Subspace Ensemble-Based Neighbor User Searching for Neighborhood-Based Collaborative Filtering

Neighborhood-based collaborative filtering (NCF) typically uses a similarity measure for finding similar users to a target user or similar products on which the target user rated. To find neighbor users, traditional similarity measures rely only on the ratings of co-rated items when calculating similarity of pairwise users. Some hybrid similarity measures can avoid this situation but they suffer from the time-consuming issue. To solve the mentioned issues, the current paper presents an effective method of subspace ensemble-based neighbor user searching (SENUS) for NCF. First, three item subspaces are constructed, or interested, neither interested nor uninterested, and uninterested subspaces. In each subspace, we calculate the co-rating support values for pairwise users. Then, SENUS combines three co-rating support values to get the total co-rating support values for pairwise users, which are utilized to generate direct neighbor users for a target user. For the target user, its neighbor users include direct and indirect ones in SENUS, where its indirect neighbors are the direct neighbors of its direct neighbors. Experimental results on public datasets indicate that the proposed method is promising in recommender systems.