Heterogeneous Graph Neural Network for Short Text Classification

Round 1

Reviewer 1 Report

The present study proposes a novel method for short text classifications and neural graphics. The authors point out the limitations of the current methods and the possible benefits. The subject is interesting and it is a good fit for the Applied Sciences Journal.

Abstract: Abstract is objective and well-constructed. I only suggest the authors to add a final phrase to closure, with the suitability of the model and/or a future perspective.

Introduction: Is clear and well written, pointing out the limitations of present models and the importance of the study. In addition, contains examples of word characterizations. However, the format (citing the contributions after the objectives) is closer to a thesis or a project application than a journal paper. I suggest changing the format of the end of the text. Moreover, the authors should define what a short text is (how many words does a short text can contain to be classified as?).

The related work section is good, embracing a mini review. However, this section is also suitable for the discussion. It might be interesting to put this review together with the results or in the discussion section, which could provide a direct comparison or hypothetical explanations for the superior obtained results.

Methodology is clear and illustrated. Nevertheless, the resolution of the images are low and the graphics could be more aesthetic appealing. The network graphics can be displayed in circular or hierarchical layouts. Provide the software and versions used.  

Experiment (results) section is well-written and objective, with tables and graphics.

Discussion: Too short. It does look like an overview of the results and not a discussion. The authors should enrich this section with some comparisons with other models/studies or putting some of the related work section here. The comparison among models in the experiment section is comprehensive and correct – however, an in-depth qualitative comparison and adding some hypotheses regarding the better results that were obtained must be in the discussion section as well. No limitations regarding the present study were pointed out. Another option is embedding the discussion within the results section.

Conclusion: This section could be more focused. Language seems too informal, employing the use of “you” and the way of putting the pros and cons.

Minor findings:

POS tags definition written

Double punctuation line 61.

GCN – line 63, definition written (it is the abstract thou)

Line 87, 94 – punctuation

Line 100 “he” is inadequate

Line 121 lack of a “(“. Also, the sentence is confusing, repeating “TF-IDF” consecutively.

Line 131 – suggestion, change gray color, too similar to the black ones. Dashed is an option

Line 134 – written is () and the abbreviation is not. Replace it.

Line 166 – space after “where”

Line 210 – space between ( and a

Line 229 – phrase is confusing

Line 290 - . ,

Line 324 - Pros and cons in classification. Use :

Language: discussion and conclusion the verbal form should be in the past form.

Author Response

Dear Reviewer:

Thank you very much for your suggestions for revision of this paper. Your comments are very valuable and very meaningful to this paper. Based on your comments, we have revised this paper, and the responses to your questions are as follows:

Abstract: This paper has increased the scope of applicability of the model in the abstract.

Introduction: References have been revised and expanded in this article, and a definition of short text has been added.

Related work: This paper has expanded the discussion with reference to content from related work.

Method: This article and the images have been modified according to your comments, the gray is changed to green, and the software used is Power Point 2013.

Discussion: This paper has added discussions related to related work, and has also outlined the research progress related to short texts, and pointed out the advantages and disadvantages of the model proposed in this paper, as well as future research directions and room for improvement. 

Conclusion: This paper has revised the conclusions used in this paper to focus on the contributions of this paper and future research directions, and the language has been revised.

Minor findings:

This paper has added the definition of POS tag, the language form in the discussion and conclusion, and revised some of the formatting issues you raised.

Thank you again for carefully reviewing this paper and for making these meaningful suggestions.

Reviewer 2 Report

This paper provides an interesting approach for text classification based on  short text information. The following issues should be addressed before it can  be accepted for publication: The quality of English should be improved. I suggest that the text goes through a professional analysis.  - The authors should mention other approaches for text classification, including those based on complex networks. See and mention e.g.: doi: 10.1016/j.physa.2017.12.054 and 10.1016/j.ipm.2018.12.008 In general those approaches are complementary to traditional text classification approaches. - I would like to see a discussion on whether the proposed approach can be used to classify papers abstracts. This is an interesting application. - Conclusions of the paper should be improved. The authors should provide perspectives for future work.  - I would like to know if the differences in performance are statistically significant.

Author Response

Dear Reviewer:

Thank you very much for your suggestions for revision of this paper. Your comments are very valuable and very meaningful to this paper. Based on your comments, we have revised this paper, and the responses to your questions are as follows:

-This article has made a lot of revisions to English expressions, making them more formalized and standardized.
-For the other text classification methods you proposed, this paper and the analysis and comparison of them are added, and the two papers you gave are studied carefully.
-This paper discusses the research used for classification of abstracts in papers and is added in the Introduction and Discussion sections. Abstracts belong to the scope of short texts, and the model proposed in this paper can be used to analyze them. However, due to the large number of professional vocabulary and specific phrases contained in the abstract of the paper, to improve the classification accuracy may require a more targeted pre-training model or a more complex network, which may become the focus of our next research. 
In this paper, the conclusion part has been modified, and the future work is analyzed and prospected from different directions.
-The comparative experiments done in this paper are carried out under the same data set and determined parameters. The results shown are the average values ​​after the model runs 10 times, and the standard deviation is provided, which has certain statistical significance.
Thank you again for your careful review of this paper and for these meaningful suggestions.

Reviewer 3 Report

The contribution of this paper is sufficient by proposing a new approach called TWPGCN. The organization of this paper is well written and the proposed approach is compared to the existing models. It shows better performance than the existing designs. However, there are some sections that could be further improved. Thus, I suggest a minor revision and the authors are suggested to address the comments as follows:

1. Please discuss the motivation of this study in detail and show how this research could fill the current research gap.

2. Please include more references from the past two years (2021 to 2022).

3. Please discuss the theoretical and practical contributions of this study.

4. Please summarize the strength and weaknesses of the proposed method compared to the existing designs.

Author Response

Dear Reviewer:

Thank you very much for your suggestions for revision of this paper. Your comments are very valuable and very meaningful to this paper. Based on your comments, we have revised this paper, and the responses to your questions are as follows:

1. This paper has added the motivation for this research in the Discussion section and explained how this paper is different, why it works for short texts, and the advantages of the model proposed in the paper.
2. More references from the past two years  have been added to this paper.
3. This paper has added the theoretical and practical contributions of this research in the Discussion section.
4. This paper has summarized the advantages and disadvantages of the proposed method compared to existing designs in the Discussion section.
Thank you again for your careful review of this paper and for these meaningful suggestions.