Quantification and Analysis of Group Sentiment in Electromagnetic Radiation Public Opinion Events

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper proposes a system for analyzing public opinion regarding electromagnetic radiation (EMR) on social media, with a focus on the NIMBY effect. The proposed system should allow the identification of sources of negative emotions in social networks and offer practical solutions for authorities to efficiently monitor and manage public opinion crises.

I recommend the authors consider the following aspects:

-the current introduction is too long and includes much general information about the internet and social media, which can be condensed or moved to the background/“Related Work” section. It is important that the introduction clearly explains, from the first paragraphs, the specific problem the paper addresses (for example, the difficulty in analyzing public emotions related to EMR and the impact of NIMBY).

-the introduction should also include references to previous studies that have addressed public opinion or sentiment analysis. In its current form, the lack of bibliography in the introduction is a major issue. Furthermore, at the end of the introduction, it is useful to indicate how the paper is organized into sections.

-in the introduction section, the author states that the paper has three innovative contributions, which is only partially true. It would be better to write that this paper applies and integrates the following methods in a current context.

- figure 1 lacks a clear description of its components, does not explain the relationships between modules, and does not show how the figure reflects the proposed innovations.

-table 1 is based on a sample that is too small (700 posts) and does not clearly explain how precision and recall were calculated, which affects the credibility of the results. There are no details about the data source or the validation methodology.

-table 2 is based on only ten subjectively analyzed posts, without a clear procedure for sentiment labeling, which makes the results weakly supported and unreproducible.

All the mathematical formulas are insufficiently explained, and a clarification of all the variables used is necessary.

At lines 444 and 445, it is not explained what the provided link contains. Moreover, the .xlsx files have columns written in Chinese, making it difficult to understand what is found in these files.

Chapter 4 has an incorrect title (“Reaserch” instead of “Research”) and does not present a proper research plan, but rather a summary of implementation. The statistical analysis is superficial, based only on average values, without significant tests or confidence measures.

At lines 466–468, the claim that the average emotional score of 0.019 indicates a generally positive trend is questionable, as the value is extremely close to zero, suggesting rather a neutral or balanced distribution than a clearly positive one. The authors do not provide any statistical analysis—such as standard deviation, confidence interval, etc.—to justify the conclusion. The interpretation of a "positive trend" is unfounded and potentially misleading.

-at line 478, Figure 2 does not explain how the emotional scores were calculated, making it impossible to verify or reproduce the results. Moreover, it contains text in Chinese without translation, which reduces clarity and accessibility in a paper written in English. In Table 2, the text is written in English but also includes usernames or terms in Chinese.

Figure 3, which is supposed to illustrate the dynamics of collective sentiment over time, provides only a general representation, without detailed explanations of the axes, calculation method, or units used. Thus, trend interpretation becomes ambiguous, and the reader lacks sufficient information to understand how the data was aggregated.

Figure 5 is not explained and is very unclear.

In the conclusion section, it is stated that the proposed system is effective and applicable to public opinion monitoring, but no relevant data is mentioned to support this claim, such as the average emotional scores or the distribution of analyzed sentiments. References to the problematic aspects identified in the data analysis are entirely missing, such as the small sample size, lack of manual validation, or issues related to deleted posts. The conclusions are limited to idealistic formulations about "social governance" and "effective response," without clearly showing how the research results support these statements.

The paper proposes a useful framework for analyzing public sentiment regarding EMR in the online environment; however, to achieve its goal, substantial reorganization is necessary. Emphasis must be placed on clarifying the methodology, expanding and validating the dataset, and rigorously interpreting the results. The paper contains numerous writing issues in English, including grammatical mistakes, vague formulations, and inconsistent terminology. Careful rewriting and thorough language editing are essential to improve the quality and credibility of the research.

Comments on the Quality of English Language

The English language must be checked, and where applicable, Chinese text in figures and tables should be replaced with English.

Author Response

Reviewer Comment 1:

“the current introduction is too long and includes much general information about the internet and social media, which can be condensed or moved to the background/“Related Work” section. It is important that the introduction clearly explains, from the first paragraphs, the specific problem the paper addresses (for example, the difficulty in analyzing public emotions related to EMR and the impact of NIMBY).”

Response:
Thank you for this valuable feedback. We agree that the initial introduction contained overly general background information. In the revised manuscript, we have significantly condensed the introductory discussion about social media platforms and general internet usage. We also restructured the first paragraphs to focus more explicitly on the specific problem the paper addresses—namely, the difficulty in analyzing public emotions related to EMR and the implications of the NIMBY effect. These adjustments help clarify the motivation and research gap from the outset.

Reviewer Comment 2:

“the introduction should also include references to previous studies that have addressed public opinion or sentiment analysis. In its current form, the lack of bibliography in the introduction is a major issue. Furthermore, at the end of the introduction, it is useful to indicate how the paper is organized into sections.”

Response:
We appreciate the reviewer highlighting this oversight. In the revised version, we have added citations to key previous works on public opinion and sentiment analysis in the introduction, such as those by Kelman (2006), Maitner et al. (2006), and Li et al. (2020). This provides better context and grounding for our study. We have also added a brief paragraph at the end of the introduction outlining the structure of the paper.

Reviewer Comment 3:

“in the introduction section, the author states that the paper has three innovative contributions, which is only partially true. It would be better to write that this paper applies and integrates the following methods in a current context.”

Response:
Thank you for this suggestion. We have revised the relevant paragraph to clarify that the paper applies and integrates several methods within a current context, rather than claiming entirely novel methodological innovations. The revised phrasing now reads: “This paper applies and integrates the following methods within the context of EMR-related public sentiment analysis…”

Reviewer Comment 4:

“figure 1 lacks a clear description of its components, does not explain the relationships between modules, and does not show how the figure reflects the proposed innovations.”

Response:
We acknowledge this shortcoming. In the revised manuscript, we have updated in-text description of Figure 1 to clearly explain the roles of each module, their interrelationships, and how they reflect the innovations proposed in the paper.

Reviewer Comment 5:

“table 1 is based on a sample that is too small (700 posts) and does not clearly explain how precision and recall were calculated, which affects the credibility of the results. There are no details about the data source or the validation methodology.”

Response:

We appreciate the concern regarding the sample size and transparency. We have added further explanation of the data source, including the Weibo API access period (Oct 2021 – Sep 2022), and provided a more detailed description of the validation process for precision and recall calculations. The dataset contains 700 posts, accompanied by over 20,000 instances of associated data such as comments, reposts, and likes, demonstrating that the information volume is sufficient to support our analysis. Furthermore, to enhance the reliability of the data, we conducted manual annotation by inviting five annotators to label the dataset, thereby ensuring a more credible and accurate ground truth.

Reviewer Comment 6:

“table 2 is based on only ten subjectively analyzed posts, without a clear procedure for sentiment labeling, which makes the results weakly supported and unreproducible.”

Response:
We agree that additional clarification was needed. In the revised manuscript, we now explain the procedure used to subjectively analyze the top-ten posts, double-checking by two annotators to reduce subjectivity. While we acknowledge the small sample size, we stress that this was intended as a qualitative check to illustrate the effect of the semantic filtering method. We have added a statement on the limitations and need for larger-scale validation.

Reviewer Comment 7:

“All the mathematical formulas are insufficiently explained, and a clarification of all the variables used is necessary.”

Response:
Thank you for pointing this out. We have revised the relevant sections to provide clear definitions of all variables used in the mathematical formulas, particularly The group emotion value of a cluster is explained in 3.2.2. Each formula now includes accompanying textual descriptions and variable definitions for clarity.

Reviewer Comment 8:

“At lines 444 and 445, it is not explained what the provided link contains. Moreover, the .xlsx files have columns written in Chinese, making it difficult to understand what is found in these files.”

Response:
We have updated the manuscript to explain the contents of the provided link, which includes the source code, dataset, and a demo video. Furthermore, we have translated the column headers in the uploaded .xlsx files into English and included an English data dictionary for easier understanding and reproducibility.

Reviewer Comment 9:

“Chapter 4 has an incorrect title (“Reaserch” instead of “Research”) and does not present a proper research plan, but rather a summary of implementation. The statistical analysis is superficial, based only on average values, without significant tests or confidence measures.”

Response:
Thank you for identifying the typo. We have corrected “Reaserch” to “Research” in the section title. We also revised the section to more explicitly present the research design, including the hypothesis, variables, analysis strategy, and rationale for selected methods. We improved statistical commentary where relevant.

 

Reviewer Comment 10:

“At lines 466–468, the claim that the average emotional score of 0.019 indicates a generally positive trend is questionable, as the value is extremely close to zero, suggesting rather a neutral or balanced distribution than a clearly positive one. The authors do not provide any statistical analysis—such as standard deviation, confidence interval, etc.—to justify the conclusion. The interpretation of a "positive trend" is unfounded and potentially misleading.”

Response:
Thank you for your insightful comment. We agree that the average sentiment score of 0.019 is numerically close to zero and does not, on its own, justify the conclusion of a “positive trend.” Our original expression was imprecise and may have misled readers. In response, we have revised the sentence to reflect a more neutral interpretation of the data and have supplemented the analysis with additional statistical information, including the standard deviation and 95% confidence interval, to more accurately characterize the emotional distribution. Specifically, we now report that the sentiment score is near-neutral, with high variance.

Reviewer Comment 11:

“at line 478, Figure 2 does not explain how the emotional scores were calculated, making it impossible to verify or reproduce the results. Moreover, it contains text in Chinese without translation, which reduces clarity and accessibility in a paper written in English. In Table 2, the text is written in English but also includes usernames or terms in Chinese.”

Response:

We have updated Figure 2 to include translated labels and a clearer explanation of how emotional scores were computed using the dictionary-based sentiment analysis algorithm in 3.2.2. This explanation is now also reflected in the caption and corresponding text.

Reviewer Comment 12:

“Figure 3, which is supposed to illustrate the dynamics of collective sentiment over time, provides only a general representation, without detailed explanations of the axes, calculation method, or units used. Thus, trend interpretation becomes ambiguous, and the reader lacks sufficient information to understand how the data was aggregated.”

Response:
We appreciate the reviewer’s observation regarding the lack of detailed explanation in Figure 3. In the revised manuscript, we have supplemented the figure caption and main text with additional information, including the calculation method for daily emotion values, clarification of the axes. Specifically, the horizontal axis represents the timeline from April 28 to September 10, 2022, and the vertical axis denotes the average emotional score per day, calculated based on sentiment analysis of related Weibo posts and comments. The daily score is the arithmetic mean of all individual sentiment values on that day.  with scores ranging from -1 (strongly negative) to +1 (strongly positive). This enhancement improves the interpretability and reproducibility of the results.

Reviewer Comment 13:

“Figure 5 is not explained and is very unclear.”

Response:
Thank you for pointing this out. We acknowledge that the original explanation of Figure 5 lacked sufficient clarity. In the revised manuscript, we have added a detailed description of how the "influence" of accounts was calculated. Specifically, we clarify that the influence metric is based on three components: cohesiveness, authority, and dissemination power. We also explain the meaning and role of each component in the context of emotional dissemination. Additionally, we have revised the figure caption to better explain the data presented, enhancing both the readability and reproducibility of the results.

Reviewer Comment 14:

“In the conclusion section, it is stated that the proposed system is effective and applicable to public opinion monitoring, but no relevant data is mentioned to support this claim, such as the average emotional scores or the distribution of analyzed sentiments. References to the problematic aspects identified in the data analysis are entirely missing, such as the small sample size, lack of manual validation, or issues related to deleted posts. The conclusions are limited to idealistic formulations about "social governance" and "effective response," without clearly showing how the research results support these statements.”

Response:

Thank you for your valuable feedback. We acknowledge the limitations you identified in the conclusion section and have revised the manuscript to address them more clearly.

First, to enhance the transparency and reproducibility of our study, we have made the dataset publicly available. We hope that this open data approach will encourage further exploration and validation by the research community, supporting future studies on public opinion and sentiment analysis.

Second, regarding data reliability, we have conducted manual validation of a portion of the dataset. Specifically, a subset of posts and sentiment labels were manually reviewed, and we performed secondary verification for part of the data to improve accuracy. While full-scale manual labeling was not feasible due to resource constraints, we are committed to improving the data quality in future iterations of our system.

Third, we recognize that deleted or missing posts may affect analysis completeness. In future work, we plan to integrate large language models (LLMs) to assist with content filtering, anomaly detection, and semantic enrichment of incomplete data. This will help enhance both the coverage and robustness of the sentiment analysis.

Finally, we have revised the conclusion to better reflect the actual findings of our study, highlighting not only the practical implications for social governance but also the methodological challenges and areas for future research. These updates aim to provide a more balanced and evidence-based summary of the study’s contributions and limitations.

Reviewer Comment 15:

“The paper proposes a useful framework for analyzing public sentiment regarding EMR in the online environment; however, to achieve its goal, substantial reorganization is necessary. Emphasis must be placed on clarifying the methodology, expanding and validating the dataset, and rigorously interpreting the results. The paper contains numerous writing issues in English, including grammatical mistakes, vague formulations, and inconsistent terminology. Careful rewriting and thorough language editing are essential to improve the quality and credibility of the research.”

Response:
Thank you for your constructive and insightful feedback. We appreciate your recognition of the potential value of our proposed framework and acknowledge the areas needing improvement that you have outlined.

In response, we have made substantial revisions throughout the manuscript to address your concerns:

1. Clarification of Methodology: We have reorganized and expanded the methodology section to provide a clearer explanation of data collection, sentiment analysis procedures, emotion score calculations, and account influence metrics.
2. Dataset Expansion and Validation: We have conducted additional manual validation of a portion of the dataset to ensure data quality and reliability. Furthermore, we have made the dataset publicly available to support open research and allow others to reproduce and extend our work. We also plan to integrate large language models in future iterations to assist in data cleaning and annotation.
3. Interpretation of Results: We revised the results and discussion sections to provide more rigorous and balanced interpretations of the findings.
4. Language and Terminology Improvements: We have conducted a thorough language revision to correct grammatical errors, improve clarity, and ensure consistency in terminology.

Reviewer 2 Report

Comments and Suggestions for Authors

Before you publish the paper, please improve any editorial mistakes, such as double spaces, titles of section on previous pages etc. Github link should also start with https, etc.

You have not mentioned any generic methods of finding relevant words, e.g. TF-IDF that sometimes helps to find the relevance of the expression, in your case, it would be EMI.

In your future work please also consider the sample of the results of sentiment analysis to be checked with the "annotation exercise", i.e. manual checking whether the sentiment is correct or not. Sometimes the ML algorithm may result in high accuracy but in fact not because it is so perfect, but because of problems with the keyword.

Because the data is not the most recent, you can consider repeating this exercise if all code to process and analyse the data already exist.

Please translate Chinese labels in Figures and Tables - you can add English translation in brackets - it is difficult to read for most of the readers who are not familiar with your language. I know it is explained in the text but it would be better to read have it directly in the figure / table.

It was also hard to find the information on processing the data. For example, first step would be to detect the language as you may have English and Chinese posts in Weibo.

Comments on the Quality of English Language

Please consider a proof-reading and editorial changes as mentioned in the review.

Author Response

Comment 1:
“Please improve any editorial mistakes, such as double spaces, titles of section on previous pages etc. Github link should also start with https.”

Response:
Thank you for pointing this out. We have carefully revised the manuscript to correct all editorial issues, including double spacing, improper section title placements, and formatting inconsistencies. The GitHub repository link has also been corrected to start with "https" for proper formatting and security compliance.

Comment 2:
“You have not mentioned any generic methods of finding relevant words, e.g. TF-IDF that sometimes helps to find the relevance of the expression, in your case, it would be EMI.”

Response:

Thank you for the suggestion. We acknowledge that methods such as TF-IDF can be useful in identifying keywords relevant to specific topics like EMI. In our study, we initially filtered relevant posts using topic-specific keywords manually selected based on prior domain knowledge (e.g., “electromagnetic radiation,” “EMR,” “radiation harm”). We recognize that incorporating automated keyword extraction techniques such as TF-IDF or topic modeling (e.g., LDA) could enhance objectivity and scalability. We plan to integrate these approaches in future work to improve the robustness of topic relevance detection.

Comment 3:
“In your future work please also consider the sample of the results of sentiment analysis to be checked with the 'annotation exercise', i.e. manual checking whether the sentiment is correct or not.”

Response:

Thank you for the valuable suggestion. In this study, we have already performed a manual annotation exercise to verify the accuracy of the sentiment analysis results. A portion of the dataset was manually labeled by trained annotators to ensure the validity of the sentiment classification, and secondary verification was also conducted to further improve reliability. In future work, we plan to expand the scale of manual annotation and consider combining it with model-assisted review (e.g., using large language models) to enhance the overall consistency and quality of sentiment evaluation.

Comment 4:
“Because the data is not the most recent, you can consider repeating this exercise if all code to process and analyse the data already exist.”

Response:
Thank you for the suggestion. We acknowledge that the current dataset has a limited temporal scope. In our follow-up research, we are extending the study to cover a longer time span and include more recent and diverse public opinion events related to EMR and other emerging topics. This ongoing work aims to validate and expand the applicability of our analysis framework using up-to-date data and additional cases.

Comment 5:
“Please translate Chinese labels in Figures and Tables - you can add English translation in brackets.”

Response:
We fully agree with this comment. All Chinese text in figures and tables has now been translated or accompanied by English translations in brackets to improve readability and accessibility for a broader audience.

Comment 6:
“It was also hard to find the information on processing the data. For example, first step would be to detect the language as you may have English and Chinese posts in Weibo.”

Response:
Thank you for pointing this out. We agree that language detection is an important part of preprocessing. In the revised version of the manuscript, we have further clarified the data processing steps in the first section. Specifically, we note that the collected Weibo content was primarily in Chinese, and we translated key text content into English for analysis and presentation. This ensures consistency in processing and improves accessibility for an international audience.