An Ensemble-Learning-Based Technique for Bimodal Sentiment Analysis
Round 1
Reviewer 1 Report
The article discusses the importance of bimodal sentiment analysis, an emerging area in natural language processing, which has gained significant attention in various fields. However, the limited studies on bimodal conversational sentiment analysis have posed challenges due to the complexity of how humans express sentiment cues across various modalities. The authors attempt to address this gap by proposing a neural network-based ensemble learning technique, achieving state-of-the-art accuracy on the MELD dataset.
The paper is well written and flows well. It was a pleasure to read, the authors make a considerable effort to convey the information in a well readable way. However, the paper show few drawbacks that in my opinion could be address in order to improve its quality. My main concern regards the contribution: in fact, the authors simply propose a study that exploits several methods that are already known in the literature. Although the authors propose an ensemble learning approach, the combination of the considered methods is not enough to be considered as a solid contribution. In light of this consideration, I suggest the authors to address some aspects that could strengthen the proposed work. These aspects are detailed in the following:
- In the Introduction, the authors should make an effort and convince the reader of why this work is needed.
- Also in the Introduction, the contribution should be stated in a more convincing and clear way.
- Although the paper flows well, it misses the “so what”. What are the implications of this study? What should a reader grasp from this study?
- More information about low-level features should be provided. As an example, advantages and disadvantages of each of them could be reported as well as whether one should be selected instead of another.
- Also some example of deep features could be reported.
- I appreciated the Section 2.3, however it is not easy to follow the overview of the related approaches. I suggest the authors to use a table to summarize the approaches and reports for each of them pros and cons.
- The distribution of the emotions in the considered dataset could be reported. It helps identifying whether the dataset is balanced or neither.
- A discussion on why the basic RoBERTa is the most performant in 5.1 could be reported.
- Figures reporting heatmaps should be way smaller.
- Sentiment analysis techniques are important in several fields, e.g., social network analysis. The authors could consider citing recent works that exploits such techniques, e.g., [https://doi.org/10.1016/j.datak.2022.101979]
Author Response
Dear Reviewer,
Thank you very much for your constructive feedback on our paper. We appreciate your efforts in thoroughly reviewing our work and providing us with valuable suggestions for improvement. We have carefully considered all your points and made the necessary revisions to address them. We believe that these revisions have significantly strengthened the paper and improved its overall quality.
Firstly, we have revised the introduction to better emphasize the need for our study, and to state our contributions more clearly and convincingly. We have also provided more information about low-level features and deep features and included a table in Section 2.3 summarizing the related approaches mentioned.
In addition, we have reported the distribution of emotions in the dataset and added a discussion on why the basic RoBERTa is the most performant. We have also addressed the issue of the size of the heatmaps in Figure 5.1.
Lastly, we have included a citation to a recent work that exploits sentiment analysis techniques in social network analysis where relevant, as suggested by you.
We hope that you will find the revised paper more convincing and of higher quality. Thank you once again for your valuable feedback.
Sincerely,
Shariq
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The topic of this research (Bimodal Sentiment Analysis) is very interesting. The authors of this papers reported this study in a verbose manner, mainly focusing into details of existing body of knowledge in sentiment analysis methods. However, there are few major points that the authors needs to address:
1) Introduction does not clearly portray the research impact (i.e., the area that this research will benefit). Without clearly highlighting the research impact, the readers wouldn’t comprehend the rationale or justification of this study. Moreover, the outcome or contribution of this study in not portrayed in a quantifiable or measurable manner. To mitigate this drawback, the authors need to clearly state the contribution along with numeric measurable progress. For example, statements like “Ensemble techniques have demonstrated superiority over techniques in various classification task” is extremely vague and non-academic. I urge the authors to portray these critical information in a scientifically measurable manner (stating the improvements in F1 scores, Precisions, Recalls in comparison with existing studies).
2) In section 2, simply stating the background and related works, paragraph after paragraph is not appealing to the readership. Use tables to portray advantage, disadvantages of existing methods in a scientific manner that generates interest among the academic community.
3) Methodology/Experimentation sections, could benefit from introducing a flow chart / conceptual model to portray the information in a visual manner.
4) There are also minor issues like Figure 1 spreading across multiple pages. Same issue with Figure 2. Figure 1 should be in a single page and same for Figure 2. You can use quadratic format (2 columns and 2 rows) depict a, b, c, d in a same page. This means that the font sizes within the figures need to be bigger (to keep them readable).
5) I am happy with results and conclusion sections.
Author Response
Dear Reviewer,
Thank you for your valuable feedback on our paper. We have carefully considered your suggestions and made the necessary revisions to address the points you raised. We believe that these revisions have significantly improved the quality of our paper.
Firstly, we have revised the introduction to portray the research impact more clearly, and to state the contribution scientifically.
Secondly, we have used tables to better illustrate the advantages and disadvantages of existing methods in Section 2. We have also included a flowchart in the methodology and experimentation sections to visually portray the information in a more organized and understandable manner.
Thirdly, we have restructured Figure 1, Figure 2, and Figure 3 to be displayed in a single page in a quadratic format. We have also increased the font sizes within the figures to keep them readable.
Lastly, we are glad to hear that you are happy with our results and conclusion sections.
We hope that you find our revised paper more scientifically measurable, visually appealing, and of higher quality. Thank you again for your valuable feedback.
Sincerely,
Shariq
Reviewer 3 Report
1. There are certain typos and grammatical mistakes. For instance: State-Of-The-Art >> State-of-the-Art. The authors should proofread the paper for possible errors.
2. Efforts are needed to make the abstract coherent and some parts of it to be re-written for ease of understanding. For instance, this sentence, “To address this gap, a comparison of the performance of multiple data modality models has been conducted on the MELD dataset, a widely-used dataset for benchmarking sentiment analysis within the research community.”
3. The manuscript overall needs better organization. For instance, the section and models within 2.2. Summary of Models Used in Bimodal Sentiment Analysis should belong to Material and Methods instead 2. Background and Related Work. Refer : DOI: 10.1109/ACCESS.2020.3043221 (https://ieeexplore.ieee.org/document/9286431 )
4. Figures (1-3) occupy 5 pages which are just Confusion matrices. It is a bad utilization of space and makes the manuscript unreasonably longer. It will be good to shrink the images for better readability and optimal use of spaces.
5. Table 5 needs a better caption (concise and representative of the work).
6. Subsection 4.5. Ensemble Learning is not understandable.
7. The comparison of the proposed neural network model with ‘max’ and other voting is incomprehensible in the absence of necessary details.
8. What is the basis of Ensemble selection when selecting an ensemble model? Refer : DOI: 10.1109/ACCESS.2020.3043221 (https://ieeexplore.ieee.org/document/9286431 )
9. I suggest the authors strictly follow the article DOI: 10.1109/ACCESS.2020.3043221 for the overall organization of papers and inferencing the sections feature extraction, ensemble approach, and DL architecture.
10. Conclusion: The work presented in this manuscript is interesting. But from the current organization of the manuscript, it is difficult to follow the work, and some sections are disjoint.
11. Therefore, it is suggested that authors sufficiently motivate the necessary sections and refer to DOI: 10.1109/ACCESS.2020.3043221 (https://ieeexplore.ieee.org/document/9286431 ) for holistic readability of the manuscript.
Author Response
Dear Reviewer,
Thank you for taking the time to review our manuscript and providing us with constructive feedback. We appreciate your thorough evaluation of our work and have taken your suggestions into consideration to improve the quality of our paper.
We have carefully proofread the paper and corrected all typos and grammatical mistakes, including the correction of "State-Of-The-Art" to "State-of-the-Art." We have also made efforts to improve the coherence and clarity of the abstract, particularly the sentence you mentioned.
Regarding the organization of the paper, we have made changes wherever we could. However, we think that our current structure is already quite like the DOI you shared and provides a clear and concise presentation of our work. The similar view is shared by other reviewers.
We have also addressed your comment on Figures 1-3 and have adjusted optimize space and readability. Table 5 now has a better caption that accurately represents our work. We have revised Subsection 4.5 on Ensemble Learning to make it more understandable and added necessary details for the comparison of the proposed neural network model with 'max' and other voting. Regarding Ensemble selection, we have made changes throughout the manuscript to establish an understanding on this point.
Finally, we have included a paragraph in the conclusion section that outlines the potential applications of our proposed approach. We hope that these changes address your concerns and improve the quality and readability of our paper. Once again, we thank you for your valuable feedback.
Best regards,
Shariq
Reviewer 4 Report
Please address the following issues:
1. Please extend this point:- "Detecting bimodal sentiment can be an important aspect of many applications, mostly in Contact Centres (CCs), which are the first point of contact with organizations for most customers". Here you must clarify what is exactly the problem and why it is important to be solved. Does it affect anyone/business/society? Please do not ignore any what, why and does questions.
2. The type of accuracy should be clear in Tables 4,5 and others.
3. Extend the implications of your approach to academia/literature and business/professionals. This you should address in the discussion section.
4. Cite the reference here and extend this point by providing the key benefits or advantages of using ensembling - "Ensemble techniques have 61 demonstrated superiority over techniques in various classification tasks. This is due to 62 their flexibility in training and updating classifiers".
5. in the conclusion please write a paragraph about the potential applications of your proposed approach.
Thank you.
Author Response
Dear Reviewer,
Thank you for taking the time to review our paper and for providing us with valuable feedback. We have carefully considered your suggestions and have made the necessary revisions to address the issues you have raised.
- We have extended the discussion in the Introduction to clarify the problem of detecting bimodal sentiment and why it is important to be solved, including its impact on Contact Centres (CCs).
- We have made it clear in Tables 4 and 5 that the accuracy reported is weighted accuracy. We have also revised other tables and wherever necessary to clarify the type of accuracy reported.
- We have extended the Results and Discussion to include the implications of our approach to academia/literature and business/professionals, highlighting its potential contributions to both areas.
- We have cited the reference and provided a summary of the key benefits or advantages of using ensemble learning.
- We have included a new paragraph in the Conclusion section that discusses the potential applications of our proposed approach, including its use in various industries and contexts, such as CCs and social media.
We hope these revisions meet your expectations and that our paper is now improved in terms of clarity and scientific rigor. Once again, we appreciate your feedback and the opportunity to improve our work.
Sincerely,
Shariq
Round 2
Reviewer 1 Report
The authors successfully addressed my concerns, even though they did not consider citing the reference I suggested [doi.org/10.1016/j.datak.2022.101979], that is an approach that could be extended with the one proposed by this paper.
Author Response
Dear Reviewer,
Thank you for your email.
We have carefully reviewed all the cited references in our manuscript and have found all of them to be relevant to our research. Each reference cited in our manuscript has contributed to our understanding of the research topic and has provided valuable insights into our study. There may be some references that are not be directly related to the topic but they still have contributed in writing our research.
Therefore, we confirm that all the cited references are relevant to our research, and we have not found any reference that we consider irrelevant.
We hope this clarifies any concerns regarding the cited references in our manuscript. Thank you for your time and consideration.
Best regards,
Shariq
Reviewer 2 Report
The authors have successfully addressed all my previous comments in this revised manuscript. It looks much better now. I do not have any further comments to add.
Author Response
Thank you!
Reviewer 3 Report
My comments are duly addressed, and I recommend the manuscript for publication.
Author Response
Thank you!
Reviewer 4 Report
The authors have tried to update the manuscript according to the suggestions. Thank you.
Author Response
Thank you!
