A New Sentiment-Enhanced Word Embedding Method for Sentiment Analysis

Round 1

Reviewer 1 Report (New Reviewer)

The paper proposes a sentiment-aware word embedding approach for sentiment classification. In general, the description of the research is relatively clear, and the research results also have specific contributions to the sentiment analysis field.

Please revise the manuscript to address the following issues:

 

1. Present the recent relevant works, such as: 

 

Zhao, A., & Yu, Y. (2021). Knowledge-enabled BERT for aspect-based sentiment analysis. Knowledge-Based Systems, 227, 107220.

 

Wang, J., Zhang, Y., Yu, L. C., & Zhang, X. (2022). Contextual sentiment embeddings via bi-directional GRU language model. Knowledge-Based Systems, 235, 107663.

 

Wu, J., Ye, C., & Zhou, H. (2021, May). BERT for sentiment classification in software engineering. In 2021 International Conference on Service Science (ICSS) (pp. 115-121). IEEE.

 

2. The following paper introduces a sentiment embedding approach using bidirectional GRU, and is a good article as a reference to compare your result with:

 

Wang, J., Zhang, Y., Yu, L. C., & Zhang, X. (2022). Contextual sentiment embeddings via bi-directional GRU language model. Knowledge-Based Systems, 235, 107663.

 

3. Consider comparing the performance of your proposed method with context-aware text analysis models like BERT and ElMo. 

 

4. Regarding the experimental results section, to compare various models, you need to do statistical tests. Model comparison requires statistical test of significance of the results. 

 

5. Figure 2 needs to be improved. An entry point presenting the input fed to the workflow makes it more precise.

 

6- Please proof-read the manuscript for occasional corrections: Examples:

- ABCDM is used in abstract without being defined. 

- Line 228, 231, (and subsequent header lines) need to be re-formatted to indicate they are sub-headers

- "The the ...."  Line 285

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report (New Reviewer)

 

1. Please confirm it: existing pre-trained word embeddings always perform poorly in sentiment analysis tasks. This might not be true always.

2. The advantage of the proposed method should be highlighted in the abstract

3. Instead of "3. The sentiment-enhanced word embedding method and its applications on 

downstream tasks", please use 'Proposed method'

4. It is difficult to understand Fig. 2, which is input and output? and example text?

5. SOTA looks older, are there any new models for the comparison?

6. Please discuss and cite the recent sentiment analysis tasks. eg

https://www.hindawi.com/journals/cin/2021/2158184/

This author has done several works in sentiment analysis. Please compare and contrast them in the revised version.

 

Author Response

Please see the attachment.

Round 2

Reviewer 1 Report (New Reviewer)

The manuscript is improved significantly based on the previous comments. Thank you. 

Reviewer 2 Report (New Reviewer)

Accepted.

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

This study establishes a link between word embedding and sentiment corpus and performs the classification task using Word2Vec and GloVe. Although the proposed model solves the problem, the proposed model is not compared with the state-of-the-art models.

Some minor but essential suggestions to revise the thesis are as follows:

- Double-check the absolute numbers in the abstract, the performance increase of the proposed model should be positive.

- Why is n-fold cross-validation not used to avoid data biasness?

- Compare the proposed method's results with/without SMOTE to present the influence of SMOTE on the proposed method, especially with the SemEval-2013.

- Compare the results of the proposed model with the state-of-the-art approach(es). 

- Code and dataset should be available to replicate the results.

- The paper presents some wrong English constructions, typo errors, and misuse of articles that should be checked.

Author Response

Please see the attachment. Thanks.

Author Response File: Author Response.docx

Reviewer 2 Report

The paper is not easy to be read and understood by normal reader. For example  the following paragraph are not clear to me:

1- "In TransE [15] method, a knowledge triple (h, l, t) is extracted from knowledge graphs, where h, l and t represent......"

2- Figure 2 the difference between the phrase " Sentiment enhancement embedding" and "Sentiment-enhanced embedding" is not clear

3- Figures 3 &4 are  not clear

and many other things.

Therefore I could not continue reading the paper

Author Response

Please see the attachment. Thanks.

Author Response File: Author Response.docx

Reviewer 3 Report

The main idea of paper is used word embedding to improve sentiment analysis therefore, the title needs to change.

I recommended to publish paper

Author Response

Please see the attachment. Thanks.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

I still see that the paper is hard to be read and benefit readers in its current form. The comments given in my earlier review were just example. The paper needs be simplified to be read independent of the references. Of course you could not explain everything you used but to make some abstractions of the previous work such that readers can benefit from your contribution. The topic you are presenting is very interesting and useful but the presentation is not clear.