AugFake-BERT: Handling Imbalance through Augmentation of Fake News Using BERT to Enhance the Performance of Fake News Classification

Round 1

Reviewer 1 Report

In the introduction, correct the statement claiming that fake news is a type of disinformation; It’s not. Fake news is an abstract term and falls merely into two categories of information disorders: misinformation or disinformation.

 

In the literature review section, line 87, change “fake news” to disinformation—disinformation is the type of information disorder that is purposedly created to deceive contrary to its counterpart misinformation (which can happen by accident due to outdated articles or even the lack of public knowledge).

 

Correct sentence break and transition (lines 252, 253).

 

Provide metrics on the performances of classifiers when performing based on balanced vis-à-vis imbalanced data—since the paper is highly anchored on the importance of data balancing.

 

In the evaluation baseline section, the metrics on which language is used as reference is missing; Are these values based on English or the Bengali language?

 

In the datasets provided, especially given the background of the pervasiveness of disinformation, it would be really useful to provide a case where the class of fake news/deceptive content proportion is reversed: i.e., under-sampling of the other class.

 

Why is there no limitation(s) provided on this article? Please attach this important section to your work.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The article addresses a very important research problem and offers well-structured study data. This academic paper is well structured, it offers a literature review in which the authors provide a critical view of various fake news detection methods.

The main part of the article consists of a detailed presentation of the research data, which offers insight into an interesting study that addresses important ML problems. The fact that the authors of the study address the issues of Bengali language analysis and offer a solution to them, overcoming the influence of the ML models of the English language dominant in academic research, is a positive thing. The research method is clearly represented, an appropriate list of literature is attached to the article.

However, the article would need the following improvements:

- the introduction is a bit chaotic so far, it should be improved by defining the purpose and conceptual framework of the study instead of simply describing the structure of the article (from 71 lines).

- in the introduction, it would be necessary to briefly describe fake news data in relation with actual situation in India, because readers in other countries may not be familiar with the obstacles of the specific country. It is not enough to announce that fake news spreads quickly in the social media environment, the authors of the study should define the reasons how their research will solve specific problems of the public communication landscape;

- the part of the conclusions is very laconic, it lacks a discussion of the research results and their significance, and a review of the limitations of the study.

English should be improved throughout the article.

I would recommend accepting the article for publication after making minor changes.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

In general, I think it is worthy of publishing. Some points should be included within the manuscript in order to improve the publication.

• In the introduction section, the current state of the research field should be reviewed carefully and key publications cited and analyzed. A brief description of corresponding studies about text augmentation would be useful.
• There is a rich body of literature on all of the topics covered in this paper and many of these papers must be reported in the Introduction section.
• The description of the augmentation (section 4.1) is relatively weak in the present form and should be strengthened with more details and justifications.
• In the conclusions section the authors summarize the main points of their study. The authors should explain the contribution of their results in comparison to the results of other researchers. More conclusions should be added. Practical applications of this study should be mentioned.
• The authors should refer to the recent papers, such as the following:

-         Chi-Chung Tao, Yue-Lang Jonathan Cheung, "Social Media Mining on Taipei's Mass Rapid Transit Station Services based on Visual-Semantic Deep Learning," WSEAS Transactions on Computers, vol. 20, pp. 110-117, 2022.

 

-      Huanzhuo Ye, Yuan Li, "Fuzzy Cloud Evaluation of Service Quality Based on DP-FastText," WSEAS Transactions on Computers, vol. 20, pp. 149-167, 2021.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 4 Report

Reviewer's comments to the author:

1.          To use BERT for language classification has been done in the last 10 years and documented a lot, your paper does not provide new insights. The methodologies used in the paper are common methods, and novelty should be emphasized.

2.          This paper provides a quite simple comparison between established classification methodologies. The comparison of the six deep learning models and the seven machine-learning is not well organized in its current form at least. While a thorough investigation between many data, or an improvement and combination of these methodologies may provide some sort of scientific contribution, the present study does not provide any strong novelty.

3.          Abstract should summarize the main findings, and provide implications or directions for future work.

4.          Since the machine learning methods in Fig. 8 and 9 are independent, a bar graph should be used instead of a line graph.

5.          Please check the entire manuscript for possible grammatical and syntax errors. I will mention just a few: Such as " Initially, Natural Language Processing (NLP) issues were handled using traditional Machine Learning (ML) methods such as Logistic Regression and Support Vector Machine (SVM) with hand-crafted features. [10,11]." There is two period at the end of the sentence. (On page 2, lines 41-43). "Masked Language Modeling (MLM): given a sentence, the method masks 15% of the words in the input and then runs the the whole masked text through the model, which must estimate the masked words." There is two "the" in the sentence. (On page 6, lines 218-220).

6.          Some parts are written comprehensively with many citations. Yet, the use of the English language is marginal in quality, while tenses and sentences are often poorly constructed.

7.          Overall, it is unclear what the study is presenting or contributing academically or practically. Descriptions of more contributions to this research are needed in the article.

I rate the article as a draft that still requires a lot of work to get published. It should be rejected unless the authors perform the experiments again as suggested and rewrite the text of the article.

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

The paper can be published in its present form.

Reviewer 4 Report

The authors have successfully addressed all of the reviewer's comments. This manuscript can be accepted without any comments. I recommend it for publication.