Location
Hilton Hawaiian Village, Honolulu, Hawaii
Event Website
https://hicss.hawaii.edu/
Start Date
3-1-2024 12:00 AM
End Date
6-1-2024 12:00 AM
Description
Crises send out early warning signals; mostly weak and difficult to detect amidst the noise of everyday life. Signal detection based on social media enables early identification of such signals supporting pro-active organizational responses before a crisis occurs. Nonetheless, social signal detection based on Twitter data is not applied in crisis management in practice as it is challenging due to the high volume of noise. With OSOS, we introduce a method for open-domain social signal detection of crisis-related indicators in tweets. OSOS works with multi-lingual Twitter data and combines multiple state-of-the-art models for data pre-processing (SoMaJo) and data filtration (GPT-3). It excels in crisis domains by leveraging fine-tuned GPT-3\textsuperscript{FT} (Curie) model and achieving benchmark results in the CrisisBench dataset. The method was exemplified within a signaling service for crisis management. We were able to evaluate the proposed approach by means of a data set obtained from Twitter (X) in terms of performance in identifying potential social signals for energy-related crisis events.
Recommended Citation
Janzen, Sabine; Saxena, Prajvi; Baer, Sebastian; and Maass, Wolfgang, ""Listening In": Social Signal Detection for Crisis Prediction" (2024). Hawaii International Conference on System Sciences 2024 (HICSS-57). 7.
https://aisel.aisnet.org/hicss-57/dg/disaster_resilience/7
"Listening In": Social Signal Detection for Crisis Prediction
Hilton Hawaiian Village, Honolulu, Hawaii
Crises send out early warning signals; mostly weak and difficult to detect amidst the noise of everyday life. Signal detection based on social media enables early identification of such signals supporting pro-active organizational responses before a crisis occurs. Nonetheless, social signal detection based on Twitter data is not applied in crisis management in practice as it is challenging due to the high volume of noise. With OSOS, we introduce a method for open-domain social signal detection of crisis-related indicators in tweets. OSOS works with multi-lingual Twitter data and combines multiple state-of-the-art models for data pre-processing (SoMaJo) and data filtration (GPT-3). It excels in crisis domains by leveraging fine-tuned GPT-3\textsuperscript{FT} (Curie) model and achieving benchmark results in the CrisisBench dataset. The method was exemplified within a signaling service for crisis management. We were able to evaluate the proposed approach by means of a data set obtained from Twitter (X) in terms of performance in identifying potential social signals for energy-related crisis events.
https://aisel.aisnet.org/hicss-57/dg/disaster_resilience/7