Analyzing the Emotions That News Agencies Express towards Candidates during Electoral Campaigns: 2018 Brazilian Presidential Election as a Case of Study
Round 1
Reviewer 1 Report
Review of “Analyzing The Emotions That News Agencies Express Towards Candidates During Electoral Campaigns: 2018 Brazilian Presidential Election as a Case of Study”
This paper presents an analysis of tweets generated by Brazilian news agencies on Twitter during the 2018 Brazilian presidential elections. After a careful evaluation of the manuscript, it is evident that the paper does not meet the standards for publication in its current form. However, the study holds potential, and I recommend that the authors undertake a thorough revision to address the following concerns. Firstly, the paper introduces two methodologies for measuring sentiments and emotions towards the presidential candidates in tweets. However, the description and elaboration of both methods, namely the Stemming and Naïve Bayes algorithms, lack clarity and detail. To ensure better comprehension, it is crucial that the authors provide explicit explanations of the predicted classes, the underlying features, and the process by which the training set was generated for the Naïve Bayes classifiers. An illustrative example would greatly aid in clarifying these aspects. Similar improvements are necessary for the Stemming method. Furthermore, the overall style and organization of the paper can be significantly enhanced. Consider revising the structure to improve the flow of ideas and ensure logical coherence between sections. Additionally, attention should be given to refining the writing style, incorporating appropriate transitions, and using concise and precise language throughout the manuscript. In conclusion, while the current version of the paper cannot be accepted for publication, I acknowledge its potential. I encourage the authors to carefully address the aforementioned concerns, revise the paper accordingly, and resubmit it for further consideration.
Comment 1: The paper presents findings that indicate the prevalence of negative emotions expressed by most news agencies towards the winning candidate, Jair Bolsonaro, during the Brazilian elections. These results suggest that relying on emotions as a basis for election prediction, at least in the context of Brazil, is ineffective. In the concluding section, the authors propose that media attention may serve as a predictive factor. However, the paper lacks sufficient evidence to substantiate this claim. Hence, I recommend modifying the tone in the abstract where it states, "Finally, we discuss that the candidate that captured the highest and most negative attention won the elections, highlighting the importance of having social media presence, regardless of generating positive or negative emotions." At best, the authors demonstrate a bias in the news reporting, which may not accurately reflect the actual support among the voting population. It remains uncertain whether similar results would be obtained with a left-wing candidate.
Comment 2: The paper exhibits numerous grammatical errors and typos that require comprehensive rewriting. For instance, Line 21 contains an incomplete sentence: "An emotional state, for example, anger or happiness." Line 77 has extra spacing that needs to be removed. Line 87 should correct "owrks" to the appropriate term. A typo is present in line 91. Line 95 includes an incomplete sentence: "In a way that whatever sentence about a politician causes the most extreme emotions in people [27]." Line 120 also contains an incomplete sentence: "Where representative and balanced samples concerning specific factors, such as genre, newspaper articles, literary fiction, spoken speech, blogs and journals, and legal documents." Furthermore, the sentence "…there is a consistency of similar over time, it can be detected.." is grammatically incorrect. These are just a few examples of the errors present in the paper that necessitate careful revision.
Comment 3. The methodology section requires extensive revision to enhance clarity and comprehensibility. To improve the discussion, the authors should begin by introducing the main metrics S and R. Subsequently, they should provide a comprehensive explanation of how each term in S is extracted using the Stemming and Naïve Bayes algorithm. It is essential for the authors to fully define terms such as "prevailing sentiment" and "modulated sentiment," along with explaining how each term was generated for every tweet. For instance, in the case of the "prevailing sentiment" determined by the Naïve Bayes classifier, it is crucial to clarify the features used by the classifier, the process of determining the training set, the classes associated with this classifier, and the specific type of Naïve Bayes classifier employed. Adding small illustrative examples at this stage would greatly aid in clarifying the input, training, and output components of the methodology. Similar improvements should also be made for the Stemming method.
Comment 4. In relation to Section 3.1, I recommend keeping Figure 1 as it provides valuable information. However, it would be beneficial to relocate the detailed description of the various steps (lines 182-247) to an Appendix. This adjustment will help to avoid distraction from the main focus of the section, which is the definition of the S and R metrics. By moving the step-by-step description to the Appendix, readers can easily refer to it for a more comprehensive understanding of the methodology without detracting from the primary points highlighted in the section.
Comment 5. The content presented in lines 207-222 appears to be repetitive, duplicating the description already provided on page 4.
Comment 6. Equation 2 appears to be incorrect as it measures the "Relation" from one agency towards a specific candidate, but it sums over agencies and candidates. There seems to be an error in the formulation.
Comment 7. The authors should provide further elaboration on the finding that R is predominantly negative. Additionally, it would be beneficial to understand the potential reasons behind the observed escalation in negativity over time. Is it possible that this trend could be attributed to a systemic error in the analysis? Alternatively, is there a deeper underlying factor that could explain the consistent negativity? Expanding upon these points will help shed light on the significance and interpretation of your findings.
Comment 8. In line 29, you mention the influence of Twitter on recent protests. It would be informative to acknowledge that the impact of news on political mobilizations and civil unrest has been demonstrated in a broader context. Reference 1 provides evidence for the effect of news on civil unrest events across various countries and spans over a century of data. Including this reference in your list would be beneficial. Additionally, reference 2 explores the role of online networks in protest mobilization and could also be a valuable addition to your references.
Comment 9. In line 243, you mention that a detailed comparison was conducted between the methods using the working dataset, resulting in the identification of the "Stemming in Portuguese" method as the most accurate for analyzing emotions, achieving 90.70% accuracy. However, the evaluation process is not clearly described in this section or throughout the paper. It is important to clarify whether training and testing sets were used, and if cross-validation techniques were applied. Additionally, details regarding the predictive performance measures employed in the evaluation are necessary. Further elaboration is required to provide a comprehensive understanding of the evaluation methodology.
Comment 10. Figure 3 is causing confusion. The y-axis represents the "Sum of Support," which is calculated using weighted values derived from various factors such as "prevailing sentiment," "polarity," and "Compound." However, the panels in the figure are described at the level of the individual NB classifier and Stemming algorithm. This misalignment between the y-axis and the panel descriptions adds to the confusion. Additionally, as mentioned in Comment 9 above, there is a need for clarification on how the evaluation between the NB classifier and the Stemming algorithm was conducted in this context. Further explanation is necessary to address these concerns.
Comment 11. In line 353, you state, "Next, we compared the sentiment detected by Naïve Bayes...". However, the subsequent paragraph does not appear to be directly related to this statement.
Comment 12. It appears that the two paragraphs in lines 369-382 contain redundant information. To improve clarity and eliminate repetition, I recommend combining these paragraphs and removing the duplicated content.
Comment 13. My final comment pertains to the overall interpretation of the results and an important aspect missing from this paper. When voters make decisions during elections, they are influenced by external factors, including news media, as well as by the influences exerted by their peers. In the context of the Brazilian elections discussed in this paper, it appears that the external influences from news media may have played a less significant role compared to the influences exerted by peers on social media platforms. Therefore, it is plausible to consider that the victory of Bolsonaro reflects the impact of social networks, particularly Twitter, in shaping people's voting decisions. This interpretation aligns with the theory and extensive empirical research presented in reference 3. In that study, the authors analyzed a century of US presidential election data, demonstrating that social contagion effects, mediated by social networks, have become increasingly instrumental in shaping collective political behavior on a large scale. It is crucial for the authors to discuss this possibility in light of these recent findings and their implications for the present study.
References:
1. Braha, D. (2012). Global civil unrest: contagion, self-organization, and prediction. PloS one, 7(10), e48596.
2. González-Bailón S, Borge-Holthoefer J, Rivero A, Moreno Y (2011) The dynamics of protest recruitment through an online network. Sci Rep 1, 197.
3. Braha, D., & De Aguiar, M. A. (2017). Voting contagion: Modeling and analysis of a century of US presidential elections. PloS one, 12(5), e0177970.
See main comments.
Author Response
First of all, we would like to thank the reviewer for the valuable comments and feedback. We have revised the manuscript taking into consideration the comments provided by the reviewers. We have marked in blue the changes. Reply to the Review (black: reviewer comments, red: author answer, blue: modified texts).
See attached file
Author Response File: 
Author Response.pdf
Reviewer 2 Report
It is unclear to me why the authors conclude that the news agencies are expressing the emotion or sentiment towards the candidates. Wouldn't the analysis simply take into account any words used in the article? So what if they are quoting the candidates, supporters, or opposing candidates? That doesn't seem to be about the news agencies' emotion or sentiment towards the candidate, no? Maybe there is a more sophisticated argument to be made about the sources being cited, etc., but it seems hard to argue that these words reflect the opinions/attitudes/emotions of the news agencies themselves.
It seems a much stronger contribution could be made by an analysis that distinguished between emotion/sentiment expressed in the articles and by people quoted in the articles.
Other notes:
Starting an academic paper with a definition from a dictionary seems very weak. The authors could look to psychology and sociology or other fields to define emotion.
The example phrase used in the methodology “The candidate answered hardly the reporter's question." is not proper English, and therefore doesn't really help illustrate the point. Did they mean "hardly answered" as in didn't give a very complete answer? Or did they mean something else? It would be helpful to use an example in proper English to help the reader understand the sentiment and emotional evaluation.
Some strange spacing in lines 77-81, 87-91, 95-96, 100-108, 137, 204, .
Line 109 says "This paper is organized as follows." and does not include any information about how the paper is organized.
The word choice throughout reflects a kind of translation from Portuguese words to "false friends" or words that appear to be equivalent but are used differently in English, and could use some revision by someone with a strong command of English usage.
Author Response
First of all, we would like to thank the reviewer for the valuable comments and feedback. We have revised the manuscript taking into consideration the comments provided by the reviewers. We have marked in blue the changes. Reply to the Review (black: reviewer comments, red: author answer, blue: modified texts).
See attached file
Author Response File: Author Response.pdf
Reviewer 3 Report
This is an investigation of interest on the use of twitter by news agencies in Brazil in the 2018 elections.
The data set is quite old. 2018, previous pandemic, in the digital age 6 years is too long to consider.
In the introduction it is necessary to explain the use of social networks around the world and in Brazil with recent statistics.
in the introduction, there is only one reference to a similar study referring to the Indonesian elections. It is totally necessary to include other references worldwide with the main results extracted from them.
the methodology for analyzing the relationship between news agencies and candidates on social media is excellent but the conclusions are very poor, it is necessary to improve this part of the article, considering the large amount of research on the matter, and putting all of them in relation to this analysis.
Do not include the macmillan dictionary at the beginning of the article, it is not necessary and less scientific
When using acronyms, please use full words first (for example: Natural Language Processing) and then the acronyms (NLP)
its mostly good
Author Response
First of all, we would like to thank the reviewer for the valuable comments and feedback. We have revised the manuscript taking into consideration the comments provided by the reviewers. We have marked in blue the changes. Reply to the Review (black: reviewer comments, red: author answer, blue: modified texts).
See attached file
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors have successfully addressed all of the referee’s comments.
Author Response
Thank you very much for your feedback
Reviewer 3 Report
Thanks for improving the text. Regarding the explanation about the fact that data is 6 years old, please explain an justify it in the text in some way, as the authors try to in the document of improvements, in the final manuscript.
Author Response
Thanks for improving the text. Regarding the explanation about the fact that data is 6 years old, please explain an justify it in the text in some way, as the authors try to in the document of improvements, in the final manuscript.
Thanks for your comments. We have added the following text in the first paragraph of the Dataset section:
The generated dataset, is and will be relevant for researchers exploring political contexts through social networks, even if the data is old, because this election presented a high polarization and most of the discussion took place through Twitter. In fact, it is actually very relevant, mainly because candidate Bolsonaro still stands as one of the main topics in the Brazilian social networks, where news agencies are constantly reporting about him.
