the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 14 Nov 2025
Tracing Emotional Evolution along Named Entity Topic Chains: A Mechanistic Study of Chinese Social Media in the 2025 Myanmar Earthquake
Abstract. This study examines how emotional responses to transboundary disasters are structured and propagated within digital discourse, using the 2025 Myanmar earthquake as a case. Drawing on a dataset of 139,473 Chinese Weibo posts collected from March 28 to April 25, we develop an emotion – entity coupling framework that integrates large language model-based emotion annotation with named entity recognition (NER) to construct a semantic-affective network. Rather than treating sentiment as a standalone attribute, this approach models emotion as a dynamic and relational process that flows through named entities, which serve as semantic anchors and emotional conduits. The analysis reveals distinct patterns in both temporal emotion dynamics and structural emotion transmission. While emotions such as fear and surprise dominated the discourse, positive sentiments – particularly those associated with humanitarian actors – formed localized zones of empathic resonance. The coupled emotion–entity network exposed asymmetric affective pathways, with certain entities acting as hubs of amplification, bridge nodes, or buffers in the transmission of emotional meaning. Subgraph analysis further highlighted how institutional memory, geographical proximity, and media narratives shaped the stability and flow of public sentiment. By reconceptualizing emotion as structurally embedded and semantically routed, this study offers both theoretical and methodological innovations in disaster risk communication. The proposed framework advances understanding of how empathy and public engagement are generated, distributed, and sustained in the digital age – particularly in response to disasters that cross national borders.
- Preprint
(4830 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2025-4507', Anonymous Referee #1, 28 Nov 2025
-
RC2: 'Comment on egusphere-2025-4507', Anonymous Referee #2, 20 Dec 2025
Tracing Emotional Evolution along Named Entity Topic 2 Chains: A Mechanistic Study of Chinese Social Media in the 3 2025 Myanmar Earthquake
Reviewer Notes 20.12.2025
Abstract
In the abstract, enhance clarity and conciseness, minimize metaphorical expressions (“flows,” “conduits,” “zones of empathic resonance”), and provide greater methodological detail.
Introduction
The introduction is excessively long and contains repetitive critiques of traditional sentiment analysis. Multiple paragraphs reiterate points such as the overly aggregate nature of sentiment and the lack of semantic grounding, without providing additional nuance. The writing sometimes emphasizes theoretical elegance at the expense of analytical clarity, which may challenge reader engagement. Furthermore, the term “mechanism” is not clearly defined early in the text. While the introduction is generally persuasive, it would benefit from greater conciseness, a more precise problem definition, and improved conceptual focus.
Literature review
The literature review demonstrates competence and currency; however, it requires a more focused scope, clearer conceptual boundaries, and more selective critical analysis.
Methodology
The overall workflow, encompassing data collection, annotation, model fine-tuning, and network construction, is logical and well structured. However, the absence of quantitative validation for both emotion classification and named entity recognition (NER), such as F1 score, accuracy, or inter-annotator agreement, is a significant limitation. Assigning a single “dominant emotion” to each post oversimplifies the emotional complexity present in disaster discourse. The inference of “emotion transitions” based solely on entity co-occurrence lacks formal justification and may conflate association with propagation. Furthermore, key modeling decisions, including edge weighting, threshold selection, and network pruning, are not sufficiently justified. In summary, while the methodology is ambitious and well-designed at a conceptual level, it lacks adequate validation, transparency, and robustness checks to fully substantiate the paper’s mechanistic claims.
Results
The results are presented in a clear sequence, moving from descriptive statistics to emotional dynamics and ultimately to semantic–affective networks. Nevertheless, several significant limitations are evident. The findings are predominantly descriptive and visual, with minimal quantitative testing or statistical validation. Assertions regarding “emotion flow,” “amplification,” and “buffering” are based on co-occurrence patterns rather than being formally substantiated. The extensive use of complex network visualizations increases the risk of interpretive overreach and reader subjectivity. Furthermore, the lack of baselines or null models hinders the assessment of whether the observed patterns are distinctive or could occur by chance. Although positive emotion clusters are emphasized, their relative magnitude and robustness are not systematically quantified. In summary, while the results are comprehensive, coherent, and visually engaging, they remain largely exploratory. The strength of interpretive claims is not consistently supported by analytical rigor.
Discussion
The discussion appears to overinterpret descriptive findings, particularly when inferring mechanisms such as amplification and buffering without formal causal evidence. While claims regarding nationalism, trust construction, and geopolitical effect are plausible, they remain untested empirically. Alternative explanations, including media agenda-setting, platform effects, censorship, and posting norms, receive insufficient consideration. The discussion reiterates conceptual contributions at length, resulting in redundancy. Although the discussion is thoughtful and theoretically ambitious, it extends beyond what the results can robustly support. Furthermore, only two references are cited in the discussion, which limits its scientific support.
Conclusion
The conclusion section reiterates claims from the Discussion but does not synthesize insights at a broader conceptual level. The use of mechanistic and causal language is overstated, given the analysis's primarily descriptive nature. Although limitations are acknowledged, their implications for interpretation are not thoroughly examined. Suggestions for future research are concise but lack specificity. Overall, while the conclusion is coherent and well written, it adds little beyond summarization.
Reviewer Suggestions
The authors’ efforts in preparing this manuscript are appreciated.
The manuscript addresses a significant and timely topic in disaster management. However, several substantive concerns remain.
The study relies heavily on emotion classification but does not report essential validation metrics (e.g., accuracy, Scores, inter-annotator agreement). Without quantitative evidence of model performance, it is difficult to assess the reliability of the core analytical outputs on which the conclusions depend.
The manuscript frequently describes the analysis as “mechanistic” and interprets emotion as “flowing” or “propagating” through entity networks. However, the empirical foundation for these claims is primarily based on co-occurrence patterns rather than on formally specified mechanisms, causal inference, or diffusion modeling. This discrepancy results in a gap between the strength of the claims and the evidentiary support provided.
Although the results are rich and visually compelling, many interpretations, particularly those concerning emotional amplification, buffering, trust construction, and digital nationalism, extend beyond what can be robustly inferred from the analyses presented.
Key concepts such as emotion propagation, semantic routing, and affective pathways are used extensively, yet remain insufficiently operationalized. Consequently, the manuscript at times blurs the distinction between analytical metaphor and empirical demonstration.
Collectively, these issues necessitate substantial reconceptualization, additional validation, and methodological strengthening that exceed the scope of a standard revision process.
Thank you.
Regards.
-
RC3: 'Comment on egusphere-2025-4507', Mirela-Adriana Anghelache, 24 Dec 2025
This article presents a good application of a model of AI in order to estimate the emotional public response in case of a natural disaster, in this case the 2025 Myanmar earthquake. The applied method and the conclusions are relevant not only for the scientific community but for the decision-makers, as well, in order to understand and respond properly to the public emotional response in case of a disaster.
I strongly recommend the article to be published and I recommend some minor technical correction for improvement:
1. In the Abstract, at Row 15, for the "large language model" to be put the abbreviation LLM in brackets.
2. the paragraph between Rows 76-92 from Introduction to be moved to the Chapter 3, Data and Methodology, it and could begin with the "The study..."
3. the phrase from Rows 121-123 repeats, it can be deleted.
4. Fig. 10 - the colors for Good and Anger don't discriminate clearly between them and on my monitor I can see bold pink lines for which I don't see the explanation at the Legend.
Citation: https://doi.org/10.5194/egusphere-2025-4507-RC3 -
RC4: 'Comment on egusphere-2025-4507', Anonymous Referee #4, 03 Jan 2026
General comments:
The manuscript analyses emotional responses on social media following the 2025 Myanmar earthquake, framed as a transboundary disaster in this work. The authors combine large language model for emotion classification and named entity recognition to then construct an emotion–entity coupled network, aiming to model the flow and transformation of emotions beyond typical aggregate sentiment analysis. The topic is interesting and relevant to the NHESS readership, particularly in the context of understanding societal perception of disasters based on social media data and NLP based methods. Such analyses can potentially provide valuable insights for disaster communication, management, and risk awareness. The dataset is large, and the analytical workflow involving LLMs and network modelling demonstrates substantial technical effort. However, the application and interpretation of these methods remain largely qualitative. Much of the analysis relies on visualisations and descriptive network representations, with no quantitative validation of the inferred emotional dynamics. As a result, the link to natural hazards science and disaster risk reduction remains mostly conceptual and is not sufficiently articulated to be actionable or informative for hazard practice.The manuscript would benefit from clearer positioning within natural hazards research, stronger methodological validation, and substantial streamlining. I therefore recommend that the authors rework this manuscript which unfortunately goes beyond major revisions.
Specific comments:
- Quantitative and statistical validation: The manuscript presents hypotheses and interpretations that are primarily supported through visual inspection of networks and figures. The authors should provide some statistical or empirical basis to evaluate these hypotheses, some quantitative metrics will demonstrate value beyond descriptive visualisations.
- Transboundary perspective and novelty: Given that the majority of posts originate from regions not directly affected by the earthquake, the authors should clarify what additional insights are revealed compared to existing studies of domestic or locally affected disasters. It would be useful to explicitly contrast the findings with patterns reported in past work.
- Introduction structure and event description: The Introduction should be rewritten to be more concise and focused. A clearer description of the earthquake event itself (damage, impacts, timeline, response actions) is needed, particularly where these aspects are later referenced in the analysis. Fig 1 has some details but that is not sufficient.
- Redundancy between Introduction and Literature Review: The literature review is thorough and well explained. However, there is considerable redundancy between the Introduction and Literature Review. I suggest combining or restructuring these sections, as several arguments introduced early on are difficult to follow without the context provided later in the literature review.
- Discussion and conclusion sections: Similar redundancy appears between the Discussion and Conclusion. These sections could be substantially shortened and made clearer. In particular, the authors should provide more concrete examples of practical insights gained from the analysis, beyond methodological improvements. That is how the public sentiment will be utilized by govt, aid agencies ? Especially in such a transboundary event?
- Use of highly abstract or philosophical language: The manuscript frequently employs theoretical or philosophical terminology that may be difficult to interpret, particularly for readers with an engineering or applied hazards background like me. The authors should aim to use clearer, more direct language and more explicitly relate their analysis to post disaster activities.
- Overall length and clarity: The manuscript would benefit significantly from tightening. Reducing repetition and focusing on key contributions would improve readability and impact.
Technical corrections and minor comments
- Title clarity: The title is difficult to interpret without reading the abstract and introduction. A clearer, more descriptive title would improve accessibility.
- Terminology consistency: Please use consistent terminology throughout (e.g. “named entity” vs “key entity”).
- Simplify complex phrasings such as “latent cognitive infrastructures of digital disaster” should be rewritten in simpler English.
- “Fine-grained” emotion annotation: But the analysis appears to rely on single-label emotion classification. The authors should clarify why this is considered fine-grained.
- Annotation examples: Providing concrete examples of emotion and entity annotations (e.g. a small table) would greatly improve transparency.
- Figures and presentation:
Figure 5 should be split into labelled subpanels, with clearer annotations and a more detailed caption.
Figure 6 needs axes, a labelled y-axis and clear indication of units or normalisation.
Figures 8–10 are very difficult to interpret in static form. While they may be effective in an interactive dashboard, the manuscript should extract and present more interpretable summaries or quantitative insights from these networks. - Keyword filtering: The keywords used for data collection can be more explicitly documented.
- Spatial context: Maps showing the geographical distribution of users would help contextualise the spatial patterns discussed in the text.
Citation: https://doi.org/10.5194/egusphere-2025-4507-RC4
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 392 | 170 | 28 | 590 | 21 | 20 |
- HTML: 392
- PDF: 170
- XML: 28
- Total: 590
- BibTeX: 21
- EndNote: 20
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Thank you for allowing me to review this manuscript. The topic of the manuscript is interesting, and the contribution is clear. Yet, the clarity of the first half of the paper could be improved, as well as that of the results. My recommendations and comments would be as follows:
1. In line 45, you claim that " social media operates not only as an information infrastructure but also as an affective interface". Please clarify how this is possible and what this implies.
2. In line 50, you referred to "evolving topic structures" without clarifying what this means. Please do this to ensure clarity.
3. The paragraph on the 2025 Myanmar earthquake (lines 53-65) appears under-supported, as several of the claims are not referenced. Please address this.
4. In line 91, you do not clarify what "semantic chains" are. This may detract from the clarity of your work.
5. In line 99, "transnational solidarity, symbolic alignment, and cognitive resonance" should be individually briefly discussed to ensure the clarity of your contribution from the beginning of the paper.
6. The contribution of your work should also be highlighted further, following line 108, by explaining the relevance and motivations behind this growing research interest in this area.
7. The dynamics mentioned in line 153 could also be discussed more explicitly.
8. The key reasons for the importance of addressing the gap identified in line 163 should also be pointed out more clearly.
9. The graphs included in the results section could be described briefly for increased clarity and accessibility.
I hope this helps. All the very best.