Explainable Multi-type Item Recommendation System Based on Knowledge Graph

Chao Chang, Junming Zhou, Weisheng Li, Zhengyang Wu, Jing Gao, Yong Tang

Abstract: Knowledge graphs and recommender systems have significant potential for improving recommendation accuracy and interpretability. However, most existing methods focus on single-type item recommendations and offer limited explainability of their recommendations. In this paper, we propose a novel framework knowledge graph transformer network (KGTN) for explainable multiple types of item recommendation, which aims to recommend items with different formats and categories to users simultaneously. Unlike previous methods that rely on predefined meta-paths, KGTN mines hidden path relationships in a collaborative knowledge graph. The KGTN integrates a transformer-based model with a meta-path-based graph convolutional network to effectively learn user-item representations and capture the complex relationships among users, items, and their corresponding attributes. Finally, we use the critical path in the learned useful meta-path graph as an explanation. Experimental results on two real-world datasets demonstrate KGTN’s superiority over state-of-the-art methods in terms of recommendation performance and explainability. Furthermore, KGTN is shown to be effective at handling data sparsity and cold-start problems.