Semantic manifold modularization-based ranking for image recommendation

Abstract As the Internet confronts the multimedia explosion, it becomes urgent to investigate personalized recommendation for alleviating information overload and improving users’ experience. Most personalized recommendation approaches pay their attention to collaborative filtering over users’ interactions, which suffers greatly from the highly sparse interactions. In image recommendation, visual correlations among images that users consumed provide a piece of intrinsic evidence to reveal users’ interests. It inspires us to investigate image recommendation over the dense visual graph of images instead of the sparse user interaction graph. In this paper, we propose a semantic manifold modularization-based ranking (MMR) for image recommendation. MMR leverages the dense visual manifold to propagate users’ historical records and infer user-image correlations for image recommendation. Especially, it constrains interest propagation within semantic visual compact groups by manifold modularization to make a tradeoff between users’ personality and graph smoothness in propagation. Experimental results demonstrate that user-consumed visual correlations play actively to capture users’ interests, and the proposed MMR can infer user-image correlations via visual manifold propagation for image recommendation.