Automatic Extraction of Effective Relations in a Knowledge Graph for a Recommendation Explanation System

A knowledge graph represents a network of real-world entities (i.e., objects, events, or concepts) and illustrates the relationship between entities. A recommender system can improve its reasoning and explainability using the knowledge graph. In this paper, we propose a hybrid and modular approach that combines path ranking with graph embedding; it can automatically eliminate the ineffective relations among entities and generate a better relation set for the explanation system. We conducted a user survey for performance evaluation and proved that our proposed approach provided the same quality of explanations for recommended items as our previous approach, which manually selected relations from the knowledge graph.