Neural multi-task collaborative filtering

Abstract

In recommendation systems, the rating matrix is often very sparse. Collaborative filtering recommendation algorithms cannot be applied to sparse matrices or used in cold start problems. Although the users’ trust relationships provide some useful additional information for recommendation systems, the existing research has not incorporated the rating matrix and trust relationships well. The trust relationship itself also has the problem of data sparsity. With a focus on the problem of sparsity and low accuracy in collaborative filtering algorithms, this paper proposes a general framework, called neural multi-task collaborative filtering (NMCF), which can simultaneously predict the rating and trust relationships. That is, the rating of the same user in e-commerce platforms and the trust relationships in social networks promote and complement each other and help to improve the prediction accuracy of both. The study results for three datasets of real-world show that our algorithm performs better in recommendation, and the accuracy of the proposed algorithm is significantly improved compared with that of the comparison models.