Buzz Tweet Classification Based on Text and Image Features of Tweets Using Multi-Task Learning

Round 1

Reviewer 1 Report

This paper is to investigate social media trends and propose a buzz tweet classification method to explore the factors that cause the buzz phenomenon on Twitter. 
An updated and complete literature review should be conducted to present the state-of-the-art of the research. Besides, additional experiments are needed to validate the effectiveness of the proposed method. It is recommended to expand the data set.
Pay attention to the presentation of the paper. Please check whether the following are appropriate.
In 5 Conclusions, we considered tweets with images as buzz tweets, and tweets with more than 1,000 likes as non-buzz tweets.
True positive rete, False positve rete.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The manuscript is centered on a very interesting and timely topic, which is also quite relevant to the themes of Applied Sciences. Organization of the paper is good and the proposed method is quite novel. The length of the manuscript is about right.

The manuscript, however, does not link well with important literature on NLP research, e.g., see works on convolutional and recurrent neural networks for multi-label text categorization. Also, check latest trends in generative models for category text generation and literature on learning word dependencies in text by means of deep recurrent belief networks. Finally, review the usage of capsule networks for challenging NLP applications and recent hybrid AI approaches that inject semantic information into deep nets, e.g., hybrid networks for aspect-based sentiment analysis.

The manuscript presents some bad English constructions, grammar mistakes, and misuse of articles: a professional language editing service (e.g., the ones offered by IEEE, Elsevier, or Springer) is strongly recommended in order to sufficiently improve the paper's presentation quality for meeting the high standards of Applied Sciences.

Finally, double-check both definition and usage of acronyms: every acronym, e.g., SNS, should be defined only once (at the first occurrence) and always used afterwards (except for abstract and section titles). Also, it is not recommendable to generate acronyms for multiword expressions that are shorter than 3 words, e.g., RT (unless they are universally recognized, e.g., AI).

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The flow of thought in the article is easy to follow, but it shows a lack of knowledge of scientific writing at several points. The methods used should be presented in much more detail and with a scientific approach. The description of related work should also be improved. 

Major comments:

• The structure of the article is not usual. It is not common to divide the Introduction into subsections. I suggest reorganizing these subsections into a new section and rethink the structure of the article.
• For articles [1-6], I suggest also introducing the developed/used machine learning algorithms. What kind of AI methods were developed/used in these articles for performing the classification of the tweets?
• Please introduce the article [9] in more detail.
• In general: the introduction of related works needs to be significantly extended and rewritten. The applied methods need to be also presented. Furthermore, there is a lot of approach to model information diffusion on networks, but the authors only mention one from them. There is a need for a more detailed presentation of related work also in this area.
• Section 2.2: Please explain why you used 768-dimensional word embeddings.
• Section 2.2: There is also no mention of the preparation of texts. Does this mean that the raw texts were given as input to the BERT model, and no preparation was done?
• Section 2.2: Please specify from which layer of the BERT model the word embedding was extracted.
• Section 2.3: The extraction of the image features needs to be presented in more detail.
• Section 2.3: Why have been used for this task 7 networks? How were these results used or combined?
• Section 2.4: It would be important to specify the exact structure of the networks used: activation functions, number of neurons, … (for each layer).
• Section 2.4: The presentation of the training phase is missing. Please introduce how was the network trained (training dataset, epoch number, etc.), and please introduce the fine-tuning of the network also.
• Section 2.4: The presentation of the text set is controversial. Please introduce the collected text set in more detail.
• The reorganisation of Section 2.4 is needed. In this chapter, the presentation of the data set, the presentation of the learning phase, the presentation of the model and the presentation of the analysis results (Figures 2 and 3) are mixed. Please reorganize this Section.
• As I see, the statement related to Figure 10. is not true. If there is a difference, it should be measured by clustering coefficients, for example.
•  

 

Minor comments:

• Line 15-16: error in the sentence: text repetition within the sentence
• Line 21-22: error in the sentence: text repetition within the sentence
• Line 31: abbreviation SNS is not explained
• Line 35, 37: ” is missing
• Line 43: “what does mean “tweets/tweets”?
• Examples between lines 35 and 37: Please describe these examples in more detail, because they are not well-known for all readers.
• Line 62: abbreviation RT is not explained
• Line 81: “account account”
• Line 82-84: “text “they applied a deep learning feature extraction approach to analyze unstructured tweet text” is repeated
• Tables 3, 6, 7: Please clarify that in all combinations BERT model was used.
• Furthermore, all tables need to be aligned centre.
• Figures 2, 3 and Figures 5, 6 have the same captions.
• The sentence in lines 254-255 “Figure 5 shows the misclassification of buzz tweets, and Figure 6 shows the misclassification of non-buzz tweets.” is not true.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The paper investigates social media trends and proposes a buzz tweet classification method to explore the factors causing the buzz phenomenon on Twitter. Experiments validates the effectiveness of the proposed method.
The authors have revised it according to the suggestions made by the reviewers.

Author Response

Thank you very much for re-reviewing. We have reviewed the manuscript for typographical errors and partially edited the manuscript again to produce the final version.

Reviewer 2 Report

The authors have addressed most of the concerns raised by the reviewers and their revisions have substantially improved the manuscript. However, there are still some minor issues to be addressed, namely:
1) presentation is better but there is still some room for improvement
2) important relevant literature is still missing

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors have taken my suggestions into account and have corrected and supplemented the article accordingly.

Author Response

Thank you very much for re-reviewing. We have reviewed the manuscript for typographical errors and partially edited the manuscript again to produce the final version.