A Deep Learning Approach for Sentiment Analysis of COVID-19 Reviews

Round 1

Reviewer 1 Report

1) Looking at some latest work on AI/Machine-Learning based Covid Application; some latest literature seem missing. AI and IoT have been predominant these days. Please include some of them in the introduction/related work to let readers know the latest trends. One example of such a missing article would be:
a) IVACS: Intelligent voice Assistant for Coronavirus Disease (COVID-19) Self-Assessment 

You can find a some more on this: b) Service Robots: Trends and Technology

However, there are more than these. I suggest authors update their literature.

2) Figure 1 is more like a testing; please restructure/add figure to define the training process as well. That is your overall architecture.

3) Why was LSTM used why nor ReLU? Based on the experiment?

4) How as optimal C and gamma parameters in SVM found to avoid overfitting?

5) Comparing this work with the work done by other researchers is recommended.

 

Author Response

Hello Reviwer,

Kindly see the attached document for your review.

Regards

Chetanpal Singh

 

Author Response File: Author Response.docx

Reviewer 2 Report

This paper proposes LSTM-RNN based method which uses an enhanced feature transformation framework via attention mechanism for Sentiment Analysis of COVID-19. Below I give some of the limitations which I have identified and my suggestions for enriching the paper:

1. I recommend you re-write the main contributions of this paper to the field with simple and clear sentences. I think that it is a little bit hard for readers to select the main contributions from long paragraphs (lines 55-74). 
2. Please add one figure that can explain your improved part of the LSTM-RNN network in Section 3. For example, it is maybe the overall network structure of the LSTM-RNN network (green color) and your contributions (red color).
3. Please check the format and style of 13, 14, 15 equations.
4. Many existing methods and algorithms (such as Naïve Bayes, logistic regression, Random Forest, etc.) are used but did not cite appropriately with reference numbers. Please cite them in Table 6 as well.
5. A more detailed evaluation and comparison of the state of the art would be appropriate. For instance, Precision-Recall curve and so on. 
6. To make the work effectively comparable, benchmarks and evaluations should be performed in comparison to the other more famous approaches in the field.
7. Add a new section named “Limitation and Discussion” to give a limitation of the proposed method and future research gaps in this field.
8. Please check the style and format of references.
9. I am not quite sure about the scientific novelty of this paper since almost all equations were used from other works.

Author Response

Hello Reviwer,

Kindly see the attached document for your review.

Regards

Chetanpal Singh

Author Response File: Author Response.docx

Reviewer 3 Report

In my opinion scientific level of this publication is below requirements of this Journal.

In particular, I do not see any breakthrough results compared to current and commonly known methods.

Author Response

Hello Reviwer,

Kindly see the attached document for your review.

Regards

Chetanpal Singh

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The revised version addresses my comment. I have no further concerns.

Author Response

Thank you for accepting the changes.

Author Response File: Author Response.docx

Reviewer 2 Report

The authors responded fully to my comments and developed the manuscript. I think that the manuscript can be accepted in the present form.

Author Response

Thank you for accepting the changes.

Author Response File: Author Response.docx

Reviewer 3 Report

Major comments:

 

The reviewed paper proposes an interesting problem, which is not only theoretically, but could be also practically important. Authors focused on Sentiment Analysis, Deep Learning and Classification on COVID-19.

 

Authors study an user-generated multi-media content such as image, text, video, and speech has recently become more popular on social media sites for people to share their ideas and opinions. One of the most popular social media sites for providing public sentiment towards events that occurred throughout the COVID-19 period is Twitter. This is because Twitter posts are short and constantly being generated. This paper presents a deep learning approach for sentiment analysis of tweet data where the dataset is based on COVID-19 reviews on Twitter. The proposed algorithm is based on LSTM -RNN based network and enhanced featured weighting by attention layer. This algorithm uses an enhanced feature transformation framework via attention mechanism. A total of four class labels (sad, joy, fear, and anger) from publicly available data of more than hundred thousand of tweets posted in Kaggle database are used in this study. Over the years, several approaches have developed for sentiment analysis. Scholars have tested the efficiency of these approaches. However, analysing huge amount data and achieving high accuracy remains a challenge. This paper presents a deep learn- ing approach for sentiment analysis of tweet data where the dataset is based on COVID- 19 reviews on Twitter. The algorithm is based on LSTM -RNN based network and en- hanced featured weighting by attention layer. This algorithm uses an enhanced feature transformation framework via attention mechanism. A total of four class labels (sad, joy, fear, and anger) from publicly available data of more than hundred thousand of tweets posted in Kaggle database are used in this study. Based on the use of attention layers with the existing LSTM-RNN approach, in comparison with the current approaches, the pro- posed deep learning approach significantly improves the performance metrics by 20% in accuracy, 10% to 12% in precision, but only 12-13% in recall. Out of a total of 179,108 COVID-19 related tweets, tweets with positive, neutral, and negative sentiments are found to be 45%, 30%, and 25%, respectively. Overall, the proposed approach is found to be efficient, practical and can be easily implemented for sentiment classification of COVID-19 reviews. This study is not free from limitations. The main limitation is with the work of features mapping. The feature weighting and features mapping are considered in the original dataset, further features which are noisy, and a combination of which may affect the clas- sification outcomes. Sometimes it is quite important, and the information vanishes during the experimental testing as we are providing weight to our features which can cause issues with the features which are not noisy. So, this has been a significant limitation of this re- search. In future, the enhancement of this work should use the validation or an optimization approach which should use by optimizing the features in an iterative process and converge for which they may implement metaheuristic or stochastic approach.

 

The paper has a logical structure and is clearly, concisely and accurately written.

 

I would suggest to update Abstract and Conclusions to highlight most important findings of this research.

 

The Introduction part should be updated, the authors did not clearly show the difference between their approach and those in the literature. Complex and expanded state-of-the-art is needed.

 

I also suggest author should add discussion about pros and cons of considered problem to clearly identify the benefits of the introduced approach.

 

It would be intersting for readers if the paper include theoretical section about extended number of different potential practical applications of presented approach in different areas. I would suggest to add it.

 

I strongly suggest to add all [1]-[2] appropriate references from the list below:

 

[1] 10.3390/ijerph19031416

 

[2] 10.3390/app11188438

 

Minor comments:

 

Paper contains some amount of typos that need to be corrected throughout the paper. There are several minor language errors in the text. Some sentences require rewriting.

Author Response

 

Point 1: I would suggest to update Abstract and Conclusions to highlight most important findings of this research.

Response 1: Both Abstract and Conclusion sections have been rewritten to show the main contribution and findings of this research.

 

Point 2: The Introduction part should be updated, the authors did not clearly show the difference between their approach and those in the literature. Complex and expanded state-of-the-art is needed.

Response 2: Lines 61-79 have been rewritten and two paragraphs are added on 80-91 for justifying the development of the proposed deep learning approach in this research area.

 

Point 3: I also suggest author should add discussion about pros and cons of considered problem to clearly identify the benefits of the introduced approach.

Response 3: Lines 417-430 have been rewritten to discuss about the problems with sentiment analysis process and the justification of the proposed deep learning approach in dealing with this kind of decision-making problem.

 

Point 4: It would be interesting for readers if the paper include theoretical section about extended number of different potential practical applications of presented approach in different areas. I would suggest to add it.

Response 4: A paragraph is added on lines 432-439 for discussing about the theoretical and practical implications of the study.

 

Point 5: I strongly suggest to add all [1]-[2] appropriate references from the list below:

[1] 10.3390/ijerph19031416

[2] 10.3390/app11188438

Response 5: The two references are now added and discussion of these two articles are included in Section 2 on page 4.

 

Point 6: Paper contains some amount of typos that need to be corrected throughout the paper. There are several minor language errors in the text. Some sentences require rewriting.

Response 6: Proof-reading has been done to correct grammar mistakes and improve the readability of the paper.

 

 

Author Response File: Author Response.docx

Round 3

Reviewer 3 Report

The current version contains the required materials as suggested and it is acceptable for publication in the Journal.

I have no further suggestions regarding the improvements of the content.

 

 

 

 

User-generated multi-media content such as image, text, video, and speech has recently 10 become more popular on social media sites for people to share their ideas and opinions. One of the 11 most popular social media sites for providing public sentiment towards events that occurred 12 throughout the COVID-19 period is Twitter. This is because Twitter posts are short and constantly 13 being generated. This paper presents a deep learning approach for sentiment analysis of Twitter 14 data on COVID-19 reviews. The proposed algorithm is based on LSTM-RNN based network and 15 enhanced featured weighting by attention layer. This algorithm uses an enhanced feature transfor- 16 mation framework via attention mechanism. A total of four class labels (sad, joy, fear, and anger) 17 from publicly available Twitter data posted in Kaggle database are used in this study. Based on the 18 use of attention layers with the existing LSTM-RNN approach, the proposed deep learning ap- 19 proach significantly improves the performance metrics by 20% in accuracy, 10% to 12% in precision, 20 but only 12-13% in recall as compared with the current approaches. Out of a total of 179,108 COVID- 21 19 related tweets, tweets with positive, neutral, and negative sentiments are found to be 45%, 30%, 22 and 25%, respectively. This shows that the proposed deep learning approach is found to be efficient, 23 practical and can be easily implemented for sentiment classification of COVID-19 reviews.

Over the years, several approaches have developed for sentiment analysis of social 417 media data. This sentiment analysis process is usually complex and time consuming, due 418 to the huge amount data and the requirement to achieve a high level of accuracy. Thus, 419 this paper presents a deep learning approach for sentiment analysis of Twitter data on 420 COVID-19 reviews. The algorithm is based on LSTM -RNN based network and enhanced 421 featured weighting by attention layer. This algorithm uses an enhanced feature transfor- 422 mation framework via attention mechanism. A total of four class labels (sad, joy, fear, and 423 anger) from publicly available Twitter data posted in Kaggle database are used in this 424 study. In comparison with the current approaches, the proposed deep learning approach 

significantly improves the performance metrics by 20% in accuracy, 10% to 12% in preci- 426 sion, but only 12-13% in recall. Out of a total of 179,108 COVID-19 related tweets, tweets 427 with positive, neutral, and negative sentiments are found to be 45%, 30%, and 25%, re- 428 spectively. Overall, the proposed deep learning approach is found to be efficient, practical 429 and can be easily implemented for sentiment classification of COVID-19 reviews. 430

This study provides theoretical and practical implications. For theoretical implica- 431 tions, this study applies a deep learning approach for sentiment analysis of individuals 432 from Twitter data on information regarding COVID-19. This deep learning approach can 433 be further applied to sentiment analysis of a general decision-making problem in various 434 industries such as marketing, government, service and academic. For practical implica- 435 tions, this study suggests that the proposed deep learning approach can be adopted and 436 modified for achieving a good level of accuracy especially when considering the complex- 437 ities entailed in textual analysis. 438

This study is not free from limitations. The feature weighting and features mapping 439 are considered in the original dataset, and further features which are noisy, as well as a 440 combination of these factors may affect the classification outcomes. In future work, the 441

deep learning approach can also be enhanced to formed simultaneously.

can be designed to optimise the features in an iterative process It 442 work with topic detection and sentiment classification to be per- 443

A Deep Learning Approach for Sentiment Analysis of COVID- 2

19 Reviews

 

Twitter; Sentiment Analysis; Deep Learning; COVID-19; Classification