Enhancing Explainable Sequential Recommendation With Disentangled Representations and Auxiliary Review Explanations

In recent years, there has been a growing focus on enhancing the explainability of recommendation systems. Most sequential recommenders embed item features and capture user sequential preference with entangled, low-transparency methods. While effective, these approaches often lack transparency and explainability. In addition to explaining user preferences as general recommendations, the challenge of explainable sequential recommendations extends to explaining the evolution of users’ preferences. This article has proposed an explainable sequential recommendation framework with a disentangled heterogeneous graph network and auxiliary review explanations, named DisenAre. First, we disentangle and refine heterogeneous graph embeddings to enhance transparency in explaining user preferences. Second, we introduce a preference evolution module to explore the relationships among user behaviors. Third, for user-friendly explanations, we propose a posthoc auxiliary review explainer that extracts keywords and key sentences as supplementary explanations. The framework is evaluated on two datasets, TMALL and Goodreads, and consistently outperforms existing state-of-the-art methods. Additionally, we conduct a case study to analyze explainability and experiments to determine the optimal disentangled coefficient. The results affirm the effectiveness and explainability of the proposed approach.