01 / MOTIVATION
Why this work matters
When crisis information needs are urgent, generated answers should be both informative and actionable. Different models can contribute different strengths.
A fusion approach combining multiple language models and retrieval for timely and actionable crisis communication.
02 / METHODOLOGY
Research approach
The earlier ISCRAM study explores retrieval-supported dynamic fusion to combine complementary language models for crisis communication.
03 / CONTRIBUTIONS
Main contribution
Comparing prompting, RAG, and dynamic fusion of LLMs for crisis information needs.
Consult the complete publication for numerical findings and experimental qualifications.
04 / PUBLICATION
Read the paper
Xiaoying Song, Anirban Saha Anik, Vanessa Frías-Martínez, and Lingzi Hong. (2025). Dynamic Fusion of Large Language Models for Crisis Communication. Proceedings of the International ISCRAM Conference, Volume 22. https://doi.org/10.59297/nqysjq45