RoleMAG: Learning Neighbor Roles in Multimodal Graphs

ArXi:2604.12271v1 Announce Type: new Multimodal attributed graphs (MAGs) combine multimodal node attributes with structured relations. However, existing methods usually perform shared message passing on a single graph and implicitly assume that the same neighbors are equally useful for all modalities. In practice, neighbors that benefit one modality may interfere with another, blurring modality-specific signals under shared propagation. To address this issue, we propose RoleMAG, a multimodal graph framework that learns how different neighbors should participate in propagation.