Personalized recommendation based on hierarchical interest overlapping community

• Published in: Information sciences Vol. 479; pp. 55 - 75
• Main Authors: Zheng, Jianxing, Wang, Suge, Li, Deyu, Zhang, Bofeng
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.04.2019
• Subjects: Hierarchical interest overlapping community; Multi-granularity similarity; Ontology user profile; Personalized recommendation; Ontology user profile; Hierarchical interest overlapping community; Personalized recommendation; Multi-granularity similarity
• ISSN: 0020-0255; 1872-6291
• DOI: 10.1016/j.ins.2018.11.054

•Content-based and structured-based interest closenesses build multi-granularity subject similarity between users.•Hierarchical interest overlapping community is detected by considering the interest linkage density.•Memberships of a community are used to predict the preferences of target user. Ontology user profiles describe users’ structural semantic interests. Studying similar relationships between user profiles is crucial to detecting interest overlapping communities. The novel view assumes that hierarchical interests of user profiles can generate multiple similarity relations, which is conducive to forming interest clusters. In this research, we develop a hierarchical interest overlapping community (HIOC) detection method and present a personalized recommendation model. First, content interest closeness and semantic interest closeness between user profiles are computed to measure multi-granularity subject similarity of users. Then, using the multi-granularity subject similarity and follow similarity of users, a heterogeneous hypergraph is constructed to represent an interest network. By application of the interest density peaks mechanism, the HIOC detection method is adopted for identifying communities of interest. Further, personalized interest prediction is implemented by consideration of the memberships of a user in a community and a subject distributed in a community. Finally, we verify the performance of the HIOC detection algorithm on several real networks and validate the effectiveness of the proposed recommendation approach. The experimental results illustrate that the proposed approach outperforms classical recommendation methods in precision and recall.