Proposing an Integrated Approach to Analyzing ESG Data via Machine Learning and Deep Learning Algorithms
Round 1
Reviewer 1 Report
Dear authors,
Multiple experiments are designed in the manuscript to verify the feasibility of various machine learning and deep learning algorithms to collect, organize, and predict Environmental Social Governance (ESG) data. ESG data is used to analyze the advantages and disadvantages of financial investments. It has certain practical significance for using ESG data to promote the sustainable development of human society. However, there are some imperfections in the manuscript. Some of the vital items are as follows:
1. In the abstract and conclusion parts, the contribution summary of this manuscript is repeated, please proofread carefully. It is mentioned in the conclusion that "there is a high positive correlation between ESG and annual returns", which has not been proved in the original text. The inaccurate conclusions need to be revisited.
2. In the manuscript, it is feasible to use multiple algorithms to process the same experimental data under the same experimental conditions. According to the different experimental results, a more suitable algorithm is obtained. However, it does not provide the basis for the selection of algorithm evaluation parameters and evaluation standards. The feasibility conclusion is unreliable.
3. At the end of Section 1.1, it is mentioned that social and governance data sets are difficult to collect and standardize. But no clear solution to this problem is given in the subsequent discussion, suggesting strengthening the connection between the context.
4. Related words are used in the manuscript many times at the connection of sentences. Please check your English grammar carefully, don't use the same related words in two adjacent sentences, and take care to optimize paragraph structure.
5. It is not compared with similar research, thus failing to highlight innovations.
6. There are some errors in the format of the manuscript. For instance, the contents of Table3 are converted into English format, the English abbreviation can be spelled out for the first time. These issues are required to revise carefully.
Sincerely,
The reviewer.
Author Response
Comment 1: “In the abstract and conclusion parts, the contribution summary of this manuscript is repeated, please proofread carefully. It is mentioned in the conclusion that "there is a high positive correlation between ESG and annual returns", which has not been proved in the original text. The inaccurate conclusions need to be revisited”
Response: Thank you for pointing this out. Since we agree with this comment, the abstract was newly written according to the reviewer's opinion, which differs from the conclusion. Furthermore, since we successfully predicted annual return with ESG variables, we concluded that there exists a positive relationship between, as mentioned in the Experiment #3 (5.3)
Comment 2: “In the manuscript, it is feasible to use multiple algorithms to process the same experimental data under the same experimental conditions. According to the different experimental results, a more suitable algorithm is obtained. However, it does not provide the basis for the selection of algorithm evaluation parameters and evaluation standards. The feasibility conclusion is unreliable”
Response: We appreciate you for pointing this out, and admit to this comment. Since we want to show that algorithms were performed under the same condition, we incorporated contents for hyper parameters settings per experiment, and it could be found in section 5.
Comment 3: “At the end of Section 1.1, it is mentioned that social and governance data sets are difficult to collect and standardize. But no clear solution to this problem is given in the subsequent discussion, suggesting strengthening the connection between the context.”
Response: We are grateful for pointing this out, and agree with this comment. Since gathering social and governance datasets is difficult, we collected them through crawling news articles, as mentioned in 3.4.
Comment 4: “Related words are used in the manuscript many times at the connection of sentences. Please check your English grammar carefully, don't use the same related words in two adjacent sentences, and take care to optimize paragraph structure”
Response: Thank you for pointing this out, and agree with this comment. We checked English grammar carefully again, and then took care to optimize paragraph structure
Comment 5: “It is not compared with similar research, thus failing to highlight innovations”
Response: Thank you for pointing this out, and agree with this comment. In the last paragraph of the Related Work section, we emphasized some features that differ from other existing related works.
Comment 6: “There are some errors in the format of the manuscript. For instance, the contents of Table3 are converted into English format, the English abbreviation can be spelled out for the first time. These issues are required to revise carefully. ”
Response: Thank you for pointing this out, and agree with this comment. We coveted all contents to English format and fixed English abbreviations issues.
Reviewer 2 Report
This research discusses a collection of popular AI approaches used in the analysis of ESG data.
The authors have done a great job reviewing the manuscript. The manuscript has little to no grammatical errors, and the writing style is coherent and easy to follow.
The reviewer recommends the following changes:-
- The authors should expand the abbreviations like MAE (mean absolute error), RMSE (root-mean-square error), and LOF (Local outlier factor) when they are mentioned for the first time.
- In the related work section, the authors have highlighted a good deal of relevant research. For each discussed work, the authors should emphasize two items: what is missing from the discussed work, and how the present research will enhance the work.
- In section 3.2, the authors should mention the categories of adversarial attacks countered by the present method.
- In section 3.3, the authors should mention the hyper-parameters of the test train and the robust-scaler methods.
- For Table 3, the authors should cite each article
- The subtitle for Figure 2 is unclear
- In section 5, for each algorithm, the authors should mention the hyper-parameters.
- In figure 7, the authors should highlight the fact that the lower values of RMSE and MAE are better. Adding a phrase like “lower values signify better performance“ will enhance readability.
- In section 7, Conclusion 1 and 4 are unclear.
Author Response
Comment 1: “The authors should expand the abbreviations like MAE (mean absolute error), RMSE (root-mean-square error), and LOF (Local outlier factor) when they are mentioned for the first time.”
Response: Thank you for pointing this out, and agree with this comment. We all fixed those abbreviations-related issues throughout this paper.
Comment 2: “In the related work section, the authors have highlighted a good deal of relevant research. For each discussed work, the authors should emphasize two items: what is missing from the discussed work, and how the present research will enhance the work.”
Response: Thank you for pointing this out, and agree with this comment. In the last paragraph of the Related Work section, we emphasized some features that differ from other existing related works.
Comment 3: “In section 3.2, the authors should mention the categories of adversarial attacks countered by the present method.”
Response: Thank you for pointing this out, and agree with this comment. We incorporated categories of adversarial attacks in the first five sentences in section 3.2.
Comment 4: “In section 3.3, the authors should mention the hyper-parameters of the test train and the robust-scaler methods.”
Response: Thank you for pointing this out, and agree with this comment. We added each hyper-parameters, the proportion of train test split, and robust-scaler methods at the end of the paragraph (section 3.3).
Comment 5: “For Table 3, the authors should cite each article”
Response: Thank you for pointing this out, and agree with this comment. We cited each article and could be found in Table 3.
Comment 6: “The subtitle for Figure 2 is unclear”
Response: Thank you for pointing this out, and agree with this comment. We renamed the subtitle to make it more clear than before.
Comment 7: “TIn section 5, for each algorithm, the authors should mention the hyper-parameters”
Response: Thank you for pointing this out, and agree with this comment. Since we want to show that algorithms were performed under the same condition, we incorporated contents for hyper parameters settings per experiment, and it could be found in section 5.
Comment 8: “In figure 7, the authors should highlight the fact that the lower values of RMSE and MAE are better. Adding a phrase like “lower values signify better performance“ will enhance readability.”
Response: Thank you for pointing this out, and agree with this comment. We added some explanations in the subtitle for figure 7 so that it could be clear that lower RMSE and MAE mean better performance.
Comment 9: “In section 7, Conclusion 1 and 4 are unclear”
Response: Thank you for pointing this out, and agree with this comment. We added sophisticated explanations in Conclusion 1 and 4 to make them more comprehensible compared to before.
Reviewer 3 Report
1. The novelty of this paper is not very clear. Readers would like to see where your contribution to the model/AI/ML is at.
2. How do you determine the impact of the current error in the real-world application. In that sense, is the error still two high?
3. Besides just accuracies, what domain knowledge can you derive for future Sustainability readers on the cultural and social sustainability of human being .
Author Response
Comment 1: “The novelty of this paper is not very clear. Readers would like to see where your contribution to the model/AI/ML is at.”
Response: Thank you for pointing this out, and agree with this comment. We show how AI can contribute to ESG analysis in the conclusion and discussion parts, and also yield how AI (NLP) helps collect social, and governance data
Comment 2: “How do you determine the impact of the current error in the real-world application. In that sense, is the error still two high?”
Response: Thank you for pointing this out, and agree with this comment. Our proposed AI-based algorithms show superior performances, therefore, as mentioned in the conclusion section, we believe it is applicable to the real world
Comment 3: “Besides just accuracies, what domain knowledge can you derive for future Sustainability readers on the cultural and social sustainability of human being”
Response: Thank you for pointing this out, and agree with this comment. Besides accuracy scores, it could be concluded that this research could help future readers approach ESG-related data more easily through this integrated point of view. Furthermore, as mentioned in the Conclusion part, this finding could be helpful to both investors and firms when it comes to approaching ESG issues.
Round 2
Reviewer 1 Report
Improvements have been made in the revised manuscript, and reasonable responses were given. Thus, my recommendation is “Accept in present form”.
Reviewer 3 Report
I don't have any other comments.
