WORK] Where are the related studies? They should be declared in a separate section. [RELATED WORK] A table of comparisons should be added at the end of the related studies section to praise the pros. and cons. of them. The year column should be added and they should be ordered by it. [EQUATIONS] The authors should follow the journal authors’ guidance in writing the equations, symbols, and variables. Please, refer to the authors guidelines on the journal official website. [EQUATIONS] Where are the equations of the used metrics? [DATASETS] Samples from the used dataset should be added and annotated. [METHODOLOGY] The suggested approach is not clearly discussed. More scientific details should be added. [METHODOLOGY] What are the used equations in the suggested approach? In other words, how the suggested approach is derived? [METHODOLOGY] Where is the overall pseudocode? Flowchart? of the suggested approach? [ABBREVIATIONS] The authors should add a table of abbreviations in the revised manuscript. [SYMBOLS] The authors should add a table of symbols in the revised manuscript. [REFERENCES] There are no citations for many sentences in the manuscript. Why? Please check. [REFERENCES] The references should be written in the same style following the journal authors’ guidance. [REFERENCES] Recent citations from 2021 should be added to the manuscript. Only one is found. [PROOFING] The authors should get editing help from someone with full professional proficiency in English. [PROOFING] The manuscript should be checked again to fix any typos such as missing spaces and commas. [CONSISTENCY] The manuscript structure is too short. It must be elaborated in their applied technology as should support more rigorous technical aspects. [CONSISTENCY] Some paragraphs are wrapped in more than 10 lines. They should be split concisely. Experimental design: [EXPERIMENTS] The experimental configurations (i.e., settings) should be declared and added to a table. [EXPERIMENTS] The working environment (i.e., software and hardware) should be declared and added to a table. [EXPERIMENTS] What are the criteria for selecting the experimental configurations? [EXPERIMENTS] More experiments should be conducted using different configurations. [EXPERIMENTS] Where is the tabular representation of the reported results? [EXPERIMENTS] The figures in the experiments section should be gridded. For example, Figure 8. [EXPERIMENTS] Why did not the authors compare their approach with others in a table? [EXPERIMENTS] Why did not the authors compare their approach with another approach to compare the suggested approach efficiency and applicability? [EXPERIMENTS] Can the authors draw the area under the curve (AUC)? [EXPERIMENTS] Why did not the authors calculate other performance metrics such as specificity and f1-score? [EXPERIMENTS] Where is the detailed and statistical discussion of the reported results? [EXPERIMENTS] More experiments should be conducted using a different dataset to prove the generalization. Validity of the findings: [NOVELTY] What is the novelty of the suggested approach? [CONCLUSIONS] The conclusions in this manuscript are primitive. Please, write your conclusions. Why did not the authors use transfer learning? More experiments should be conducted using it. Why the authors added Figure 8? In Table 1: Why the authors selected that model? What is the used criteria? [LIMITATIONS] What are the limitations of the current study? It should be added in a separate section. Additional comments: The manuscript presented “a verification of partially damaged handwritten Arabic documents approach”. However, the major and critical weak points are that: (1) Their proposed work discussion is weak distributed to be described or analyzed. (2) The novelty is not guaranteed. (3) Their work is not compared with state-of-the-art approaches nor related studies. (4) Their experiments leak from the descriptive and statistical analysis.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: WRITER VERIFICATION OF PARTIALLY DAMAGED HANDWRITTEN ARABIC DOCUMENTS BASED ON INDIVIDUAL CHARACTER SHAPES Review round: 1 Reviewer: 2

Basic reporting: The article proposes an approach for offline text independent writer verification of partially damaged handwritten Arabic documents, based on individual characters. The extraction of "alphabets" from handwritten text is a manual process. The idea is excellent, however, my main reservation is on the procedure, dataset used and results reached. The dataset collection itself is excellent, however, I suggest to expand it and make it based on all Arabic Character Shapes. Although the write up is professional there are some inconsistencies in some terms. Example of "handwritten" is spelled in 3 different ways: Two words (hand written) like in the abstract and line 88, (hand-written) as in line 254 and one word (handwritten) as in line 252 etc.. The authors use the term alphabets, letters, characters for Arabic. The most accepted term in the literature is character shape. Arabic characters take different shapes depending on their position in a word. Better to write the Arabic character shape in Arabic as in line 47 it helps the reader to see the character shape in question. The name of the character shape is not enough. Experimental design: The authors claim they are using all Arabic Alphabets. Their Extracted Alphabets Dataset (EAD) is collected from users using text in Figure 5. The text uses all Arabic Alphabet, however, it does not use all Arabic Character Shapes. Only about a quarter of the shapes are present in the text used. I suggest to expand the collection for other shapes as some missing shapes have very discriminatory features that could improve the writer verification and will be useful for an extension to writer identification. In this way I claim your reduced model will perform even better. I don't see the advantage of having IAD and EAD separate. IAD is only one character shape among the 4 possible shapes that an Arabic alphabet can take. The difference is only in the segmentation of Words into characters. If the extraction is manual where is the advantage, please clarify. Not clear: * our domain of work .. does not address the problem of writer identification from the Arabic letters. Our main focus is on how to identify the writer based on the "given" handwritten Arabic letters. * Comparison of isolated and extracted Arabic alphabet based approaches for writer verification. * Isolated Alphabet: Samples of user written isolated alphabets in Figure 4, is the middle column as I understand it. What is isolated alphabets in Figure 3, these are not the isolated shapes. Both columns are for middle shape or begin shape characters. Validity of the findings: I have a serious reservation about the findings. Why the same character body shape is giving different average error or average accuracy like in Figure 9 or Figure 11. Examples: Yaa-begin, Beh-begin, Noon-begin Teh, Theh, Haa, Khaa Feh-begin, Qaf-begin Feh-middle, Qaf-middle etc.. Additional comments: I encourage you to address the comments and reservation above and resubmit for eventual publication of your work.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: WRITER VERIFICATION OF PARTIALLY DAMAGED HANDWRITTEN ARABIC DOCUMENTS BASED ON INDIVIDUAL CHARACTER SHAPES Review round: 2 Reviewer: 1

Basic reporting: Thanks to the authors for updating the manuscript. After checking their responses to my comments, I can declare that they have made a suitable major revision. Experimental design: Thanks to the authors for updating the manuscript. After checking their responses to my comments, I can declare that they have made a suitable major revision. Validity of the findings: Thanks to the authors for updating the manuscript. After checking their responses to my comments, I can declare that they have made a suitable major revision. Additional comments: Journal: PeerJ Computer Science Manuscript Title: Writer Verification of Partially Damaged Handwritten Arabic Documents based on Individual Character Shapes Manuscript ID: PeerJ 67980 R1 Reviewer Number: 1 Submission Date: Wednesday, March 16, 2022 Thanks to the authors for updating the manuscript. After checking their responses to my comments, I can declare that they have made a suitable major revision. However, I have two minor comments: (1) Increase the size (i.e., width) of Figure 7 and Figure 9. (2) The authors should get another editing help from someone with full professional proficiency in English. For example, Table 2 caption “List of Symbols used” should be “List of Symbols”. For the authors in case of the authors got a chance to review the manuscript and submit the revised one after the editor’s decision, please, provide a table in the revised manuscript mentioning (1) the comment, (2) the authors’ response, and (3) the authors’ change (if applicable). Please, consider all of the comments and don’t ignore any of them. Please, refer to the attached file "PeerJ 67980 R1 Reviewer.pdf" for the same comments in an organized format.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: WRITER VERIFICATION OF PARTIALLY DAMAGED HANDWRITTEN ARABIC DOCUMENTS BASED ON INDIVIDUAL CHARACTER SHAPES Review round: 3 Reviewer: 1

Basic reporting: Thanks to the authors for updating the manuscript. After checking their responses to my comments, I can recommend the acceptance of the manuscript in its current version. Experimental design: Thanks to the authors for updating the manuscript. After checking their responses to my comments, I can recommend the acceptance of the manuscript in its current version. Validity of the findings: Thanks to the authors for updating the manuscript. After checking their responses to my comments, I can recommend the acceptance of the manuscript in its current version. Additional comments: Thanks to the authors for updating the manuscript. After checking their responses to my comments, I can recommend the acceptance of the manuscript in its current version.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ADJUSTING EYE ASPECT RATIO FOR STRONG EYE BLINK DETECTION BASED ON FACIAL LANDMARKS Review round: 1 Reviewer: 1

Basic reporting: interesting topic and the presented writing is adequate. However, majority of the literature references are too old (more than 5 years). The gap of knowledge of the proposed topic is unclear. Too short prior arts have been discussed, thus, not much of gap of knowledge can be identified from the present works. The landmark used in the study are rely on prior reported works from Dlib, and dataset utilized the publicly available resources from eyeblink8. Suggest the author to test on their own dataset. Experimental design: the proposed methods are inline with their problem statement. Again, I don't see much of novelty of their proposed works here although the methodology are explained sufficiently. Validity of the findings: Both qualitative and quantitative analyses are adequate. Additional comments: NA

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ADJUSTING EYE ASPECT RATIO FOR STRONG EYE BLINK DETECTION BASED ON FACIAL LANDMARKS Review round: 1 Reviewer: 2

Basic reporting: 1. A good motivational sentence is needed in the Introduction section of the study. 2. The problem space of the study should be expressed more clearly. With this explanation, it will be clearer how eye blink detection handles problem solving in the study. 3. Perhaps, the literature section can be improved by mentioning that there are different fields of study related to eyeblink. Experimental design: 1. It should be stated more clearly how the process is followed for the selection of the threshold. If this value is determined manually, how this value is selected should be explained. 2. It has been stated that environmental conditions are important in taking camera images. However, there is no experimental design related to this. Validity of the findings: well prepared. Additional comments: I think the study will get better after these edits and corrections.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ADJUSTING EYE ASPECT RATIO FOR STRONG EYE BLINK DETECTION BASED ON FACIAL LANDMARKS Review round: 2 Reviewer: 1

Basic reporting: The authors performed the necessary corrections and arrangements with a good effort. I think it can be published as it is. Experimental design: . Validity of the findings: . Additional comments: .

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PYVHR: A PYTHON FRAMEWORK FOR REMOTE PHOTOPLETHYSMOGRAPHY Review round: 1 Reviewer: 1

Basic reporting: Not all abbreviations are explained in text. For e.g. pyVHR, GPU, RGB, BVP ... Please check all abbreviations. Experimental design: No comment. Validity of the findings: No comment. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PYVHR: A PYTHON FRAMEWORK FOR REMOTE PHOTOPLETHYSMOGRAPHY Review round: 1 Reviewer: 2

Basic reporting: The topic for this study is not new but very interesting, and the methods are clearly described. The proposed system is important for health and biomedical applications. However, some major points are required before any progress. Experimental design: # The technical approach isn't defined in detail (e.g., no mathematical equations and no justifications for design choices). The experimental design was prepared in good style. Validity of the findings: Good Additional comments: Article title: An end-to-end Python Framework for Remote Photoplethysmography # Overall statement or summary of the article: The topic for this study is not new but very interesting, and the methods are clearly described. The proposed system is important for health and biomedical applications. However, some major points are required before any progress. # Please add some of the most important quantitative results to the Abstract and conclusion. # In section 1.1, the authors should clearly mention the weakness point of former works (identification of the gaps) and shows the key differences between the different previous methods and the proposed method. A comparative overview table should solve this point. # The technical approach isn't defined in detail (e.g., no mathematical equations and no justifications for design choices). # Please describe your software algorithm in a flowchart or block diagram. # The authors should describe more details for the used dataset in a table instead of mentioning the reference only. # Please state whether the study was conducted in accordance with the Declaration of Helskinki (the author should provide the human ethics protocol number).

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PYVHR: A PYTHON FRAMEWORK FOR REMOTE PHOTOPLETHYSMOGRAPHY Review round: 1 Reviewer: 3

Basic reporting: Main comments: I commend the authors for this project. It would indeed be very valuable to have an open-source Python library available that allows for consistent comparison of state-of-the-art approaches on multiple datasets. The main weakness of the project and this accompanying paper (in their current state) is that they seem to be stuck in 2018. Although the paper purports to be based on the most recent literature, there exists a whole slew of more modern, state-of-the-art rPPG methods (see detailed comments) and newer, larger datasets (e.g., VIPL-HR, OBF) which are unfortunately not addressed. To be accepted for publication in its current form, the paper needs to at least acknowledge that it focuses on legacy rPPG methods (pre-deep learning period). To make this project considerably more useful for the rPPG research community, the authors would future-proof the framework. This means relaxing the rPPG framework assumptions that existed until 2017 and supporting addition of modern rPPG models to the library, which take an entire stack of (cropped) video frames as input. There are also several other minor issues in the detailed comments below. Experimental design: Not applicable Validity of the findings: Not applicable Additional comments: Detailed comments: l. 1: I got slightly confused about the usage of the phase “end-to-end” in the title and throughout this paper. It seems to be taken to mean “a method which goes from raw video all the way to to HR estimates” – but then shouldn’t all rPPG methods that estimate a HR from video be described as end-to-end? It could also be seen as also contentious because the phrase has a special meaning when it comes to deep learning. l. 18: ... analyzing ... l. 18: Sentence is too long. Break up into two sentences. l. 20: Today, virtually all novel rPPG methods use machine learning to estimate the pulse signal or the heart rate. It is not obvious how those methods fit into the framework introduced in this paper. l. 52: This framework assumes that the selection of the region of interest is external to the definition of rPPG methods; and that all rPPG methods are defined based on the averaged RGB traces. While this framework is applicable to the methods implemented here, it is not compatible with more modern learning-based rPPG methods that usually regard entire video frames (possibly after face detection) as input. l. 70: real time l. 118: The introduction for Section 3 confused me – two main building blocks are mentioned here, but the following subsection structure is not organized accordingly. l. 123/Figure 2: Again, I believe the sequence of steps defined here do not apply “for the vast majority of rPPG methods proposed in the literature” – in fact, they probably only apply for very few methods proposed since 2018. Some examples of recent work that it does not apply to: - DeepPhys: Video-Based Physiological Measurement Using Convolutional Attention Networks. https://openaccess.thecvf.com/content_ECCV_2018/html/Weixuan_Chen_DeepPhys_Video-Based_Physiological_ECCV_2018_paper.html - RhythmNet: End-to-end Heart Rate Estimation from Face via Spatial-temporal Representation. https://arxiv.org/abs/1910.11515 - AutoHR: A Strong End-to-end Baseline for Remote Heart Rate Measurement with Neural Searching. https://arxiv.org/abs/2004.12292 - Multi-Task Temporal Shift Attention Networks for On-Device Contactless Vitals Measurement. https://arxiv.org/abs/2006.03790 - The Benefit of Distraction: Denoising Remote Vitals Measurements using Inverse Attention. https://arxiv.org/abs/2010.07770 - MetaPhys: few-shot adaptation for non-contact physiological measurement. https://arxiv.org/abs/2010.01773 - The Way to My Heart Is Through Contrastive Learning: Remote Photoplethysmography From Unlabelled Video. https://openaccess.thecvf.com/content/ICCV2021/html/Gideon_The_Way_to_My_Heart_Is_Through_Contrastive_Learning_Remote_ICCV_2021_paper.html - PhysFormer: Facial Video-based Physiological Measurement with Temporal Difference Transformer. https://arxiv.org/abs/2111.12082 l. 241: See the previous point. The authors does not seem to be aware of new learning-based methods published since 2018. l. 312: I like the detailed explanations here. Maybe the authors could use some of the information from here to qualify the statement about real-time applicability made earlier in line 70? l. 322: In my own investigations, I found that after optimizing the algorithms, at some point the computation required for spatial averaging of the skin regions can become a bottleneck. Have any investigations been made as to how the performance differs when varying the resolution of the original raw video? l. 355: Well done on including statistical assessments, these get ignored too often. l. 356: outperform (without s) l. 399: Many papers on rPPG also use the SNR (signal to noise ratio) metric. Is there any reason this wasn’t included here? l. 580: In light of my previous comments, and to future-proof this framework, I think it would be useful to add support for rPPG methods which rely on an entire video (potentially previously cropped to include a face) as input with dimensions [T, H, W, C]. l. 668: As pointed out in earlier comments, framework is not general enough to accommodate recent work in the field.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PYVHR: A PYTHON FRAMEWORK FOR REMOTE PHOTOPLETHYSMOGRAPHY Review round: 2 Reviewer: 1

Basic reporting: The authors have addressed all my concerns. Thank you for this project and putting in the effort to future-proof it by supporting DL model. Experimental design: The authors have addressed all my concerns. Thank you for this project and putting in the effort to future-proof it by supporting DL model. Validity of the findings: The authors have addressed all my concerns. Thank you for this project and putting in the effort to future-proof it by supporting DL model. Additional comments: The authors have addressed all my concerns. Thank you for this project and putting in the effort to future-proof it by supporting DL model.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: (RE)SHAPING ONLINE NARRATIVES: WHEN BOTS PROMOTE THE MESSAGE OF PRESIDENT TRUMP DURING HIS FIRST IMPEACHMENT Review round: 1 Reviewer: 1

Basic reporting: Hi, I hope you all are doing well. The work by the authors is good and the presentation of data is also good, personally, I like the topic. The author should improve the article to make it significant as I am suggesting some points. 1- Abstract is too long and even the main points of studies are missing in the abstract. it's too general. 2- Add contributions in the introduction section with bullets. 3- Literature is missing at least add some NLP-related studies. I am suggesting some below: a- A performance comparison of supervised machine learning models for Covid-19 tweets sentiment analysis b- Tweets classification on the base of sentiments for US airline companies c- Determining the Efficiency of Drugs under Special Conditions from Users’ Reviews on Healthcare Web Forums d- US Based COVID-19 Tweets Sentiment Analysis Using TextBlob and Supervised Machine Learning Algorithms e- Sentiment Analysis and Topic Modeling on Tweets about Online Education during COVID-19 f- Deepfake tweets classification using stacked Bi-LSTM and words embedding 4- Add some description of the used model's architectures. 5- Add comparison with Textblob and perform the comparison of machine learning models with textblob sentiments. 6- We didn't used neural networks models? Experimental design: Mention above Validity of the findings: Mention above Additional comments: Mention above

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: (RE)SHAPING ONLINE NARRATIVES: WHEN BOTS PROMOTE THE MESSAGE OF PRESIDENT TRUMP DURING HIS FIRST IMPEACHMENT Review round: 1 Reviewer: 2

Basic reporting: The article provides a novel approach to the treatment of bots, taking the learnign machine technique as the main quantitative measurement tool. However, the authors should consider the following requirements and recommendations in order to improve all areas. Title: I woud add the date of the political trial to the title to diferentiate from other impeachments. Abstract: The objective of the research should be clearer in the abstract, in line with the established hypothesis. The study should be expanded in future research and the current limitations should be overcome. This type of study, more than the numerical value, requires the qualitative and discursive-rhetorical analysis of the content of the bots. It does not matter if it is necessary to reduce the sample and resort to manual measurement. The final selection of actors is not clear; in the Discussion section in Figure 3 eight actors are discussed ... In previous graphs, only four are discussed. They should also be reflected in the abstract and in the Method (selected sample). Is the sample considered balanced based on the ideology and membership of the Democratic and Republican parties of these actors? Literature: The literatura missing references to the use of disinformation, political corrupción, twitter-rhetoric and critical analysis of speech. In general, we value the updated bibliography that is provided related to the study. Some recommended papers are related in the bibliography. As the article focuses on the Twitter network, we consider the following quote relevant: Campos-Domínguez, Eva (2017). “Twitter y la comunicación política”. El profesional de la información, v. 26, n. 5, p p . 785-793. https://doi.org/10.3145/epi.2017.sep.01 Experimental design: Research questions are not defined, essential in the methodological section of a study with these characteristics. The criteria for selecting the sample of political actors analyzed are not clearly described. A variable sheet is recommended to help identify which measurement elements have been taken into account: Actors Post Type: Bot, Human, Unknown Type of feeling: positive, negative, neutral ... Method: I would recommend to apply the triangulated content analysis methodology with a triple approach (qualitative, quantitative, discursive) on a smaller sample of bots. The fact of explaining the variables in the methodology would help to better understand the subsequent tables and graphs and the results of the analysis. We value the justified selection of sampling dates. As keywords, It would include the word disinformation, fallacy, lie, or something that reflects the purpose of the bot. Define the selection criteria for these actors; see if there is a balance in the selection, between Democratic and Republican leaders. The arbitrary selection of 3000 tweets enables a study of qualitative and discursive marks, which would get better the methodology of the article. We recommend incluiding: -Krippendorff, Klaus (2004). Content analysis. Sage. ASIN: B01B8SR47Y -Van-Dijk, Teun A. (2015). “Critical discourse studies. A sociocognitive Approach”, Methods of Critical Discourse Studies, v. 3, n. 1. -Pérez‐Curiel, Concha, Rivas‐de‐Roca, Rubén; García‐Gordillo, Mar (2021). “Impact of Trump’s Digital Rhetoric on the US Elections: A View from Worldwide Far‐Right Populism”, Social Sciences, v. 10, n. 152. https://doi.org/10.3390/ socsci100501 We value the detailed technical explanation of using botometer. Validity of the findings: Working with most of the data is appreciated; differential value of BERT. However, the results of a bot analysis should not be limited only to the quantitative, but should explain the topic of the bot, the profile of actors it praises or attacks ... if the bot includes hyperlinks, know what content they do reference... If the study also proposes an analysis of feelings, it is considered advisable to consult other investigations focused on the use of the fallacy and the propaganda of political discourse: Pérez-Curiel, Concha; Velasco-Molpeceres, Ana-María (2020). "Impact of political discourse on the dissemination of hoaxes about Covid-19. Influence of misinformation in the public and the media ”, Revista Latina de Comunicación Social, n. 78, pp. 65-97. https://www.doi.org/10.4185/RLCS-2020-1469 More than a recommendation, it could be considered as a need to complete the quantitative part, linked exclusively to machine learning. It would be appropriate to include some captures of Twitter messages cataloged as Bots. Results: The Figure 1 is very interesting. I would delve further into the interpretation of it. For example, analyzing the authorship factor of the post. The value of Human ahead of Bot or Unknowwn. It would also be convenient to interpret the tone of the sentiment shown in each message beyond the numerical. The question is: what variables determines the feeling, considering that it is a subjective factor? Very interesting the study of the tone of feelings. Discussion: Correct use of antecedents and previous context to justify the origin of the bots It is necessary to explain the criterion of why in some tables the study is limited to four actors and in others (Figure 3) a larger sample is taken. The limitations of the type of method used are obvious; the study is complete but only if we consider quantitative measurement. When the object of study are bots or false messages spread with the strategy of influencing, discursive analysis is required as the main methodology. It is highly recommended to keep this perspective in mind for present and future research. Conclusions: The conclusions should be completed much more in spite of the fact that the discussion is quite developed. One possibility is to link the conclusions to the discussion, at the same point. It is important that the conclusions refer to research questions that are not detailed in this study (see comment on methodology). They could also turn these conclusions around the objectives of the investigation. Neither are they well defined in the abstract or the text of the article. Considering that only quantitative measurement is performed, the significant role of bots in intensifying negative tone during impeachment cannot be affirmed. Numerical values ​​cannot measure message intentionality. The conclusions also refer to the most relevant literature on which the study has been based and to what extent this research provides novelty compared to previous ones. The conclusions should be completed much more in spite of the fact that the discussion is quite developed. One possibility is to link the conclusions to the discussion, at the same point. It is important that the conclusions refer to research questions that are not detailed in this study (see comment on methodology). They could also turn these conclusions around the objectives of the investigation. Neither are they well defined in the abstract or the text of the article. Considering that only quantitative measurement is performed, the significant role of bots in intensifying negative tone during impeachment cannot be affirmed. Numerical values ​​cannot measure message intentionality. The conclusions also refer to the most relevant literature on which the study has been based and to what extent this research provides novelty compared to previous ones. Additional comments: The article should generally deal more in depth with the literature related to the proposed topic. In addition, it is essential to explain the variables, the sample and ask whether the sentiment analysis requires a qualitative and descriptive approach, as well as a qualitative one.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: (RE)SHAPING ONLINE NARRATIVES: WHEN BOTS PROMOTE THE MESSAGE OF PRESIDENT TRUMP DURING HIS FIRST IMPEACHMENT Review round: 1 Reviewer: 3

Basic reporting: In this manuscript, the authors examine the role of bots in the spreading of Donald Trump’s tweets during the first impeachment trial. Specifically, the authors collected and analyzed more than 13 million tweets, collected during six key dates during the impeachment trial, to perform (1) sentiment analysis and (2) bot detection. Overall, the setup of the study and the analysis of the data were done well. I recommend Accept. Experimental design: To assess activities on Twitter during Trump’s first impeachment trial, the authors collected more than 13 million tweets on six key dates. For the sentiment analysis portion, the authors used a supervised learning approach, via the natural language processing model BERT. Though I was initially concerned with using only a training set of 3,000 tweets and applying the model to such a large dataset, the authors alleviated those concerns with adequate citations and associated discussions. For the bot detection portion, the authors used the botometer library and applied three different supervised learning techniques to the data. Overall, the research design was well-crafted, the methodology detailed and rigorous, and the discussion thorough. Validity of the findings: Data is provided via online repository. The results presented in the manuscript are sound and follow from the analyses. Statistically, the results and analyses are sound and rigorous. Consistent with their research question, the authors show that bots contributed significantly to the negative rhetoric on Twitter. Not only do bots comprise of approximately 20-25% of all tweets, bots are also overwhelmingly negative and appeared to be strategic in nature. Bots not only targeted Democratic politicians more than Republicans, tweets directed at Democratic politicians were more negative than those targeted at Republican politicians. More broadly, these results add to our understanding of political communications on Twitter, as well as the impact of bots on the political system. Additional comments: • Misspelling of the word “analyses.” In the manuscript, the authors misspelled the noun “analyses” as “analyzes.” See, for instance, line 15, 66, and 67. • Unnecessary capitalizations. For example, the word “tweets” is generally lowercase, but is uppercase in multiple spots in this manuscript (line 123, etc.) • Misspelling of the word “Twitter.” In the manuscript, Twitter is sometimes misspelled as “Tweeter.” (line 139)

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: (RE)SHAPING ONLINE NARRATIVES: WHEN BOTS PROMOTE THE MESSAGE OF PRESIDENT TRUMP DURING HIS FIRST IMPEACHMENT Review round: 2 Reviewer: 1

Basic reporting: Authors have done good work in revision and my recommendation for the paper is accepted. Experimental design: No Comments Validity of the findings: Comments Additional comments: Comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: (RE)SHAPING ONLINE NARRATIVES: WHEN BOTS PROMOTE THE MESSAGE OF PRESIDENT TRUMP DURING HIS FIRST IMPEACHMENT Review round: 2 Reviewer: 2

Basic reporting: The authors have made the requested modifications. These changes greatly improve the quality of the proposal. Experimental design: The methodology is now described in a more detailed and rigorous way. Introducing research questions has helped to better structure the results and the discussion. Validity of the findings: The results section has improved considerably as well as the discussion. Following the instructions of the reviewer, the position of the different parties, the type of bias, the authorship of the bots, among other topics of interest, are argued. Bearing in mind that the data is adequately analyzed, the discussion introduces a justified interpretation and reflection. Therefore, the conclusion is related to the objectives and research questions. Additional comments: We thank the authors for all the changes made to improve the proposal.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: (RE)SHAPING ONLINE NARRATIVES: WHEN BOTS PROMOTE THE MESSAGE OF PRESIDENT TRUMP DURING HIS FIRST IMPEACHMENT Review round: 2 Reviewer: 3

Basic reporting: This article is well-written, with some minor errors (detailed later). The authors adequately addressed the suggestions made by other reviewers and myself from the initial round of review. Experimental design: In my original review, I summarized, “Overall, the research design was well-crafted, the methodology detailed and rigorous, and the discussion thorough.” The revised manuscript, to address the questions raised by other reviewers, is even better. Validity of the findings: I like the revised manuscript. The authors made clearer their contributions (i.e., the three research questions), and the presentation of the empirical results is well-aligned and presented to the noted contributions. Additional comments: In this manuscript, the authors examine the role of bots in the spreading of Donald Trump’s tweets during the first impeachment trial. Specifically, the authors collected and analyzed millions of tweets, collected during six key dates during the impeachment trial, to perform (1) sentiment analysis and (2) bot detection. Specifically, the authors asked: (1) Are bots actively involved in the debate? (2) Do bots target one political affiliation more than another? And (3) Which sources are used by bots to support their arguments? In my original review, I recommend Accept. The revised manuscript is significantly better. After reading the comments from other reviewers and the authors’ response, I again recommend acceptance of this manuscript, with suggestions below for minor revisions. Minor points • Discrepancy in number of tweets collected. In the abstract, the authors noted that they collected more than 10 million tweets. However, in the Methods section (Line 141), the authors noted they collected more than 13 million tweets. This is likely a typo, but it needs to be corrected • To be consistent with official spelling, I recommend change these company names: “youtube” to “YouTube,” “facebook” to “Facebook,” “civiqs” to “Civiqs,” “google” to “Google,” “wordpress” to “WordPress,” “change.org” to “Change.org,” “blogspot” to “Blogspot,” etc. This specific comment is particular to Lines 308-315. Interestingly, company names are correctly capitalized in Lines 316-330, so I recommend revising to be consistent. I also recommend checking the rest of the manuscript (text, figures, tables, etc.) for these issues.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ENHANCING DATA TRANSMISSION IN DUCT AIR QUALITY MONITORING USING MESH NETWORK STRATEGY FOR LORA Review round: 1 Reviewer: 1

Basic reporting: In the introduction section authors discuss about previous research works related to this topic along with the potential communication technologies for this application. The paper purposes implementation of LoRa for air quality monitoring of large buildings. in this section they should clearly mention what is the novelty and contribution of this paper. In chapter two authors discuss about the previous research work done related to this field. In chapter 3 authors discuss about the experimental setup of this project. In this chapter authors discuss about the experimental setup of the project the. Two communication architecture has been tested. In first architecture end node is directly connected to the master node and in second architecture a repeater node has been used to hop data to the master node. In the results section authors conclude that the mesh network performs better than the direct end-node to master communication on the basis of experimental data. Experimental design: Two communication architecture has been used • Direct communication between end-node to master node • Mesh network-based architecture where an intermediate node has been used for data hopping in case of weak signal strength. Authors haven't mentioned what is the uniqueness of their experiment as compared to previous research in this area. Validity of the findings: Authors have compared the performance of the single hop and two hop based data transmission based on Packet delivery rate for different scenarios. From the experimental results the authors concluded that mesh network-based system provides better Comments • Data from table 2-4 can be represented more simply in terms of a chart. • Figures should be made clearer. • Use same kind of chart to represent PDR in different scenario in figure 11. Either both histogram of both lines. • Novelty not assessed. Additional comments: • My question is why use LoRa for this specific application? Why not use preexisting communication infrastructure available in the building like WIFI or LAN? • Authors should clearly mention what is the nobility and contribution of this paper. • Authors are suggested to include analysis regarding data rate of the network, Frequency of the data extraction from each node and how does the size of network effect the data extraction frequency. • One of the main abilities of LoRa is the ability to transmit data on multiple spreading factor which allows it to trade data rate and range of communication. So instead of using mesh configuration variable spreading factors can be used for reliable communication.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ENHANCING DATA TRANSMISSION IN DUCT AIR QUALITY MONITORING USING MESH NETWORK STRATEGY FOR LORA Review round: 1 Reviewer: 2

Basic reporting: 1.Some sentences in the text are not clear，for example "The repeater nodes were configured to receive the data and forward it to the end node while the end node was programmed to receive data from the repeater node instead of receiving it directly from the master node." 2.Literature references are sufficient. 3.Figure 1 is not necessary for illustration. Figure 4 is too simple to illustrate the algorithm flow. 4.The experimental results are convincing. 5.Please state the conclusion in more professional terms. Experimental design: 1.The content described in the article is within the objectives and scope of the magazine. 2.Research question well defined, relevant & meaningful. However, the description of the method used is too simple and the description of the network structure is not clear. 3.For the technical description, please specify the function of the mobile phone. 4.The steps in the method description are reasonable and the detailed information is insufficient. For example, the flow chart is too simple. Validity of the findings: 1.The article has certain novelty. 2.The data provided are reliable and sufficient. 3.Please carefully summarize the innovation of the article and write the conclusion in academic words. Additional comments: 1.It is suggested to merge the contents of "2. RELATED WORKS" into "1. INTRODUCTION" because they are all literature analysis.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ENHANCING DATA TRANSMISSION IN DUCT AIR QUALITY MONITORING USING MESH NETWORK STRATEGY FOR LORA Review round: 1 Reviewer: 3

Basic reporting: - Your related works need more detail. I suggest that you consider the other works related to LoRa mesh networks and routing protocols for the wireless mesh networks. - Figures 4, 5, 6 should be provided in high quality (for example with 600dpi) - In Figures 11, 12 the comparison of PDRs should be described in 2 columns as the comparison in the Fig. 13. Experimental design: - In the experiments, the authors added some repeater-nodes between master-nodes and end-node, which organizes multi-hop transmission between them. However, the author should consider the influence of the number of master-nodes and the number of repeater-nodes to PDR. Validity of the findings: - The results have shown that using additional repeater-nodes increases the PDR and the network coverage in the duct air monitoring application. Additional comments: - Amit Mullick et al have shown the significant of using a LoRa mesh network for monitoring application, in particularly in duct air monitoring. The authors have proposed a mesh network architecture based on LoRa technology. However, there are repeater-nodes used to relay packets from a master-node to an end-node, and the authors haven’t provided which routing protocols used in the LoRa mesh network. - The authors should provide more detail about the proposed mesh architecture.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: HOW TO GET BEST PREDICTIONS FOR ROAD MONITORING USING MACHINE LEARNING TECHNIQUES Review round: 1 Reviewer: 1

Basic reporting: The importance of the application domain should be shortly mentioned and emphasized in the abstract. The study being retrospective and no independent assessor for the outcome would carry great bias in different issues including selection and exclusion of patients and outcome assessment. Experimental design: In each part, in the experimental results and analysis section, before giving the driven conclusions, the results should be summarized, explaining what is given in Table, column, row, etc. This is necessary, so that the reader can validate the outcomes obtained. Include more technical discussions on the observations would strengthen the paper's contribution. Validity of the findings: The comparison is not fair to verify the proposed method. Include more technical discussions on the observations would strengthen the paper's contribution. More recent references should also be included: Only 3 out of 25 cited papers are published in the last 5 years. Language, grammar errors need to be addressed and add latest references. Additional comments: The study being retrospective and no independent assessor for the outcome would carry great bias in different issues including selection and exclusion of patients and outcome assessment.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: HOW TO GET BEST PREDICTIONS FOR ROAD MONITORING USING MACHINE LEARNING TECHNIQUES Review round: 1 Reviewer: 2

Basic reporting: 1. extracting 30 features what are those and not explained properelly Experimental design: architecture is not available. if possible add it Validity of the findings: good Additional comments: NO

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: HOW TO GET BEST PREDICTIONS FOR ROAD MONITORING USING MACHINE LEARNING TECHNIQUES Review round: 1 Reviewer: 3

Basic reporting: The article flows well. However, there are several grammatical and spelling errors in the submission. Please ensure you correct them. Experimental design: The datasets used in the paper should be further defined so that the reader understands different nuances involved in it. In the current state, the paper only describes the datasets in couple of sentences. Validity of the findings: I am fine with the methods used. Please include the majority class baseline for your models in table 4. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: HOW TO GET BEST PREDICTIONS FOR ROAD MONITORING USING MACHINE LEARNING TECHNIQUES Review round: 2 Reviewer: 1

Basic reporting: 1. Found a few Spelling Mistakes in the paper 2. literature is not there 3. the Problem: If there was no problem, there would be no reason for writing a manuscript, and definitely no reason for reading it. So, please tell readers why they should proceed with reading. Experience shows that for this part a few lines are often sufficient. Experimental design: ok Validity of the findings: good Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: HOW TO GET BEST PREDICTIONS FOR ROAD MONITORING USING MACHINE LEARNING TECHNIQUES Review round: 2 Reviewer: 2

Basic reporting: Authors have addressed all my concerns from my previous review Experimental design: Authors have addressed all my concerns from my previous review Validity of the findings: Authors have addressed all my concerns from my previous review Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: HOW TO GET BEST PREDICTIONS FOR ROAD MONITORING USING MACHINE LEARNING TECHNIQUES Review round: 3 Reviewer: 1

Basic reporting: ok Experimental design: ok Validity of the findings: ok Additional comments: Modified as per the comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: CLASSIFICATION OF ZOPHOBAS MORIO AND TENEBRIO MOLITOR USING TRANSFER LEARNING Review round: 1 Reviewer: 1

Basic reporting: The presentation quality of this paper is good. The background, the task, and the model are introduced clearly. However, the technical contribution of this paper is less. It seems that this paper just simply applies the VGG-19 model for solving the Zophobas Morio and Tenebrio Molitor classification problem. Experimental design: 1. It seems that the authors do not perform baselines on the dataset and provide comparisons. Specifically, what is the performance of non-transfer-learning models on the dataset? What is the superiority of the VGG-19 model compared with other transfer learning approaches on the given dataset? 2. Since the dataset is very small, the authors are suggested to use cross-validation to avoid the impact of sampling. Validity of the findings: The novelty of this paper has not been assessed. The dataset of this paper has not been provided. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: CLASSIFICATION OF ZOPHOBAS MORIO AND TENEBRIO MOLITOR USING TRANSFER LEARNING Review round: 1 Reviewer: 2

Basic reporting: This is a good technical paper. Figures are clear and have at least 300 dpi. On the related works, the authors should move Table 2 to the Discussions section as the benchmarking between their proposed methods and other papers. Table 2 should discuss the summary of the work, methods, strengths, and weaknesses of other papers. Experimental design: The use of VGG-19 should be justified. The dataset is appropriate. No data augmentation performed? Validity of the findings: The original Table 2 could be moved here to improve proposed method validation. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: CLASSIFICATION OF ZOPHOBAS MORIO AND TENEBRIO MOLITOR USING TRANSFER LEARNING Review round: 1 Reviewer: 3

Basic reporting: 1. The English article is written quite clear and professionally. 2. The background and context are sufficient for this topic. 3. The article structure is quite reasonable 4. Relevant results to the hypothesis are quite self-contained. 5. Formal results are clear and have detailed proofs. Experimental design: 1. This article is within the aims and scope of the journal. 2. The research questions are well defined and fill the specific problems of difficulties to diffenetiate these two worms. 3. This investigation was performed to a high technical and ethical standard. 4. The methods are described with sufficient detail and information to replicate. Validity of the findings: 1. This article gives impact and novelty for worms recognition algorithm. 2. All underlying data have been provided and robust, statistically sound, and controlled. 3. Conclusions are well stated, linked to original research questions and limited to supporting results. Additional comments: 1. In general, we are not confused by distinguishing these two worms in my lab. 2. The app will help beginners to differentiate these two worms.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: CLASSIFICATION OF ZOPHOBAS MORIO AND TENEBRIO MOLITOR USING TRANSFER LEARNING Review round: 2 Reviewer: 1

Basic reporting: I believe the authors have improved the paper based on the previous reviewer comments. Experimental design: The authors have justified the use of VGG16. They even added other traditional classifiers and another deep learning classifier, such as Inception v3. Cross-validation has been evident as well in this revised paper. The justification of the authors to not use data augmentation is rather acceptable. Validity of the findings: The findings have been properly validated and benchmarked. The revised paper looks better than the original. Additional comments: The previous issues raised have been addressed by the authors.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A NOVEL CAPTCHA SOLVER FRAMEWORK USING DEEP SKIPPING CONVOLUTIONAL NEURAL NETWORKS Review round: 1 Reviewer: 1

Basic reporting: This research primarily used deep learning methods to break state-of-the-art CAPTCHA codes. Overall the idea is good and the study is organized, however, there are still some improvements that need to be done. Please use the following comments and revise the manuscript accordingly. Experimental design: 1. The proposed approach used two different types of datasets on the same model. How can the same prediction model predict the different types of captchas? Please elaborate. 2. The previous studies have used dataset generation, where the proposed study used only a public study dataset. Why? Validity of the findings: 1. It is said that 5-folds have been used, where no detail of each fold is given that how it is calculated. Please evaluate it in terms of table or diagram etc. 2. The 'conclusions' are a crucial component of the paper. It should complement the 'abstract' and is normally used by experts to value the paper's engineering content. In general, it should sum up the most important outcomes of the paper. It should simply provide critical facts and figures achieved in this paper for supporting the claims. Additional comments: 1. The proposed study has used a proposed CNN, where in previous studies, most of the researchers have used similar CNNs. How the proposed CNN is different and novel as compared to previous studies? Clearly mention and highlight. 2. The abstract is too general and not prepared objectively. It should briefly highlight the paper's novelty as the main problem, how it has been resolved, and where the novelty lies? 3. The existing literature should be classified and systematically reviewed instead of being independently introduced one by one. 4. It seems that Skip connection is used in the proposed CNN; how does it work differently? No detail of this part is given. Please evaluate it specifically in some separate sections to get into depth.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A NOVEL CAPTCHA SOLVER FRAMEWORK USING DEEP SKIPPING CONVOLUTIONAL NEURAL NETWORKS Review round: 1 Reviewer: 2

Basic reporting: The author proposed deep learnig method for captcha breaking security. The paper technically sounds, however, there are still several issues needs to be fixed. In the Literature review, the table shows only CNN-based approaches in 2021, mainly where if the table is created, it must highlight different aspects of studies from different years. The proposed study has used text-based CAPTCHA images that contain multiple characters where single character recognition results and predictions are shown? How is the segmentation processed? Are previous studies similarly doing segmentation? What is the reason for choosing these specific datasets? When researchers using their own generated dataset, that is more in several images as well. What does mean by 5-fold validation in CNN? Please clearly make a separate section, also mention what its significance? In the comparison section, in the table, it is said “different” type of characters! What does it mean by that? Please elaborate on the results and comparison section. There are some typos and grammatical errors in the manuscript. It is strongly suggested that the whole work to be carefully checked by someone has expertise in technical English writing. Key contribution and novelty has not been detailed in manuscript. Please include it in the introduction section What are the limitations of the related works Are there any limitations of this carried out study? How to select and optimize the user-defined parameters in the proposed model? There are quite a few abbreviations are used in the manuscript. It is suggested to use a table to host all the frequently used abbreviations with their descriptions to improve the readability Explain the evaluation metrics and justify why those evaluation metrics are used? It seems that the authors used images of equations, please use editable equation format. The Related Works section is also fair, yet the criteria behind the selection of the works described should be explained. Experimental design: please see above Validity of the findings: please see above Additional comments: please see above

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A NOVEL CAPTCHA SOLVER FRAMEWORK USING DEEP SKIPPING CONVOLUTIONAL NEURAL NETWORKS Review round: 1 Reviewer: 3

Basic reporting: What are the research gaps in previous studies, and what are your contributions? Have not to mention clearly. Does the proposed CNN have a skip connection? What is that? How is it working? Why do you use it? How is the segmentation step processed? It is missing in the manuscript. It should be separately discussed. Experimental design: The substantial contributions should be highlighted and discussed. The results and comparative analysis should be discussed in detail. Every time a method/formula is used for something, it needs to be justified by either (a) prior work showing the superiority of this method, or (b) by your experiments showing its advantage over prior work methods - comparison is needed, or (c) formal proof of optimality. Please consider more prior works. Validity of the findings: The data is not described. Proper data description should contain the number of data items, number of parameters, distribution analysis of parameters, and target parameter for classification. Figure resolutions are bad. The text in figures became unreadible due to digitizing. Some figures have black background. Why the background is black for them? 8. The tables are too long. What te authors are trying to show in the tables are not clear? 9. Introduction contains too many citations. Please keep them at literature review. Authors are not supposed to explain the results in the introduction part. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A NOVEL CAPTCHA SOLVER FRAMEWORK USING DEEP SKIPPING CONVOLUTIONAL NEURAL NETWORKS Review round: 2 Reviewer: 1

Basic reporting: The entire paper has been written perfectly. Literature is sufficient. The organization of the paper is good. Experimental design: Experiments are valid. Results are presented systematically. Moreover, the quality of the paper has been improved. Validity of the findings: All findings are well written and presented. Additional comments: The paper can be accepted.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A NOVEL CAPTCHA SOLVER FRAMEWORK USING DEEP SKIPPING CONVOLUTIONAL NEURAL NETWORKS Review round: 2 Reviewer: 2

Basic reporting: The manuscript is relatively improved. However, I still think the manuscript needs reorganization, and contribution needs to be more precise. Please see the following comments, and hence, I would suggest one more round of revision. Experimental design: *Please fix Table 1 position; It should be in line with the LR section. *Similar Issue with Table 3, also please rewrite Table 3 caption. Validity of the findings: *Table 5 borders are out of page width; please fix this as well. *Would you please add two-three more points on the contribution of the mathdology? Additional comments: See above

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A NOVEL CAPTCHA SOLVER FRAMEWORK USING DEEP SKIPPING CONVOLUTIONAL NEURAL NETWORKS Review round: 2 Reviewer: 3

Basic reporting: I recommend to accept the current version Experimental design: I recommend to accept the current version Validity of the findings: I recommend to accept the current version Additional comments: I recommend to accept the current version

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A NOVEL CAPTCHA SOLVER FRAMEWORK USING DEEP SKIPPING CONVOLUTIONAL NEURAL NETWORKS Review round: 2 Reviewer: 4

Basic reporting: The manuscript is written in poor English that not suitable for publish in this journal until it be improved to ensure that audiences can clearly understand the text. There are a lot of short sentences that lack context. Experimental design: Not bad. Validity of the findings: Not bad. Additional comments: Not bad.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A NOVEL CAPTCHA SOLVER FRAMEWORK USING DEEP SKIPPING CONVOLUTIONAL NEURAL NETWORKS Review round: 2 Reviewer: 5

Basic reporting: Initially, I would like to appreciate the authors efforts for their interesting work. The work is a timely one since the deep learning based CAPTCHA breaking has been gained momentum within the recent years. Generally, the work can be evaluated as good and worth reading one, however there are several issues which should be taken into account before it could be accepted for publication. 1. The text, throughout the manuscript and more specifically in the abstract, introduction and literature review is not integrated, well-written. It mostly seems several short, and in some cases unrelated sentences without any coherence. So, the text does not seem natural and should be revised, maybe by a native person. 2. Some of the references in Introduction section are not necessary, since those are related to obvious and common information not something specific and important. For example, in line 49, "Azad and Jain (2013). CAPTCHA can be used for authentication in login forms with various web credentials". Instead, the authors can refer to some of the important survey papers (such as [1,2]) in the this field for all of such information. [1]Roshanbin, N., & Miller, J. (2013). A survey and analysis of current captcha approaches. Journal of Web Engineering, 001-040. [2]Xu, X., Liu, L., & Li, B. (2020). A survey of CAPTCHA technologies to distinguish between human and computer. Neurocomputing, 408, 292-307. 3. In the Literature Review section, specifically lines 101 to 109, the organization of the information is poor and need to be revised to be more clear and readable. 4. Generally, the Literature Review section is not well-organized and both its text and structure should be improved. For example, the authors may introduce some of classical CPATCHA breaking methods and elaborate on their deficiency. Then, they can speak about benefits of deep learning based methods and mention the notable works in this domain, either chronologically or based on underlying approach. 5. Some of the tables are not appropriately located in the manuscript. It may be due to inconsistency in Latex but anyway the issue should be managed. 6. Figure 1 should be revised. There are some problems with the figure. For example, the connection between different sections is not illustrated appropriately. This is not clear, for example, how input image will be passed to the next step. Also, the circle in the upper section of the figure is not large enough to fit the "preprocessing" term. 7. The section which is entitled "Comparison" seems to be unnecessary since its information could be presented in the discussion section. 8. The authors are encouraged to tell about the possible future works in the field. 9. The major motivation of the study should be clearly stated, may be in a separate section. What is the advantage of this study and what it brings to the community? What problem(s) this work is intended to solve and what is new in this research? 10. Since the most of today's CAPTCHAs are image-based, the authors should clarify why such works are still useful and worth considering. Can such research come in handy for other applications? Experimental design: The experiment is well-designed and conducted, so it seems appropriate. It is suggested that the authors speak about the efficiency of the proposed method for text-based CAPTCHAs in other languages, either experimentally or theoretically. Validity of the findings: The findings are acceptable and worth considering. Additional comments: If the authors could revise their manuscript appropriately and according to the comments, it could be accepted for publication.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A NOVEL CAPTCHA SOLVER FRAMEWORK USING DEEP SKIPPING CONVOLUTIONAL NEURAL NETWORKS Review round: 3 Reviewer: 1

Basic reporting: Accept Experimental design: Accept Validity of the findings: Accept Additional comments: Accept

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A NOVEL CAPTCHA SOLVER FRAMEWORK USING DEEP SKIPPING CONVOLUTIONAL NEURAL NETWORKS Review round: 3 Reviewer: 2

Basic reporting: Although the author has improved the manuscript, I still have some concerns on the benchmark models. I check the references of the investigated models, and find the benchmark models are not convincing. Here, I suggest authors to compare the proposed algorithm with some recent state-of-art algorithms from some top tier journals (IEEE Transactions on Information Forensics and Security, Computers & Security) in computer security. Experimental design: Not bad. Validity of the findings: Authors have to compare the proposed algorithms with some state-of-art methods in some top journals. I find many work on the CAPTCHA solver framework from the google scholar as below. 1. Alqahtani, Fatmah H., and Fawaz A. Alsulaiman. "Is image-based CAPTCHA secure against attacks based on machine learning? An experimental study." Computers & Security 88 (2020): 101635. 2. Ouyang, Zhiyou, et al. "A cloud endpoint coordinating CAPTCHA based on multi-view stacking ensemble." Computers & Security 103 (2021): 102178. 3. Osadchy, Margarita, et al. "No bot expects the DeepCAPTCHA! Introducing immutable adversarial examples, with applications to CAPTCHA generation." IEEE Transactions on Information Forensics and Security 12.11 (2017): 2640-2653. 4. Shi, Chenghui, et al. "Adversarial captchas." IEEE Transactions on Cybernetics (2021). Additional comments: The current experiment is not convincing.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A NOVEL CAPTCHA SOLVER FRAMEWORK USING DEEP SKIPPING CONVOLUTIONAL NEURAL NETWORKS Review round: 3 Reviewer: 3

Basic reporting: The reviewer would like to appreciate the authors' work and effort in revising the manuscript. So, the manuscript has been improved and can be published in its current form. Experimental design: The reviewer would like to appreciate the authors' work and effort in revising the manuscript. So, the manuscript has been improved and can be published in its current form. Validity of the findings: The reviewer would like to appreciate the authors' work and effort in revising the manuscript. So, the manuscript has been improved and can be published in its current form. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A NOVEL CAPTCHA SOLVER FRAMEWORK USING DEEP SKIPPING CONVOLUTIONAL NEURAL NETWORKS Review round: 4 Reviewer: 1

Basic reporting: The manuscript can be accepted. Experimental design: I have no more comment. Validity of the findings: I have no more comment. Additional comments: I have no more comment.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: STATE-OF-THE-ART VIOLENCE DETECTION TECHNIQUES IN VIDEO SURVEILLANCE SECURITY SYSTEMS: A SYSTEMATIC REVIEW Review round: 1 Reviewer: 1

Basic reporting: no comment Experimental design: no comment Validity of the findings: no comment Additional comments: This paper provided a survey by the assessment of video violence detection problems that occurs in state-of-the-arts for both qualitative and quantitative studies in terms of procedure, datasets, and performance indicators. Some challenges and future directions are also covered. The topic seems to be very interesting, however, I have some major concerns about the paper that will further enhance the paper quality and its body structure. The wording style of the abstract is too sloppy and instead of actual contents description, the background studies is largely covered. Authors need to make the abstract more compact and representative for the whole paper contents. A reader gets confuse when face the sections without numbering that is an important aspect of paper body. Resolve this issue. I did not find any novelty after going through the contribution lists that are already the practice of existing surveys such as methods coverage, their features extraction sets, datasets, etc. Authors are suggested to clearly highlight their contributions and mention why this survey is needed if there already exist several violence detection surveys. I did not find the any visual statistical information of year-wise violence detection papers distribution. Authors need to include the details of paper coverage from each year, present their taxonomy, broadly categorize them into machine learning or conventional techniques and deep learning by investigating each year for the category. Similarly, authors are suggested to present visual or tabular representation of working flow of the survey for the ease of readers. Next, I did not most recent violence detection literature that authors need to includes such as: An intelligent system for complex violence pattern analysis and detection. International Journal of Intelligent Systems. 2021 Jul 5 AI assisted Edge Vision for Violence Detection in IoT based Industrial Surveillance Networks. IEEE Transactions on Industrial Informatics. 2021 Sep 29 Violence detection and face recognition based on deep learning. Pattern Recognition Letters. 2021. Challenges and future directions are not well-structured as they are explained in wordy manners. It is suggested to explain each challenge in separate small section. Finally, I recommend to consider the English proficiency in terms of spelling and grammatical corrections.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: STATE-OF-THE-ART VIOLENCE DETECTION TECHNIQUES IN VIDEO SURVEILLANCE SECURITY SYSTEMS: A SYSTEMATIC REVIEW Review round: 1 Reviewer: 2

Basic reporting: This systematic review provide a comprehensive assessment of the video violence detection problems that have been described in state-of-the-art researches. This paper deals with an interesting topic especially in context of video violence. Therefore, it requires revisions to improve the quality of work. 1. More suitable title should be selected for the article. Please use different terms in the "Title" and the "Keywords". The abstract should be ordered by answering questions such as Originality of the manuscript, Objectives, Method (indicate how many papers you located in the different stages of the SLR and how many you were finally left with) and Finding. Experimental design: 2. I suggest a new Table with SLR and bibliometric analysis have been used in ``several previous research papers''. 3. A flowchart should be added to the article to show the research methodology. In this sense, I propose to update the methodology as this paper does: B. Kitchenham and S. Charters, "Guidelines for performing sandstematic literature reviews in software engineering version 2.3", Engineering, vol. 45, no. 5, pp. 1051, 2007. In this paper you will see that the flowchart in figure 2 presents the Number of articles in each stage after applying the inclusion and exclusion criteria. Abarca, V. M. G., Palos-Sanchez, P. R., & Rus-Arias, E. (2020). Working in virtual teams: a systematic literature review and a bibliometric analysis. IEEE Access, 8, 168923-168940. Validity of the findings: 4. Consider the length of the conclusions. 5. Add DOI for all references. 6. It is suggested to include a SLR protocol and question research. 7. It is suggested to compare the results of the present research with some similar systematic literature review which is done before. Additional comments: The manuscript is of interest and examines a problem of great interest. However, the methodology applied needs to be improved.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: STATE-OF-THE-ART VIOLENCE DETECTION TECHNIQUES IN VIDEO SURVEILLANCE SECURITY SYSTEMS: A SYSTEMATIC REVIEW Review round: 2 Reviewer: 1

Basic reporting: - Experimental design: - Validity of the findings: - Additional comments: The authors have made significant improvements. I accept the paper. I have added an optional minor comment. Section 4 (CLASSIFICATION OF VIOLENCE DETECTION TECHNIQUES), the text details in the subsections are too long. Please add few headings in each subsection for readers understanding.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: RANDOMNESS ANALYSIS OF END-TO-END DELAY IN RANDOM FORWARDING NETWORKS Review round: 1 Reviewer: 1

Basic reporting: In Line 12 in the abstract: both receiver and receiver I believe this is an unforgivable mistake, please modify it The introduction isn't written well This manuscript needs substantial copyediting and English writing revisions Literature references are insufficiently provided. Experimental design: No comment Validity of the findings: No comment Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: RANDOMNESS ANALYSIS OF END-TO-END DELAY IN RANDOM FORWARDING NETWORKS Review round: 1 Reviewer: 2

Basic reporting: The authors propose a mathematical model of random forwarding networks (FNs) and derive the expression of end-to-end delay distribution in different FNs. The topic is interesting, but the quality of the manuscript can be improved in terms of its problem novelty and main contribution. The following comments are provided for the authors’ consideration: 1. Please try to demonstrate more results in comparing different parameter settings and benchmarks. It would be better that some comparisons between existing works and the proposed algorithm are provided. 2. Regarding the system model part, is the model specified for only one sender and one receiver? What is the adjustment if we increase the number of senders and receivers? 3. The abstract should be revised. For instance, “As a random quantity easily obtained by both receiver and receiver …” should be replaced by “As a random quantity easily obtained by both sender and receiver”. 4. Some references are wrongly referred. For example, Zhang, J., Li, G., Marshall, A., Hu, A., and Hanzo, L. (2020). A new frontier for iot security emerging from three decades of key generation relying on wireless channels. IEEE Access, PP(99) should be corrected. In fact, the page numbers is 138406-138446 and the volume is 8. Please check that. 5. Some equations are not numbered. 6. The caption of Figure 8 should be revised (H_d as a function of p for m = 3 and N = 2; and put H_d (bits) as ylabel). Experimental design: Regarding the system model part, is the model specified for only one sender and one receiver? What is the adjustment if we increase the number of senders and receivers? Validity of the findings: Please try to demonstrate more results in comparing different parameter settings and benchmarks. It would be better that some comparisons between existing works and the proposed algorithm are provided. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: RANDOMNESS ANALYSIS OF END-TO-END DELAY IN RANDOM FORWARDING NETWORKS Review round: 2 Reviewer: 1

Basic reporting: the authors have addressed all the mentioned comments, i have no more comments Experimental design: The authors have addressed all the mentioned comments, i have no more comments Validity of the findings: The authors have addressed all the mentioned comments, i have no more comments Additional comments: The authors have addressed all the mentioned comments, i have no more comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: GPUTREESHAP: MASSIVELY PARALLEL EXACT CALCULATION OF SHAP SCORES FOR TREE ENSEMBLES Review round: 1 Reviewer: 1

Basic reporting: The submission is well-written and structured appropriately. The key ideas of the research are communicated clearly. The literature review and background are sufficient to understand the article. The results are presented clearly and are supported by the availability of source-code and datasets. Experimental design: There is a clear research question in that there is no (to the best of my knowledge) previous efficient implementation for computing the TreeShap algorithm on GPU architectures. This knowledge gap is filled by presenting and evaluating an optimised and interesting GPU implementation. The conclusions of the article are supported by the results. The experiments are well-designed and would be straightforward to replicate (very easily given the availability of the datasets and source-code used). Validity of the findings: As above, the conclusions of the the article are supported by the results. Additional comments: The GPU implementation described by this article presents both an insightful rearranging of the algorithm to make it suitable for GPU architectures as well as optimisation that shows a thorough understanding of the architectural features available.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: GPUTREESHAP: MASSIVELY PARALLEL EXACT CALCULATION OF SHAP SCORES FOR TREE ENSEMBLES Review round: 1 Reviewer: 2

Basic reporting: The writing is generally good. The introduction, literature review, and background are sufficient and well-written. Some minor issues are: 1. In Figure 1, please explain in the caption or text what the left/right arrows, solid/dash arrows mean for more clarification. 2. In Figure 2, it is better to indicate how these two unique paths correspond to the two leaf nodes (and which two) in Figure 1. 3. Please double-check Lines 22-24 of Algorithm 1, as well as related parts (Algorithm 2) and your code implementation. They are inconsistent with either (Lundberg et al, 2020) or their earlier publication (https://arxiv.org/abs/1802.03888). If it is not an error, can you confirm or explain the difference? 4. In Section 3.2, Line 184, please double-check which of "Line 12" or "Line 13" (of Algorithm 1) is more suitable for the context. 5. In Table 2, "rows" and "cols" need explanation. Does a column mean a feature? Does it include the class label? Why does fashion_mnist contain 785 columns while the image size is 28x28=784? For cal_housing, does it include longitude and latitude features? 6. In Line 305, LaTeX symbol \gg may be considered for "much greater than". 7. In Line 319, are s(i) and I defined before? 8. For easier reading/comparison, is it possible to reorganize Table 5 so that models and algs are in different dimensions (e.g. models in rows, algs in columns)? If not possible due to limited space, can horizontal split lines be added to separate different models? 9. Please double-check the caption of Table 7, as it is said in Line 373 of the main text that the number of test rows was reduced to 200 but the caption said 10000. Experimental design: The research and experiments are original and carried out well, supporting the original findings in general, but here is one issue/suggestion to enhance the experiments. In Line 339, the authors claimed that greater utilization directly translates to performance improvement. Is it possible to add additional experiments that support this statement? The authors may consider comparing the runtime speedups of BFD vs NF, or BFD vs None. Validity of the findings: Source code is publicly available. Refer to issue 3 in basic reporting. Otherwise no comment. Additional comments: Can GPUTreeSHAP result in exact SHAP values and interaction values, or just provide their approximations due to approximated heuristics in bin packing? If they are approximations, what is the error?

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: GPUTREESHAP: MASSIVELY PARALLEL EXACT CALCULATION OF SHAP SCORES FOR TREE ENSEMBLES Review round: 1 Reviewer: 3

Basic reporting: This paper is about a GPU implementation of the well-known machine learning interpretability method SHAP. The method is recursive polynomial time and until now was only possible to use it on relatively small datasets. The proposed GPU implementation achieves up to 19x speedups. I have not seen any similar papers. The paper is well-written and includes all the required sections. The proposed GPU algorithm is described using enough details and pseudocode is included in the paper. The code is available on github and the datasets are available online. Things to improve: - In section 2.3 GPU Computing, the authors could mention that both the LightGBM and Catboost have GPU implementations. - In section 5 Conclusions, the authors could comment on extending the proposed algorithms to the other gradient boosting algorithms. Is the proposed method good for solving other problems? - The abstract says 19x speedups, but section 4.2 says between 13-18x for medium and large models. - Please add a description of the multi-core CPU implementation. - The paper mentions the theoretical upper bounds of memory consumption, but no experimental memory consumption number are reported in the paper. Experimental design: The authors evaluate the proposed method using several sets of parameters for the XGBoost trees. The method is evaluated in 4 different datasets. I wish the authors would have run experiment with even larger datasets. Validity of the findings: The proposed GPU implementation in novel. The reported results could help researchers run experiments faster in various domains. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: GPUTREESHAP: MASSIVELY PARALLEL EXACT CALCULATION OF SHAP SCORES FOR TREE ENSEMBLES Review round: 2 Reviewer: 1

Basic reporting: No comment Experimental design: No comment Validity of the findings: No comment Additional comments: If the authors agree, the caption of Figure 1 may need more explanation on this question: Which arrow (left or right) corresponds to the path where the condition of the parent node is met? Otherwise, the manuscript may be accepted as is.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: GPUTREESHAP: MASSIVELY PARALLEL EXACT CALCULATION OF SHAP SCORES FOR TREE ENSEMBLES Review round: 2 Reviewer: 2

Basic reporting: No comment. Experimental design: No comment. Validity of the findings: No comment Additional comments: The current submitted updates are sufficient for paper publication.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: SOFT AND HARD SKILLS IDENTIFICATION: INSIGHTS FROM IT JOB ADVERTISEMENTS IN THE CIS REGION Review round: 1 Reviewer: 1

Basic reporting: - In general, I found the manuscript to be mostly well written. The text could however benefit from some academic proofreading with regards to more minor stuff related to punctuation and minor readability edits. -While Table 1 is briefly introduced at the start of “Related Work”, it could be introduced better to the reader. Currently, you mention many of these terms in the starting paragraph, but how they are shown in Table 1 is a bit obscure at first. In my opinion, by having a paragraph strictly introducing the structure of Table 1, the readability of the paper would be improved. The way the table classifies prior work based on sample size and used approach is commendable, however. -All other Tables could also be introduced to the reader in a more comprehensive manner. - How did you get the cluster names for Tables 2, 3, and 4? Did you name the clusters by yourself? This could be mentioned briefly in the manuscript. Experimental design: - Three research goals are mentioned in the introduction section, and the author has results for these goals in the results section. - How does this paper differ from the paper “Demand for skills on the labor market in the IT sector”, by the same author? By glancing over the previous paper, you also cite in your manuscript, it seems the novelty lies in the hard and soft skill classification and the association rule mining. I think the paper would be improved by clearly stating the differences between these two papers, perhaps towards the end of the introduction, near your contributions, with a sentence of two. - I find the way you report soft and hard skills identification confusing. How does semi-manual identification work exactly? Do you use the same list as Calanca et al.? If so, will you miss soft skills not contained in the list? Is this a validity threat that should be discussed? - While the code used to analyze the data is shared, the data itself is not. Thus, I was not able to rerun the analysis done by the author. Sharing the data with the completed manuscript would greatly improve the reproducibility of the work. Validity of the findings: - Personally, I didn't find anything surprising in your results. You compare your results to previous findings, noting the region specificity in certain skills. However, I don't see if your results miss something from previous results. That is, are some skills which are present and popular in previous findings not present in your results? Investigating this could perhaps expand both the results and discussion sections. - You mention that the study was done on the data related to the Commonwealth of Independent States. In my opinion, this contextual information is important enough, so it could be added to the title, e.g., “Insights from IT Job Advertisements in the CIS region”. At the very least, I think this should be mentioned in the abstract. - While the code used to analyze the data is shared, the data itself is not. Thus, I was not able to rerun the analysis done by the author. Sharing the data with the completed manuscript would greatly improve the reproducibility of the work. - Some limitations on the discussion section are already mentioned, but some of them are very brief. This could be expanded to its own section, e.g. internal, external, and construct validity. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: SOFT AND HARD SKILLS IDENTIFICATION: INSIGHTS FROM IT JOB ADVERTISEMENTS IN THE CIS REGION Review round: 1 Reviewer: 2

Basic reporting: Summary: The author of this paper studied the demand for skills in the IT-sphere to discover the mapping between required skill-sets and job occupations. The proposed methodology for skills identification uses natural language processing, hierarchical clustering, and association mining techniques. The dataset used covers the 2015–2019 period, with 351,623 observations. A set of 3,034 unique frequent skills was extracted and prepared for further standardization. 
 The results show explicit information about the combinations of "soft" and "hard" skills required for different professional groups. The author explains the goal of the study highlighting that the findings provide valuable insights for supporting educational organizations, human resource (HR) specialists, and state labor authorities in the renewal of existing knowledge about skill-sets for IT professionals. The paper is well structured and easy to follow. The related works are well presented, putting the study in context. The raw data have been shared, along with the code used to perform the study. The results include the definitions of all terms. Experimental design: The research is within the aims and scope of the journal. The research questions are defined on page 2. To help the reader the author might restructure the goals of the paper in a question format, assigning the number to each question, and providing a specific answer to each question in the results section. The goals of the study are relevant and meaningful, and it is stated how this study fills the gap in the literature. The methodology is well explained and performed to a high technical standard, and all the methods are described with sufficient details which allow replication studies. Validity of the findings: All the underlying data have been provided, and the statistics explained. However, I strongly encourage the author to dedicate a specific section of the paper to better explain the limitations of the study (threats to validity). Conclusions are well stated and linked to the research questions, but the author could help the reader in re-organizing the RQ as suggested in the previous section. Additional comments: Overall, the paper is well written and interesting. It fits the journal, but the presentation can be improved.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: SOFT AND HARD SKILLS IDENTIFICATION: INSIGHTS FROM IT JOB ADVERTISEMENTS IN THE CIS REGION Review round: 2 Reviewer: 1

Basic reporting: The author has improved on all points raised in my previous review well. Most commendable are the new threats section and the shared raw data. In my opinion, the manuscript can be published. Experimental design: - Validity of the findings: - Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: SOFT AND HARD SKILLS IDENTIFICATION: INSIGHTS FROM IT JOB ADVERTISEMENTS IN THE CIS REGION Review round: 2 Reviewer: 2

Basic reporting: The paper is clear and well written. Experimental design: The research is original and within Aims and Scope of the journal. Validity of the findings: All the data have been provided and the analysis are well explained. Additional comments: Thanks to the author for the revised version of the paper. I feel that the comments have been addressed and I think the paper can be accepted.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: RIDER WEED DEEP RESIDUAL NETWORK-BASED INCREMENTAL MODEL FOR TEXT CLASSIFICATION USING MULTIDIMENSIONAL FEATURES AND MAPREDUCE Review round: 1 Reviewer: 1

Basic reporting: In this paper, author presented a MapReduce model for text classification in big data. However, there are some limitations that must be addressed as follows. 1. The abstract is very lengthy and not attractive. Some sentences in abstract should be summarized to make it more attractive for readers. 2. In Introduction section, it is difficult to understand the novelty of the presented research work. This section should be modified carefully. In addition, the main contribution should be presented in the form of bullets. 3. The most recent work about text classification and big data should be discussed as follows (‘An intelligent healthcare monitoring framework using wearable sensors and social networking data’, ‘Traffic accident detection and condition analysis based on social networking data’, ‘Fuzzy Ontology and LSTM-Based Text Mining: A Transportation Network Monitoring System for Assisting Travel’, and ‘Merged Ontology and SVM-Based Information Extraction and Recommendation System for Social Robots’). 4. It is better to merge subsection 2.1 and 2.2. 5. The authors should avoid the use of too many colors in figure (see figure1). 6. Equations should be discussed deeply. 7. Captions of the Figures not self-explanatory. The caption of figures should be self-explanatory, and clearly explaining the figure. Extend the description of the mentioned figures to make them self-explanatory. 8. The whole manuscript should be thoroughly revised in order to improve its English. 9. More details should be included in future work. Experimental design: no comment Validity of the findings: no comment Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: RIDER WEED DEEP RESIDUAL NETWORK-BASED INCREMENTAL MODEL FOR TEXT CLASSIFICATION USING MULTIDIMENSIONAL FEATURES AND MAPREDUCE Review round: 1 Reviewer: 2

Basic reporting: The increasing demand for information and rapid growth of big data have dramatically increased textual data. The amount of different kinds of data has led to the overloading of information. For obtaining useful text information, the classification of texts is considered an imperative task. This paper develops a technique for text classification in big data using the MapReduce model. The goal is to design a hybrid optimization algorithm for classifying the text. This work is meaningful and potential in this field. Experimental design: This paper develops a technique for text classification in big data using the MapReduce model. Validity of the findings: The pre-pressing is done with the steaming process and stop word removal. In addition, the Extraction of imperative features is performed wherein SentiWordNet features, contextual features, and thematic features are generated. Furthermore, the selection of optimal features is performed using Tanimoto similarity. Additional comments: 1 This work should be polished by native English speaker. Some spelling and grammar mistakes should be avoided in this manuscript. 2 There are several typical machine learning classification model, such as SVM, neural network and so on. So, authors should compare the proposed method with other typical machine learning methods. 3 Some deep learning methods, including LSTM, should be compared with this method. 4 There are some typographical errors. Authors should polishted them.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: RIDER WEED DEEP RESIDUAL NETWORK-BASED INCREMENTAL MODEL FOR TEXT CLASSIFICATION USING MULTIDIMENSIONAL FEATURES AND MAPREDUCE Review round: 2 Reviewer: 1

Basic reporting: The increasing demand for information and rapid growth of big data has dramatically increased textual data. For obtaining useful text information, the classification of texts is considered an imperative task. Experimental design: This paper develops a hybrid optimization algorithm for classifying the text. Here, the pre-pressing is done by the stemming process and stop word removal. In addition, the extraction of imperative features is performed, and the selection of optimal features is performed using Tanimoto similarity, which estimates the similarity between the features and selects the relevant features with higher feature selection accuracy. After that, a deep residual network trained by the Adam algorithm is utilized for dynamic text classification. In addition, the dynamic learning is performed by the proposed Rider invasive weed optimization (RIWO)-based deep residual network along with fuzzy theory. The proposed RIWO algorithm combines Invasive weed optimization (IWO) and the Rider optimization algorithm (ROA). Validity of the findings: The aim is to devise an optimization-driven deep learning technique for classifying the texts using the MapReduce framework. Initially, the text data undergoes pre-processing for removing unnecessary words. Here, the pre-processing is performed using the stop word removal and stemming process. After that, the features, such as SentiWordNet features, thematic features, and contextual features, are extracted. These features are employed in a deep residual network for classifying the texts. Here, the deep residual network training is performed by the Adams algorithm. Finally, dynamic learning is carried out wherein the proposed RIWO-based deep residual network is employed for incremental text classification. Here, the fuzzy theory is employed for weight bounding to deal with the incremental data. Additional comments: This paper develops a hybrid optimization algorithm for classifying the text. Here, the pre-pressing is done by the stemming process and stop word removal. In addition, the extraction of imperative features is performed, and the selection of optimal features is performed using Tanimoto similarity, which estimates the similarity between the features and selects the relevant features with higher feature selection accuracy. This work is meaningful and potential in this field. 1 Some errors, including spelling errors and grammar ones should be polished. 2 Some typical references should be discussed in this work.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A STORAGE-EFFICIENT ENSEMBLE CLASSIFICATION USING FILTER SHARING ON BINARIZED CONVOLUTIONAL NEURAL NETWORKS Review round: 1 Reviewer: 1

Basic reporting: The authors submitted a very interesting work, my suggestions to improve it are: 1.I suggest to add it to github “The source code for reproducing the experiments will be available upon publication of the manuscript” 2.The contribution is no introduced clearly on the theoretical analysis; also if very good results are reported. If feasible add some more details on the theoretical analysis of your approach. 3.You should better review the literature on ensemble of classifiers, e.g. https://arxiv.org/abs/1802.03518 https://doi.org/10.1016/j.eswa.2020.114048 https://arxiv.org/pdf/2104.02395.pdf 4.Some typos, E.g. row 153 “BNN model can be used in a low poer” -> “...power” Experimental design: 5.You run many experiments and you have reported many results, this is appreciated, anyway please better stress the novelty of your method respect the literature on pruning and quantization approaches: https://www.sciencedirect.com/science/article/pii/S0031320321000868 https://arxiv.org/pdf/2103.13630.pdf Validity of the findings: no comment Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A STORAGE-EFFICIENT ENSEMBLE CLASSIFICATION USING FILTER SHARING ON BINARIZED CONVOLUTIONAL NEURAL NETWORKS Review round: 1 Reviewer: 2

Basic reporting: 1.The last word on line 153 is wrong. Suggest to examine each word and sentence carefully. 2. Suggest to polish the language of the writing. Experimental design: No comment Validity of the findings: Figure 5(b) does not contained data when in the fusion ensemble due to the limitation of GPU resources. I suggest you to supplement the result when , in order to ensure the integrity of the experiment. You can use smaller batch size or use CPU for retraining. Additional comments: In this manuscript, the authors proposed a storage-efficient ensemble classification to overcome the low inference accuracy of binary neural networks (BNNs). The work indicates that proposed method reduces the storage burden of multiple classifiers in the lightweight system. This is a good idea, which can be used to improve the accuracy and reduce the storage burden of BNN. In addition, this manuscript also provides a solution for the application of neural network in lightweight system. There are some suggestions as follows: 1. Fusion, voting, and bagging schemes were applied to evaluate ensemble-based systems. However, there is no comparison between these methods in the paper. 2. Your introduction at lines 296-310 needs more detail. I suggest that you can add figure to describe the experiment result.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A STORAGE-EFFICIENT ENSEMBLE CLASSIFICATION USING FILTER SHARING ON BINARIZED CONVOLUTIONAL NEURAL NETWORKS Review round: 1 Reviewer: 3

Basic reporting: This is the review report of the paper entitled "A storage-efficient ensemble classification using filter sharing on binarized convolutional neural networks". The paper presents a very important topic and is well presented. However, I have some comments to improve the paper. 1- In the abstract, add the value of classification accuracy to support the theory. 2- The authors show the results of their proposed method with the state-of-the-art model (ResNet), I would suggest showing the results of the ResNet on the same dataset without the use of the proposed method. 3- training parameters are required to mention. 4- Paper code with a nice demo is important to upload on any public platform. 5- I would suggest citing the following reference when referring to CNN so new reference from 2021 can be used https://link.springer.com/article/10.1186/s40537-021-00444-8 6-Comparison with state-of-the-art is necessary to add on the same used dataset. 7-explain more on the research gap of previous methods. 8-The contributions of the article have to be clear for the readers, I would suggest making them as bullet points at the end of the introduction. Experimental design: See the first box " Basic reporting" Validity of the findings: See the first box " Basic reporting" Additional comments: See the first box " Basic reporting"

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A STORAGE-EFFICIENT ENSEMBLE CLASSIFICATION USING FILTER SHARING ON BINARIZED CONVOLUTIONAL NEURAL NETWORKS Review round: 2 Reviewer: 1

Basic reporting: Revision well done Experimental design: Revision well done Validity of the findings: Revision well done Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A STORAGE-EFFICIENT ENSEMBLE CLASSIFICATION USING FILTER SHARING ON BINARIZED CONVOLUTIONAL NEURAL NETWORKS Review round: 2 Reviewer: 2

Basic reporting: No comment Experimental design: No comment Validity of the findings: No comment Additional comments: No comment

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A STORAGE-EFFICIENT ENSEMBLE CLASSIFICATION USING FILTER SHARING ON BINARIZED CONVOLUTIONAL NEURAL NETWORKS Review round: 2 Reviewer: 3

Basic reporting: The authors addressed the comments in a very good way. Experimental design: The authors addressed the comments in a very good way. Validity of the findings: The authors addressed the comments in a very good way. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: LORM: A NOVEL REINFORCEMENT LEARNING FRAMEWORK FOR BIPED GAIT CONTROL Review round: 1 Reviewer: 1

Basic reporting: This paper proposes a biped control framework based on reinforcement learning. The expert trajectory of the traditional controller is introduced to accelerate the training. And the exploration of reinforcement learning ensures the final model outperforms the expert instead of simply imitating the expert. To improve the training efficiency and performance, some improvements are also introduced into the framework. The method is validated by various experiments, including two tasks (walking as fast as possible & tracking specific velocity) and several different environments (plain, up-hill, down-hill and uneven floor). The work is conducted well and is promising in different control tasks. Both the framework and the tricks are inspiring for other works. Experimental design: Problems in the content and explanation of the paper: 1. The description of the PPO algorithm is not detailed enough. The meaning of lambda and gamma in Table1 and their usage are not illustrated. 2. The experiments are rich for the LORM models. However, the performance of reference motion should be illustrated to show the improvement or difference between the proposed algorithm and reference motion. Validity of the findings: Other details should also be edited: 1. Some paragraphs have indentation while others do not. For example Line 295 and Line 297; Line 273 and Line 276. 2. Please avoid capital letters in sentences. For example, Line 15 “ Learn and Outperform the Reference Motion (LORM), an RL based framework ... “ in the abstract. 3. Equation 17: f(x) = ..., however, there is no variable x or function f(x). I think it should be Criterion = ... . 4. The names of curves in Figures can be polished, for example, Fig11. And the captions can be used in figures to make the meaning of curves more clear. 5. The language should be further polished. Especially in the subsection “Symmetrization of actions and states”, I wonder whether it can be more clear? Though the description is understandable, it takes time to read and understand it. Additional comments: In addition, I have two questions to be answered by the authors: 1. In the subsection “Symmetrization of actions and states”, why only the angles of joints are symmetrized while other observations keep unchanged? 2. The input observation of the agent contains many items which can be obtained in simulation software (base position, position of centre of mass). However, is it possible to obtain them in the real world?

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: LORM: A NOVEL REINFORCEMENT LEARNING FRAMEWORK FOR BIPED GAIT CONTROL Review round: 1 Reviewer: 2

Basic reporting: An RL based framework for gait controlling of biped robot is proposed in this paper to overcome the complications of dynamics design and calculation. The results validated the efficiency and the advantages of the proposed method. However, there are several suggestions for the authors 1. As the proposed method is claimed to be a novel method, there should be more literature discussion in the introduction part to clarify the state-of-art of the field and thus the novelty of the paper. 2. In the result and discussion part, it is better to compare and validate the result with published works to make it more convincing. 3. There are some typo and grammar errors in the paper, please give it a proofreading for the language check. The results are sufficient enough to validate the aim of the paper. However, more discussions are expected to emphasize the novelty and significance of the method. Experimental design: In the result and discussion part, it is better to compare and validate the result with published works to make it more convincing. Validity of the findings: As the proposed method is claimed to be a novel method, there should be more literature discussion in the introduction part to clarify the state-of-art of the field and thus the novelty of the paper. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: LORM: A NOVEL REINFORCEMENT LEARNING FRAMEWORK FOR BIPED GAIT CONTROL Review round: 2 Reviewer: 1

Basic reporting: The manuscript has been well-revised. Literature review is sufficient with good background provided. The results contains clear definition of all terms. Experimental design: The experiment design is clear with sufficient details and justified. The manuscript has met the standards of the journal. Validity of the findings: The findings have been rigorously validated with sufficient details. Conclusions are well stated. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: LORM: A NOVEL REINFORCEMENT LEARNING FRAMEWORK FOR BIPED GAIT CONTROL Review round: 2 Reviewer: 2

Basic reporting: all the issues I raised have been addressed Experimental design: all the issues I raised have been addressed Validity of the findings: all the issues I raised have been addressed Additional comments: all the issues I raised have been addressed

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: VEGETATION INDICES’ SPATIAL PREDICTION BASED NOVEL ALGORITHM FOR DETERMINING TSUNAMI RISK AREAS AND RISK VALUES Review round: 1 Reviewer: 1

Basic reporting: 1. There is a need to check the grammar, spelling, sentence structures, etc. of the paper. In the title “A New Algorithm for …….on Spatial Predicition of 3 Vegetation Index”. The word “Index” should be in plural form since more than one index is used in the paper. In other parts of the paper, I found several errors on the spelling of words, verb agreement, correct usage of articles, etc. There is a need to go over the entire paper to check on these errors and reconstruct or rephrase some sentences which are redundant (e.g.Lines 239-242 needs to be reconstructed to avoid redundancy). I would suggest that the authors use an APP to correct these errors. 2. A reorganization of the Abstract is recommended. The results of the algorithm should be written in the last lines and the presentation of the algorithm and the indexes used should be more profound to emphasize the objective of the paper. 3. The tsunami risk levels (Levels 1, 2, and 3) are based on indicators resulting from the algorithm (Lines 358-360). First, the basis of these indicators was not explicitly explained. Second, how would you classify the risk if TRUE is within the range (1,2) or (2,3)? Is it always the case that the algorithm will result to TRUE=2 in your simulations? 4. In your regression model (Equation 10), the authors need to explain how they obtained the constant coefficients (e.g. 0.95, 0.007, etc.) 5. Comments on Figures Are the location maps taken by the authors themselves? If not, then there is a need to indicate the source below the figure. Figure 3, although it’s a captured image, its’ source should be written below. Also, the figure is more likely to be identified as a table rather than a figure. The x-axis and y-axis must be given labels on Figures 5,6,7,8 and the titles whould be more clear and concise. In Figures 9,10,11, the legend uses a color coding to show the level of risk is not properly defined. Since you are identifying the risk levels with color codes, some colors are not properly identified. Also, it would be appropriate to give the range of values for the risk levels so that readers of the paper can properly distinguish the different risk levels in the graph. 6. Comments on Tables Since the purpose of the authors is to make a comparative analysis on the amount of error generated using different algorithms and that of their proposed algorithm, I would suggest either of the following improvements: a. Tables 2,3,4,5 can be reconstructed as one table showing the amount of error (MSE. RMSE, MAE) of each algorithm based on each index, or b. Three separate tables for each classified error (i.e. MSE. RMSE, MAE) comparing the algorithms and the amount of error based on the 3 indexes. Since each table enumerates the amount of error of each index with prediction results using the varied machine learning algorithms in your study, it would be better to fuse all 4 tables as one where all the algorithms are listed together with the indexes and the classification of errors. In this way, you can make a clear comparison of the amount of error. Alternatively, you may create three tables, each for the identified error MSE, RMSE, MAE separately, and compare the results of the algorithms. Experimental design: the experimental design is well explained and detailed Validity of the findings: The findings are valid based on the provided data. The objectives of the study were achieved. Additional comments: No additional comments