HyperSegRec: enhanced hypergraph-based recommendation system with user segmentation and item similarity learning

Social connections are frequently leveraged as supplementary data to enhance recommendation systems (RS). In real-world scenarios, the relationships between multiple users within social networks exhibit considerable complexity and diversity. However, prevalent recommendation approaches based on graphs tend to assume pairwise interactions to uncover user preferences. This approach overlooks the influence of intricate and multifaceted social relationships on user preferences, representing a neglect of the higher-order complexities inherent in user relations. It is also observed that most approaches neglect the fact that similar items have similar shared interests when grouped with users. This shows that there is an active connection among the attributes of items and users. Hence, this study works on the drawback of current Social RS: 1. Relations among the items are not explored well, 2. Social relations and homogeneity are not diverse. To address these issues, this study introduces a novel methodology based on clustering and heterogenous hypergraph, termed as HyperSegRec. HyperSegRec (Hypergraph Segmentation based RS) encompasses three parallel processes: User segmentation through Clustering, Item Similarity through extended graph using the cosine similarity matrix method, and a Weighted aggregation process, these processes models the user-item (where user-user as well as user-item relations is mined through clustering and hypergraph representation) and the item-item relations are mined through an extended graph matrix which is constructed with the cosine similarity method and hypergraph representation. These process in the proposed approach are trained individually over BPR (Bayesian Personalized Ranking) based loss function and evaluated against state-of-the-art models based on Recall and NDCG for top-K (K = 10, 20, 50) Recommendations. The proposed model is evaluated on two datasets: Movieslens 100 k and Amazon beauty products review. Compared to other state-of-the-art models, HyperSegRec achieves a 9.92% improvement on NDGC@50. Hence, HyperSegRec demonstrates superior performance in capturing user preferences and providing precise recommendations compared to existing state-of-the-art models, showcasing its potential to enhance recommendation systems for diverse datasets.