A Regularized Hypergraph Recommendation Algorithm with Attention Mechanism

Abstract

Collaborative filtering is one of the most commonly used recommendation technologies in recommender systems. However, it has the problems of sparse score data and low recommendation accuracy. To address these problems, we propose a regularized hypergraph recommendation algorithm with attention mechanism. We can fully mine the high-order relationships in the hyperedges without loss of information. In addition, we apply users’ attributes information to calculate the attraction of an item to the attributes, and then get the similarity between items. In order to calculate the similarity more accurately, the relations of items to attributes are divided into two cases: likes and dislikes. Moreover, according to the different attention of users to different items, we introduce the attention mechanism. Finally, we build and optimize the regularization function to obtain the predictive scores and recommendation lists. Experimental results on Movielens-100K and Movielens-1M demonstrate the effectiveness of our proposed algorithm.