CAMPA: Efficient and Aligned Multimodal Graph Learning via Decoupled Propagation and Aggregation

TL;DR

CAMPA introduces a novel decoupled multimodal graph learning framework that effectively aligns cross-modal information during propagation and aggregation, leading to improved efficiency and performance.

Contribution

It proposes a two-stage alignment mechanism in decoupled MGNNs to address modal conflict, enhancing scalability and semantic consistency.

Findings

CAMPA outperforms existing methods on benchmark datasets.

Decoupled MGNNs with CAMPA are more scalable for large graphs.

The framework maintains efficiency while improving accuracy.

Abstract

Multimodal Graph Neural Networks (MGNNs) have shown strong potential for learning from multimodal attributed graphs, yet most existing approaches rely on tightly coupled architectures that suffer from prohibitive computational overhead. In this paper, we present a systematic empirical analysis showing that decoupled MGNNs are substantially more efficient and scalable for large-scale graph learning. However, we identify a critical bottleneck in existing decoupled pipelines, namely modal conflict, which arises in both the propagation and aggregation stages. Specifically, independent multi-hop diffusion causes cross-modal semantic divergence during propagation, while naive fusion fails to align multi-hop feature trajectories during aggregation, jointly limiting effective representation learning. To address this challenge, we propose CAMPA, a Cross-modal Aligned Multimodal Propagation & Aggregation framework for decoupled multimodal graph learning. Concretely, CAMPA introduces a two-stage alignment mechanism: (1) cross-modal aligned propagation, which injects cross-modal similarity priors into message passing to preserve semantic consistency without additional parameter overhead; (2) trajectory aligned aggregation, which leverages trajectory-level self-attention and cross-attention to capture and align long-range dependencies across modalities and hops. Extensive experiments on diverse benchmark datasets and tasks demonstrate that CAMPA consistently outperforms strong coupled and decoupled baselines while preserving the efficiency advantages of the decoupled paradigm.