MODEE: A Graph-Based LLM Approach for Open-Domain Event Extraction

Researchers propose MODEE (Multimodal Open-Domain Event Extraction), a novel approach combining large language models with graph-based document modeling to extract events from text without predefined event types. Existing open-domain methods overlook LLMs and fail to capture document-level contextual, structural, and semantic reasoning. MODEE addresses these gaps by integrating graph structures to mitigate LLM issues like lost-in-the-middle and attention dilution. The approach is designed for tasks such as summarization and emergency decision-making. The paper is available on arXiv under ID 2604.21885.