VISPATH Framework for Multimodal Knowledge Graph Reasoning

VISPATH introduces visual-intent-guided path reasoning to prevent multimodal KGQA systems from reverting to text-only search during multi-hop reasoning. The framework uses multimodal cues at intermediate hops to improve entity grounding and path discovery in complex knowledge graphs.