the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 17 Apr 2023
Brief communication: Small-scale geohazards cause significant and highly variable impacts on emotions
Abstract. The impact of geohazards on the mental health of the local populations, is well recognized but understudied. We used natural language processing (NLP) of Twitter posts to analyse the sentiments expressed in relation to a pre-eruptive seismic unrest, and a subsequent volcanic eruption in Iceland 2021. We show that despite the small size and negligible material damage, these geohazards were associated with a measurable change in expressed emotions in the local populations. The seismic unrest was associated with predominantly negative sentiments, but the eruption with predominantly positive. We demonstrate a cost-effective tool for gauging public discourse that could be used in risk management.
- Preprint
(916 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on nhess-2023-6', Anonymous Referee #1, 03 Apr 2024
Review of nhess-2023-6 “Brief communication: Small-scale geohazards cause significant and highly variable impacts on emotions” by Evgenia Ilyinskaya, Vésteinn Snæbjarnarson, Hanne Krage Carlsen, Björn Oddsson.
This paper is a well written focussed brief communications that needs only minor revision for publication.
My suggestions (in no particular order of importance) are as follows:
- Please add more specific summary information in it. For example, how many tweets, examples of sentiments found, what was the measurable change in sentiments. It is currently very hard level and does not work well as a summary.
-
- This communications was submitted in 2023 but only one of the 12 references is from 2022 and one from 2021. This feels a bit dated, as this field has been moving on over the past few years. Introduced a few more recent examples of the literature surrounding social media and geohazards.
- You might want to briefly mention the significance of your findings in the broader context of geohazard risk management and public health.
- These are well written, succinct, and refer the reader to the appendix for further information.
- Results and Discussion.
- I suggest you ‘introduce’ Figures 1 and 2 in terms of what is being shown, rather than just refer to results from them. Provide even 1-2 sentences pointing out to the reader what they are seeing. In Figure 1 is presented…. One can observe…
- See comments under Figure 1 and 2 below for other comments (that sometimes relate to explanations in the text).
- I’d like a bit more depth about uncertainties and limitations (beyond the brief statement at end of methods). In particular, can you briefly discuss limitations of NLP in interpreting the nuances of human emotions, especially in languages with complex idiomatic expressions like Icelandic.
- Including specific tweet examples as evidence of these sentiment changes could help the reader’s understanding. I realize that this is a short communications but I was left without a good feeling for what was categorised positive vs. negative and whether you had categories within them (and neutral) or ‘more’ or ‘less’ positive/negative/neutral.
- Further elaboration on how your insights could be integrated into existing geohazard management frameworks might enhance the paper's applicability.
- Figure 1.
- Overall a really nice figure.
- Suggest that blue and red for Fig. 1C might be difficult for colour blind people unless you add shading.
- I’m not getting the ‘yellow’ refferred to in the figure caption—do you mean orange? This might be an issue of the PDF.
- In figure caption reiterate how many total tweets there are.
- In Figure 1a, earthquakes from what minimum to maximum magnitude?
- Refer reader back to Appendix A for the detailed list of keywords.
- Figure 2.
- This figure does not work as well.
- See comments on Figure 1 re earthquake magnitude and labelling.
- I’d be curious to see this figure log-log, given the clustering of values in the lower decades of n. Consider doing a four-part figure rather than two part showing the y-axis linear and log and x-axis always log.
- I suggest you have two different variables for x- and y-axis (not n for each, that is confusing).
- Appendix A.
- This feels (like introduction) a tad dated for references on previous work.
- Stating ‘in recent years’ but citing a reference from 2017 does not work well.
- For keywords, how were plural, and slight spelling errors deal with? If NLP dealt with these, then state that.
- For which were sentiments were negative, positive and netural, it would be nice to show a table of examples.
Citation: https://doi.org/10.5194/nhess-2023-6-RC1 -
RC2: 'Comment on nhess-2023-6', Anonymous Referee #2, 09 Apr 2024
The manuscript presents an innovative study on the impact of small-scale geohazards, specifically pre-eruptive seismic unrest and a volcanic eruption in Iceland in 2021, on the local population's expressed sentiments via social media, utilizing natural language processing (NLP) techniques. The study's novelty lies in using social media data to evaluate the emotional and mental health impacts of geohazards, offering insights that could be instrumental in risk management and emergency response planning. The methodological approach, combining manual and AI-assisted sentiment analysis, is particularly commendable for its attempt to navigate the challenges of language specificity and context sensitivity in sentiment analysis. I recommend this manuscript for publication following a 'minor revision' to address the points raised.
Strengths of the manuscript:
- Leveraging NLP to analyze social media data for sentiment analysis related to geohazards is innovative and provides a scalable method for real-time sentiment tracking.
- The discovery that small-scale geohazards can cause significant emotional impacts, with a distinction between negative sentiments during seismic unrest and positive sentiments during the eruption phase, is an important contribution to both geohazard risk management and mental health fields.
- The detailed description of the methodology, including the adaptation of the model to handle Icelandic sentiment analysis and the efforts to mitigate bias and inaccuracies, showcases a commendable level of methodological rigor.
Areas for Improvement
- The study's reliance on Twitter data, while innovative, raises questions about the representativeness of the findings. With Twitter's user demographic not fully representing the entire population, future research could benefit from incorporating data from multiple social media platforms to capture a wider range of sentiments. I suggest that the authors explicitly state this limitation.
- While the adaptation of the model for Icelandic is a strength, the reliance on a model initially trained on English data and the challenges associated with keyword masking deserve further discussion. The implications of these methodological choices on the findings' accuracy and generalizability should be addressed more thoroughly.
- The study establishes that small-scale geohazards have significant emotional impacts but does not quantify these impacts in a way that could be useful for risk management or emergency response planning. Future studies could explore methodologies for quantifying emotional impacts regarding mental health outcomes or risk perception changes.
- A deeper discussion on the limitations of using an NLP model trained primarily on English data for analyzing Icelandic tweets, including any potential biases or inaccuracies introduced, would strengthen the study.
- Developing and discussing methods for quantifying the emotional and mental health impacts of geohazards could significantly enhance the practical implications of the research.
- While the study focuses on immediate sentiments expressed during the geohazards, a discussion on the potential long-term mental health impacts would provide a more comprehensive view of the geohazards' effects.
This manuscript provides valuable insights into the emotional impacts of geohazards on local populations and introduces a novel methodological approach to sentiment analysis in this context. With improvements in representativeness, model transparency, and impact quantification, this work could significantly contribute to the fields of geohazard risk management and mental health. I recommend this manuscript for publication following a 'minor revision' to address the points raised.
Citation: https://doi.org/10.5194/nhess-2023-6-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 418 | 111 | 26 | 555 | 19 | 23 |
- HTML: 418
- PDF: 111
- XML: 26
- Total: 555
- BibTeX: 19
- EndNote: 23
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1