CHARM: A Multimodal Graph Foundation Model with Hierarchical Context Modeling for Zero-Shot Transfer
CHARM targets zero-shot transfer for graphs whose nodes carry text, images, or other modalities.
The paper says existing graph foundation models either need downstream adaptation or stay too narrow for multimodal, cross-domain use. CHARM replaces raw nodes with hierarchical graph contexts meant to capture multimodal semantics and cross-modal relations. Those contexts are encoded into graph tokens for a large language model, with experiments reporting consistent gains on zero-shot multimodal graph tasks. ArXiv · AI/CL/LG's note
The paper says existing graph foundation models either need downstream adaptation or stay too narrow for multimodal, cross-domain use. CHARM replaces raw nodes with hierarchical graph contexts meant to capture multimodal semantics and cross-modal relations. Those contexts are encoded into graph tokens for a large language model, with experiments reporting consistent gains on zero-shot multimodal graph tasks. ArXiv · AI/CL/LG's note
score 4