Social Recommender Systems using Collaborative User Network Embedding with Bias
Chuanzhen Li, Han Xiao, Juanjuan Cai, Hui Wang · 2019
Traditional recommendation algorithms suffer from the problems of cold-start users and data sparsity, which significantly degrade prediction accuracy. This paper proposes a novel method CUNE-bias(Collaborative User Network Embedding with bias) to resolve these issues. Firstly, top-k semantic friends are identified only using information from user-item feedbacks. Then, we incorporate the top-k semantic friends information generated by CUNE into matrix factorization as a regularization term. Finally, the bias, independent of any interactions, is involved in rating to improve accuracy. Experimental results on two datasets demonstrate that the proposed method gets higher accuracy than CUNE-MF and other classical methods.