elmurod1202
commited on
Commit
•
91af102
1
Parent(s):
4088ec2
Update README.md
Browse files
README.md
CHANGED
@@ -1,9 +1,9 @@
|
|
1 |
# uzbek-sentiment-analysis
|
2 |
-
Sentiment analysis in Uzbek language and new Datasets of Uzbek App reviews for Sentiment Classification
|
3 |
|
4 |
Feel free to use the dataset and the tools presented in this project, a paper about more details on creation and usage [here](http://www.grupolys.org/biblioteca/KurMatAloGom2019a.pdf).
|
5 |
|
6 |
-
If you find it useful,
|
7 |
```
|
8 |
@inproceedings{kuriyozov2019deep,
|
9 |
author = {Kuriyozov, Elmurod and Matlatipov, Sanatbek and Alonso, Miguel A and Gómez-Rodríguez, Carlos},
|
@@ -18,11 +18,11 @@ If you find it useful, plese make sure to cite the paper:
|
|
18 |
|
19 |
The main contributions of this project are:
|
20 |
|
21 |
-
1. The creation of the first annotated dataset for sentiment analysis in Uzbek language, obtained from reviews of the top 100 Google Play Store applications used in Uzbekistan. This manually annotated dataset contains 2500 positive
|
22 |
|
23 |
-
2. The
|
24 |
|
25 |
|
26 |
The results obtained through the research:
|
27 |
-
![Main Results Table](
|
28 |
Table: Accuracy results with different training and test sets.ManualTT- Manually annotated Training and Test sets.TransTT- Translated Training and Test sets.TTMT- Translated dataset for Training, Annotated dataset for Test set.
|
|
|
1 |
# uzbek-sentiment-analysis
|
2 |
+
Sentiment analysis in the Uzbek language and new Datasets of Uzbek App reviews for Sentiment Classification
|
3 |
|
4 |
Feel free to use the dataset and the tools presented in this project, a paper about more details on creation and usage [here](http://www.grupolys.org/biblioteca/KurMatAloGom2019a.pdf).
|
5 |
|
6 |
+
If you find it useful, please make sure to cite the paper:
|
7 |
```
|
8 |
@inproceedings{kuriyozov2019deep,
|
9 |
author = {Kuriyozov, Elmurod and Matlatipov, Sanatbek and Alonso, Miguel A and Gómez-Rodríguez, Carlos},
|
|
|
18 |
|
19 |
The main contributions of this project are:
|
20 |
|
21 |
+
1. The creation of the first annotated dataset for sentiment analysis in the Uzbek language, obtained from reviews of the top 100 Google Play Store applications used in Uzbekistan. This manually annotated dataset contains 2500 positive and 1800 negative reviews. Furthermore, we have also built a larger dataset by automatically translating (using Google Translate API) an existing English dataset of application reviews. The translated dataset has≈10K positive and≈10K negative app reviews, after manually eliminating the major machine translation errors by either correcting or removing them completely.
|
22 |
|
23 |
+
2. The definition of the baselines for sentiment analyses in Uzbek by considering both traditional machine learning methods as well as recent deep learning techniques fed with fastText pre-trained word embeddings. Although all the tested models are relatively accurate and differences between models are small, the neural network models tested do not manage tosubstantially outperform traditional models. We believe that the quality of currently available pre-trained word embeddings for Uzbek is not enough to let deep learning models perform at their full potential.
|
24 |
|
25 |
|
26 |
The results obtained through the research:
|
27 |
+
![Main Results Table](results-table.png)
|
28 |
Table: Accuracy results with different training and test sets.ManualTT- Manually annotated Training and Test sets.TransTT- Translated Training and Test sets.TTMT- Translated dataset for Training, Annotated dataset for Test set.
|