PRISM: Personalized Recommendation via Information Synergy Module

TL;DR

PRISM addresses multimodal sequential recommendation by explicitly disentangling modality interactions into uniqueness, redundancy, and synergy using an Interaction Expert Layer. It introduces an Adaptive Fusion Layer to personalize fusion weights based on user preferences, achieving interpretable and targeted use of multimodal signals. The method forms four experts (uni-i, uni-t, syn, rdn) trained with PID-inspired losses and integrates seamlessly with existing SR backbones, maintaining training objectives with the backbone's native loss, L = L_rec + L_exp. Empirical results on four real-world datasets show consistent gains across backbones and demonstrate improved interpretability and robustness, validating PRISM as a versatile solution for multimodal SR.

Abstract

Multimodal sequential recommendation (MSR) leverages diverse item modalities to improve recommendation accuracy, while achieving effective and adaptive fusion remains challenging. Existing MSR models often overlook synergistic information that emerges only through modality combinations. Moreover, they typically assume a fixed importance for different modality interactions across users. To address these limitations, we propose \textbf{P}ersonalized \textbf{R}ecommend-ation via \textbf{I}nformation \textbf{S}ynergy \textbf{M}odule (PRISM), a plug-and-play framework for sequential recommendation (SR). PRISM explicitly decomposes multimodal information into unique, redundant, and synergistic components through an Interaction Expert Layer and dynamically weights them via an Adaptive Fusion Layer guided by user preferences. This information-theoretic design enables fine-grained disentanglement and personalized fusion of multimodal signals. Extensive experiments on four datasets and three SR backbones demonstrate its effectiveness and versatility. The code is available at https://github.com/YutongLi2024/PRISM.

Figures (5)

• Figure 1: Illustration of multimodal interactions using a red wine example. Higher-level properties, such as the wine’s giftability and collectibility, can only be inferred by synergistically integrating both modalities. The impact of each modality interaction type depends on both user intent and task context.
• Figure 2: The overall architecture of the proposed PRISM.
• Figure 3: Performance comparison of PRISM+InDiRec with different hyperparameter settings on the Yelp dataset.
• Figure 4: Case study of PRISM+InDiRec with and without user preference guidance. With preference guidance, the model increases emphasis on synergistic information, bringing the Ground Truth item to a higher rank.
• Figure 5: The t-SNE visualization of item embeddings on the Yelp dataset.