A hypergraph-based framework for personalized recommendations via user preference and dynamics clustering. (15th October 2022)

• Record Type: Journal Article
• Authors: Wang, Zhihui; Chen, Jianrui; Rosas, Fernando E.; Zhu, Tingting
• Is Part Of: Expert systems with applications. Volume 204(2022)
• Publication Date: 2022-10-15
• Subjects: Hypergraph model -- Attention relations -- Payoff function -- Similarity measurement -- Recommendation system; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.117552
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: The ever-increasing number of users and items continuously imposes new challenges to existent clustering-based recommendation algorithms. To better simulate the interactions between users and items in the recommendation system, in this paper, we propose a collaborative filtering recommendation algorithm based on dynamics clustering and similarity measurement in hypergraphs (Hg-PDC). The main idea of Hg-PDC is to discover several interest communities by aggregating users with high attention, and make recommendations within each community, thereby improving the recommendation performance and reducing the time cost. Firstly, we introduce a hypergraph model to capture complex relations beyond pairwise relations, while preserving attention relations in the network. In addition, we construct a novel hypergraph model, which defines a user and his evaluated items to form a hyperedge. Secondly, an extended game dynamics clustering method is proposed for the constructed hypergraph to aggregate users with high attention into the same interest community. Here, we combine the payoff function in game theory with the traditional dynamics clustering method. Finally, we apply the dynamics clustering results and a new similarity measurement strategy with user preferences to recommend items for target users. The effectiveness of Hg-PDC is verified by experiments on six real datasets. Experimental results illustrate that our algorithm outperforms state-of-the-art algorithms in prediction errors and recommendation performance. Highlights: Hypergraph reduces complexity and preserves high-order structure of users and items. Attention relations of users to items considers personal preferences. Novel similarity filters out the items that the two users have different opinions. Dynamics clustering approach combines with game-theoretic payoff function.