Cut to the Chase: Training-free Multimodal Summarization via Chain-of-Events

Researchers introduce CoE, a training-free multimodal summarization framework that uses a Chain-of-Events approach with Hierarchical Event Graph to better understand and summarize content across videos, transcripts, and images. The system achieves significant performance improvements over existing methods, showing average gains of +3.04 ROUGE, +9.51 CIDEr, and +1.88 BERTScore across eight datasets.

• →CoE framework addresses three key challenges in multimodal summarization: domain-specific supervision reliance, weak cross-modal grounding, and flat temporal modeling.
• →The system uses a Hierarchical Event Graph to encode textual semantics and scaffold cross-modal reasoning without requiring training.
• →Testing across eight diverse datasets shows consistent outperformance of state-of-the-art video Chain-of-Thought baselines.
• →The framework demonstrates strong cross-domain generalization and interpretability capabilities.
• →Source code is publicly available on GitHub for research and development purposes.