Datasets:

MihirRajeshPanchal

/

IITPatnaPeerReview

like 0

The answers to the critiques referenced in the paper are convincing, though I must admit that I don't know how crucial it is to answer these critics, since it is difficult to assess wether they reached or will reach a large audience.[answers-POS], [IMP-NEU]

answers

IMP

POS

NEU

Details: - p. 4 please do not qualify KL as a distance metric [null], [EMP-NEG]

EMP

NEG

- Section 4.3: Every GAN variant was trained for 200000 iterations, and 5 discriminator updates were done for each generator update is ambiguous: what is exactly meant by iteration (and sometimes step elsewhere)?[Section-NEU], [EMP-NEU]

Section

EMP

NEU

NEU

- Section 4.3: the performance measure is not relevant regarding distributions. The l2 distance is somewhat OK for means, but it makes little sense for covariance matrices. [Section-NEU], [EMP-NEU]

Section

EMP

NEU

NEU

For the JCP-S model, the loss function is unclear to me.[model-NEG], [EMP-NEG]

model

EMP

NEG

NEG

L is defined for 3rd order tensors only; how is the extended to n > 3?[null], [EMP-NEU]

EMP

NEU

Intuitively it seems that L is redefined, and for, say, n 4, the model is M(i,j,k,n) sum_1^R u_ir u_jr u_kr u_nr.[null], [EMP-NEU]

EMP

NEU

However, the statement since we are using at most third order tensors in this work I am further confused.[statement-NEG], [EMP-NEG]

statement

EMP

NEG

NEG

Is it just that JCP-S also incorporates 2nd order embeddings?[null], [EMP-NEU]

EMP

NEU

I believe this requires clarification in the manuscript itself.[manuscript-NEU], [EMP-NEG]

manuscript

EMP

NEU

NEG

For the evaluations, there are no other tensor-based methods evaluated, although there exist several well-known tensor-based word embedding models existing: Pengfei Liu, Xipeng Qiuu2217 and Xuanjing Huang, Learning Context-Sensitive Word Embeddings with Neural Tensor Skip-Gram Model, IJCAI 2015 Jingwei Zhang and Jeremy Salwen, Michael Glass and Alfio Gliozzo.[evaluations-NEG], [CMP-NEU]

evaluations

CMP

NEG

NEU

Additionally, since it seems the main benefit of using a tensor-based method is that you can use 3rd order cooccurance information, multisense embedding methods should also be evaluated.[methods-NEU], [EMP-NEU]

methods

EMP

NEU

NEU

There are many such methods, see for example Jiwei Li, Dan Jurafsky, Do Multi-Sense Embeddings Improve Natural Language Understanding?[methods-NEU], [EMP-NEU]

methods

EMP

NEU

NEU

and citations within, plus quick googling for more recent works.[citations-NEU], [EMP-NEU]

citations

EMP

NEU

NEU

I am not saying that these works are equivalent to what the authors are doing, or that there is no novelty, but the evaluations seem extremely unfair to only compare against matrix factorization techniques, when in fact many higher order extensions have been proposed and evaluated, and especially so on the tasks proposed (in particular the 3-way outlier detection).[novelty-NEU, evaluations-NEG], [CMP-NEG, EMP-NEG]

novelty

evaluations

CMP

EMP

NEU

NEG

NEG

NEG

Observe also that in table 2, NNSE gets the highest performance in both MEN and MTurk.[table-NEU], [EMP-NEU]

table

EMP

NEU

NEU

Frankly this is not very surprising; matrix factorization is very powerful, and these simple word similarity tasks are well-suited for matrix factorization.[null], [EMP-NEG]

EMP

NEG

So, statements like as we can see, our embeddings very clearly outperform the random embedding at this task is an unnecessary inflation of a result that 1) is not good[statements-NEG, result-NEG], [EMP-NEG]

statements

result

EMP

NEG

NEG

NEG

and 2) is reasonable to not be good.[null], [EMP-NEG]

EMP

NEG

Overall, I think for a more sincere evaluation, the authors need to better pick tasks that clearly exploit 3-way information and compare against other methods proposed to do the same.[evaluation-NEU], [EMP-NEG]

evaluation

EMP

NEU

NEG

The multiplicative relation analysis is interesting,[analysis-POS], [EMP-POS]

analysis

EMP

POS

POS

but at this point it is not clear to me why multiplicative is better than additive in either performance or in giving meaningful interpretations of the model.[performance-NEU, model-NEU], [EMP-NEG]

performance

model

EMP

NEU

NEU

NEG

In conclusion, because the novelty is also not that big (CP decomposition for word embeddings is a very natural idea) I believe the evaluation and analysis must be significantly strengthened for acceptance. [novelty-NEG], [NOV-NEG, IMP-NEG, REC-NEG]

novelty

NOV

IMP

REC

NEG

NEG

NEG

NEG

Summary: The authors take two pages to describe the data they eventually analyze - Chinese license plates (sections 1,2), with the aim of predicting auction price based on the luckiness of the license plate number.[null], [EMP-NEU]

EMP

NEU

The authors mentions other papers that use NN's to predict prices, contrasting them with the proposed model by saying they are usually shallow not deep, and only focus on numerical data not strings.[papers-NEU, proposed model-NEU], [CMP-NEU]

papers

proposed model

CMP

NEU

NEU

NEU

In section 7, the RNN is combined with a handcrafted feature model he criticized in a earlier section for being too simple to create an ensemble model that predicts the prices marginally better.[section-NEU], [CMP-NEU]

section

CMP

NEU

NEU

Sec 3 The author does not mention the following reference: Deep learning for stock prediction using numerical and textual information by Akita et al. that does incorporate non-numerical info to predict stock prices with deep networks.[Sec-NEG], [PNF-NEG]

Sec

PNF

NEG

NEG

Sec 4 What are the characters embedded with? This is important to specify.[Sec-NEU], [EMP-NEU]

Sec

EMP

NEU

NEU

Is it Word2vec or something else?[null], [EMP-NEU]

EMP

NEU

What does the lookup table consist of?[table-NEU], [EMP-NEU]

table

EMP

NEU

NEU

References should be added to the relevant methods.[References-NEU], [EMP-NEU]

References

EMP

NEU

NEU

Sec 5 I feel like there are many regression models that could have been tried here with word2vec embeddings that would have been an interesting comparison.[Sec-NEU], [SUB-NEU, CMP-NEU]

Sec

SUB

CMP

NEU

NEU

NEU

LSTMs as well could have been a point of comparison.[null], [EMP-NEU]

EMP

NEU

Sec 6 Nothing too insightful is said about the RNN Model.[Sec-NEG], [SUB-NEG]

Sec

SUB

NEG

NEG

Sec 7 The ensembling was a strange extension especially with the Woo model given that the other MLP architecture gave way better results in their table.[Sec-NEG, results-NEG], [CMP-NEG]

Sec

results

CMP

NEG

NEG

NEG

Overall: This is a unique NLP problem, and it seems to make a lot of sense to apply an RNN here, considering that word2vec is an RNN.[problem-NEU], [EMP-NEU]

problem

EMP

NEU

NEU

However comparisons are lacking and the paper is not presented very scientifically.[comparisons-NEG, paper-NEG], [SUB-NEG, CMP-NEG, PNF-NEG]

comparisons

paper

SUB

CMP

PNF

NEG

NEG

NEG

NEG

NEG

The lack of comparisons made it feel like the author cherry picked the RNN to outperform other approaches that obviously would not do well.[comparisons-NEG, approaches-NEG], [SUB-NEG, CMP-NEG]]

comparisons

approaches

SUB

CMP

NEG

NEG

NEG

NEG