Knowledge-Aware Multispace Embedding Learning for Personalized Recommendation

Recommender systems help users filter items they may be interested in from massive multimedia content to alleviate information overload. Collaborative filtering-based models perform recommendation relying on users' historical interactions, which meets great difficulty in modeling users' interests with extremely sparse interactions. Fortunately, the rich semantics hidden in items may be promising in helping to describing users' interests. In this work, we explore the semantic correlations between items on modeling users' interests and propose knowledge-aware multispace embedding learning (KMEL) for personalized recommendation. KMEL attempts to model users' interests across semantic structures to leverage valuable knowledge. High-order semantic collaborative signals are extracted in multiple independent semantic spaces and aggregated to describe users' interests in each specific semantic. The semantic embeddings are adaptively integrated with a target-aware attention mechanism to learn cross-space multisemantic embeddings for users and items, which are fed to the subsequent pairwise interaction layer for personalized recommendation. Experiments on real-world datasets demonstrate the effectiveness of the proposed KMEL model.