JCSE, vol. 20, no. 1, pp.16-30, 2026

DOI: http://dx.doi.org/10.5626/JCSE.2026.20.1.16

Knowledge-Aware Recommendation Method Based on Multigraph Mining

Zhongxuan Li, Hairong Wang, and Beijing Zhou
College of Computer Science and Engineering, North Minzu University, Yinchuan, Ningxia, China

Abstract: Recommendation methods based on user-item interactions and knowledge graphs often fail to explicitly capture homogeneity information (i.e., similarities among users and items) and tend to overlook the long-tail distribution of relations in knowledge graphs. To address these issues, this paper proposes a knowledge-aware recommendation framework grounded in multi-graph mining. This method constructs a collaborative neighbor graph (i.e., user-user and item-item interaction graph) on the basis of user-item interaction information to mine neighboring relationships between users and items. It performs relation clustering and generates latent semantic relations that mitigate the impact of the long-tail effect and reconstructs the knowledge graph. By jointly exploiting the structural and semantic information of the reconstructed graph, our model learns personalized user representations. Furthermore, by integrating local collaborative signals with global semantic features, the method captures both commonalities and personalized preferences, leading to refined representations of users and items. To evaluate the performance of the proposed method, experimental evaluations were carried out across the Last.FM, MovieLens-1M, and Book-Crossing datasets. Findings confirm the method’s effectiveness, highlighting that leveraging semantic information from multiple graphs and addressing the long-tail relationship distribution in knowledge graphs can boost recommendation accuracy.

Keyword: Knowledge graph; Long tail effect; Multi-graph mining; Recommendation method; Collaborative neighbor graph

Full Paper:   9 Downloads, 15 View