Cross-Modal Attention Network with Dual Graph Learning in Multimodal Recommendation

TL;DR

CRANE tackles two core issues in multimodal recommendations: shallow fusion and asymmetric user/item representations. It combines a heterogeneous user-item graph with a recursive cross-modal attention mechanism and a symmetric item-item semantic graph, optimized by BPR and a self-supervised contrastive objective to fuse behavioral and semantic signals. The approach yields consistent improvements (around 5% on average) over state-of-the-art baselines across four real-world datasets, while maintaining practical efficiency and scalable convergence. The results demonstrate robust, high-order cross-modal fusion that enhances both predictive accuracy and interpretability in large-scale multimodal recommender systems.

Abstract

Multimedia recommendation systems leverage user-item interactions and multimodal information to capture user preferences, enabling more accurate and personalized recommendations. Despite notable advancements, existing approaches still face two critical limitations: first, shallow modality fusion often relies on simple concatenation, failing to exploit rich synergic intra- and inter-modal relationships; second, asymmetric feature treatment-where users are only characterized by interaction IDs while items benefit from rich multimodal content-hinders the learning of a shared semantic space. To address these issues, we propose a Cross-modal Recursive Attention Network with dual graph Embedding (CRANE). To tackle shallow fusion, we design a core Recursive Cross-Modal Attention (RCA) mechanism that iteratively refines modality features based on cross-correlations in a joint latent space, effectively capturing high-order intra- and inter-modal dependencies. For symmetric multimodal learning, we explicitly construct users' multimodal profiles by aggregating features of their interacted items. Furthermore, CRANE integrates a symmetric dual-graph framework-comprising a heterogeneous user-item interaction graph and a homogeneous item-item semantic graph-unified by a self-supervised contrastive learning objective to fuse behavioral and semantic signals. Despite these complex modeling capabilities, CRANE maintains high computational efficiency. Theoretical and empirical analyses confirm its scalability and high practical efficiency, achieving faster convergence on small datasets and superior performance ceilings on large-scale ones. Comprehensive experiments on four public real-world datasets validate an average 5% improvement in key metrics over state-of-the-art baselines.

Figures (8)

• Figure 1: An illustration of how visual and textual modalities provide complementary signals to bridge the semantic gap. The system identifies latent user preferences by aligning fine-grained attributes, such as canopy design and outdoor durability, between the historical interaction and candidate recommendations.
• Figure 2: The overall framework of CRANE. Initially, the model leverages the heterogeneous user-item graph (left) to extract fundamental collaborative signals from historical interactions. Afterwards, the core RCA mechanism (right) iteratively aligns and refines visual and textual features to uncover deep intra- and inter-modal dependencies. These refined features are then employed to construct a homogeneous item-item semantic graph (middle), injecting high-order semantic affinities into item representations. Finally, the collaborative and semantic views are integrated to predict user preferences, optimized jointly by BPR and contrastive losses.
• Figure 3: The pipeline of Item-Item Semantic Graph construction. It details the transformation from raw multimodal features to the final adjacency matrix. The process involves latent feature alignment, dimension reduction, and the application of the Cross-Modal Attention Network to derive sparse, high-order semantic connections between items.
• Figure 4: Evaluation of Multimodal Fusion Strategies
• Figure 5: Sensitivity analysis of key hyperparameters on. The darker color indicates better performance.