Research on disaster event situational awareness based on social media big data
Accurate perception of disaster event situational relies on the timely and effective acquisition of relevant data carrying event information, as well as in-depth understanding and analysis of such data. Social media big data contains a wealth of event information. However, its characteristics of bei...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | Chinese |
| Published: |
China InfoCom Media Group
2025-01-01
|
| Series: | 大数据 |
| Subjects: | |
| Online Access: | http://www.j-bigdataresearch.com.cn/zh/article/doi/10.11959/j.issn.2096-0271.2025080/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Accurate perception of disaster event situational relies on the timely and effective acquisition of relevant data carrying event information, as well as in-depth understanding and analysis of such data. Social media big data contains a wealth of event information. However, its characteristics of being voluminous, unstructured, and spatiotemporally sensitive pose significant challenges for the dynamic and complex awareness of disaster event situational. From the perspective of social media big data, we firstly constructed a causal knowledge graph of disaster events to effectively integrate heterogeneous information from social media big data, addressing the issues of unstructured data and spatiotemporal sensitivity. Secondly, we leveraged large language models and fine-tuning techniques to enhance the reasoning capability of disaster event evolution processes. Moreover, through the fine-tuned generative pre-trained model, we could more accurately identify causal sub-events of disaster events that were targeted and practical, effectively addressing the challenges brought by the large volume of data and information redundancy. Finally, a disaster event awareness system was designed to assist relevant personnel in quickly and comprehensively understanding and analyzing disaster situations through user-system interaction. Experimental results show that the system achieves an average F1 score of 0.891 in disaster event-related text classification tasks, significantly outperforming baseline models. In terms of causal relationship generation, the fine-tuned generative pre-trained model can more accurately identify targeted and practical causal sub-events of disaster events, effectively improving the accuracy and efficiency of disaster event situational awareness. |
|---|---|
| ISSN: | 2096-0271 |