Sentiment Analysis of Text Reviews Using Lexicon-Enhanced Bert Embedding (LeBERT) Model with Convolutional Neural Network

Round 1

Reviewer 1 Report

In this paper, the authors proposed a new sentiment classification model, called LeBERT, which integrates Sentiment Lexicon, N-grams, BERT, and CNN in one framework.  They performed experiments on three datasets and demonstrated that LeBERT is effective and outperforms all baseline models in the comparison.

The paper is well-structured and well written. The authors gave an informative introduction and reviewed related work with references wherever needed. The model description, experiment setup, and result discussion are all straightforward and easy to follow. There are just a few minor issues listed below –

1. Abstract: the three datasets are not all from Yelp – make sure it is consistent with the dataset description in Section 4
2. Section 2, Paragraph 2, Line 1: replace “uses” with “use a”
3. Section 2, Paragraph 3, Line 5: insert a “,” after “For instance”
4. Page 4, Paragraph 2, Line 25: insert a “,” after “technique”
5. Page 4, Paragraph 3, Line 1: replace “works” with “work”
6. Page 5, Section 3, Paragraph 2, Line 8, replace the first “where” with “with”
7. Page 8, Paragraph 2, Line 1: remove the “;” after “Let”
8. Page 8, Paragraph 3, Line 1: add a subscript “i” to the text vector “v”
9. Page 9, Section 4 title: the “.” is misplaced
10. Issues with formula numbers: (2) and (3) are missing; align them up vertically
11. Page 10, Section 5, Paragraph 1: rephrase the sentence “In the training … tensor flow.”
12. Page 14, Paragraph 2, Line 1: add “our” to “proposed model”

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

The whole experiment is sufficient, but there are problems that need to be revised  

1. The author puts forward three contributions. But in fact, points 1-3 can be integrated into one.  

2. The related work is too redundant and should be divided into two categories: traditional algorithm and deep learning algorithm. Or according to other classifications, rather than a whole chapter. At present, this is not very friendly to readers.    

3. Most of reference need to be updated. In addition, two articles about NLP are recommended to the author. 10.1155/2022/8660828 10.3390/ijgi10100653  

4. The specific parameter settings should be placed in the Experiments section instead of the Results and Discussion section.  

5. Figure 5 should use professional software to draw  

6. The conclusion should be discussed in combination with the result

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The author has carefully revised the article and it is suitable for publication in this journal. Two additional suggestions

1. Article about “the lexicon based techniques” can be recommended to authors. 10.1109/ACCESS.2019.2920091

2. The author may misunderstand. Figure 5 should be drawn using special drawing software instead of using Microsoft Excel.

Author Response

Please see the attachment

Author Response File: Author Response.docx