Hypergraphs with Attention on Reviews for Explainable Recommendation

Code

• github.com/PreferredAI/cornac
  tf★ 1,024
• github.com/PreferredAI/Hypar
  none★ 8

Abstract

Given a recommender system based on reviews, the chal-lenges are how to effectively represent the review data and how to explainthe produced recommendations. We propose a novel review-specific Hy-pergraph (HG) model, and further introduce a model-agnostic explaina-bility module. The HG model captures high-order connections betweenusers, items, aspects, and opinions while maintaining information aboutthe review. The explainability module can use the HG model to ex-plain a prediction generated by any model. We propose a path-restrictedreview-selection method biased by the user preference for item reviewsand propose a novel explanation method based on a review graph. Ex-periments on real-world datasets confirm the ability of the HG model tocapture appropriate explanations.

Tasks

Reproductions