A note on recommendation with explainability based on knowledge graph reasoning using graph masked autoencoders

Abstract

In this paper, we present a novel explainable recommendation method based on Graph Masked Autoencoders (GMAE) and knowledge graph reasoning. Explainable recommendation, which aims to provide reasons for recommendations, is an important research challenge in the field of recommender systems. In recent years, knowledge graph reasoning emerges as one of the prominent solutions for achieving explainable recommendation. However, since conventional methods perform reasoning based solely on features obtained from knowledge graph embedding models, they result in recommendation performance degradation compared to latent factor-based recommendation methods. To tackle this problem, we attempt to capture the high-order relationships between user interaction patterns and items by GMAE, a self-supervised learning method for complex graphs, into the reasoning process. Specifically, we incorporate item features obtained from GMAE into the reward function of a reinforcement learning agent reasoning the knowledge graph. Experiments with real-world datasets have validated the effectiveness of our proposed method.