A Plug-and-play Model-agnostic Embedding Enhancement Approach for Explainable Recommendation

2025-09-03

AI Summary

Contemporary multimedia recommendation systems rely heavily on media similarity modeling, yet their embedding representations suffer from limited semantic fidelity and interpretability due to insufficient attention to representational expressiveness and interaction-level value assessment. To address this, we propose RVRec—a model-agnostic, plug-and-play embedding enhancement framework. First, it introduces a negative 2-Wasserstein distance contrastive loss to strengthen the semantic discriminability of embeddings. Second, it reweights user–media interactions via multivariate Shapley values, explicitly quantifying each interaction’s contribution to personalized recommendation. By integrating probabilistic embedding optimization with interpretability-driven contrastive learning, RVRec consistently improves both recommendation accuracy and transparency across multiple state-of-the-art base models and real-world benchmarks, outperforming all SOTA baselines.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Knowledge Representation and Reasoning: Preferences

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Existing multimedia recommender systems provide users with suggestions of media by evaluating the similarities, such as games and movies. To enhance the semantics and explainability of embeddings, it is a consensus to apply additional information (e.g., interactions, contexts, popularity). However, without systematic consideration of representativeness and value, the utility and explainability of embedding drops drastically. Hence, we introduce RVRec, a plug-and-play model-agnostic embedding enhancement approach that can improve both personality and explainability of existing systems. Specifically, we propose a probability-based embedding optimization method that uses a contrastive loss based on negative 2-Wasserstein distance to learn to enhance the representativeness of the embeddings. In addtion, we introduce a reweighing method based on multivariate Shapley values strategy to evaluate and explore the value of interactions and embeddings. Extensive experiments on multiple backbone recommenders and real-world datasets show that RVRec can improve the personalization and explainability of existing recommenders, outperforming state-of-the-art baselines.

Problem

• Enhancing embedding representativeness and value for explainable recommendations
• Improving personalization and explainability in multimedia recommender systems
• Addressing utility drop from unsystematic interaction and embedding evaluation

Innovation

• Plug-and-play model-agnostic embedding enhancement
• Contrastive loss with negative 2-Wasserstein distance optimization
• Multivariate Shapley values for interaction value assessment