Exploration of Model-Agnostic Explainable Recommender System Using Association Rule Mining

Abstract

A recommender system that can simultaneously recommend and explain the reason for the recommendation is called an explainable recommender system. Among them, an approach that generates explanations independently from the recommender algorithm is called a model-agnostic approach. In this study, we propose Explanation Variety and Explanation Diversity as diversity metrics of recommendation explanations and a model-agnostic approach that uses association rule mining to generate explanations. In the existing model, only items that the user has evaluated in the past could be used for explanation, but the proposed model can also present auxiliary information that users have. According to the experiment with matrix factorization and MovieLens and LD-gourmet datasets, we confirmed that the proposed model improved 41 points for Model Fidelity and 77 points for Explanation Variety.