Saved in:
Bibliographic Details
Main Authors: Zuo, Yilong, Li, Xunkai, Zhang, Zhihan, Li, Ronghua, Wang, Guoren
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.12271
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910127949348864
author Zuo, Yilong
Li, Xunkai
Zhang, Zhihan
Li, Ronghua
Wang, Guoren
author_facet Zuo, Yilong
Li, Xunkai
Zhang, Zhihan
Li, Ronghua
Wang, Guoren
contents 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. Concretely, RoleMAG distinguishes whether a neighbor should provide shared, complementary, or heterophilous signals, and routes them through separate propagation channels. This enables cross-modal completion from complementary neighbors while keeping heterophilous ones out of shared smoothing. Extensive experiments on three graph-centric MAG benchmarks show that RoleMAG achieves the best results on RedditS and Bili\_Dance, while remaining competitive on Toys. Ablation, robustness, and efficiency analyses further support the effectiveness of the proposed role-aware propagation design. Our code is available at https://anonymous.4open.science/r/RoleMAG-7EE0/
format Preprint
id arxiv_https___arxiv_org_abs_2604_12271
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle RoleMAG: Learning Neighbor Roles in Multimodal Graphs
Zuo, Yilong
Li, Xunkai
Zhang, Zhihan
Li, Ronghua
Wang, Guoren
Machine Learning
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. Concretely, RoleMAG distinguishes whether a neighbor should provide shared, complementary, or heterophilous signals, and routes them through separate propagation channels. This enables cross-modal completion from complementary neighbors while keeping heterophilous ones out of shared smoothing. Extensive experiments on three graph-centric MAG benchmarks show that RoleMAG achieves the best results on RedditS and Bili\_Dance, while remaining competitive on Toys. Ablation, robustness, and efficiency analyses further support the effectiveness of the proposed role-aware propagation design. Our code is available at https://anonymous.4open.science/r/RoleMAG-7EE0/
title RoleMAG: Learning Neighbor Roles in Multimodal Graphs
topic Machine Learning
url https://arxiv.org/abs/2604.12271