Multimodal approach improves diverse and accurate recommendations
Multimodal Decoupling for Diversified Recommendation through Semantic and Structural Shaping
Information RetrievalArtificial Intelligence
Summary
Recommendation systems often suggest too many similar items when they focus only on what is most likely to be liked, making recommendations less varied. The authors found that using information from multiple types of data (like images and text) is not always done well, and this limits diversity. They created a new method called MOSS that separates and organizes different types of data better and encourages a wider spread of recommendations while keeping accuracy high. Tests on real data show their approach balances accurate and diverse suggestions better than others.
What this means in practice
- •For ecommerce platform developers: Improve online product suggestions by balancing accuracy with variety using multimodal data integration and category-focused embedding techniques.
- •For media streaming engineers: Enhance multimedia content recommendations by avoiding over-concentration on popular items through semantic and structural shaping methods.
Authors
Changran He, Jiaming Wei, Jianghong Ma, Tianjun Wei, Haijun Zhang, Xiaofeng Zhang
Abstract
Recommender systems often suffer from over-concentrated results when optimized for accuracy, limiting recommendation diversity. While diversity has become an important objective, existing methods still face fundamental limitations from both semantic and structural perspectives. On the one hand, multimodal recommender systems provide rich semantic signals, yet are often treated as auxiliary to ID embeddings, leading to ineffective semantic utilization for diversity. On the other hand, the accuracy--diversity trade-off persists during training, where accuracy gains often come at the cost of diversity. We show that this phenomenon is fundamentally governed by both the structural properties of learned representations and the dynamics of optimization. To address these challenges, we propose MOSS, a unified framework for multimodal decoupling through semantic and structural shaping. Specifically, we first introduce an MLLM-based Decoupled Feature Extraction module to disentangle multimodal content into informative representations. To enhance semantic utilization, we design a Category-aware Embedding Selection module that extracts compact, category-aware semantic features. To address structural limitations, we propose two complementary modules: Alternating Dual-Graph Propagation, which alleviates representation collapse by integrating collaborative and multimodal propagation, and Entropy-based Diversity Regularization, which mitigates optimization bias by promoting balanced category exposure. Extensive experiments on real-world datasets demonstrate that MOSS consistently achieves a superior accuracy--diversity trade-off over strong baselines.