IEEE Access · 2019 · 19 citations · 51 references
ComputingData ScienceMachine LearningRecommender SystemsSocial InfluenceCold-start RecommendationCollaborative FilteringMatrix FactorizationRecommendation SystemsSocial RecommendationTrust-aware RecommendationData SparsitySocial Collaborative FilteringSocial Influence In RecommendationSocial Trust NetworksTrust Network AnalysisTrust-aware Matrix FactorizationTrust-based Collaborative FilteringUser-based Collaborative Filtering
As one of the most successful recommendation techniques, collaborative filtering provides a useful recommendation by associating an active user with a crowd of users who share the same interests. Although some achievements have been achieved both in theory and practice, the efficiency of recommender systems has been negatively affected by the problems of cold start and data sparsity recently. To solve the above problems, the trust relationship among users is employed into recommender systems to build a learning model to further promote the prediction quality and users' satisfaction. However, most of the existing social networks-based recommendation algorithms fail to take into account the fact that users with different levels of trust and backgrounds, that is, user's social status and homophily have different degrees of influence on their friends. In this paper, a novel social matrix factorization-based recommendation method is proposed to improve the recommendation quality by fusing user's social status and homophily. User's social status and homophily play important roles in improving the performance of recommender systems. We first build a user's trust relationship network based on user social relationships and the rating information. Then, the degree of trust is calculated through the trust propagation method and the PageRank algorithm. Finally, the trust relationship is integrated into the matrix factorization model to accurately predict unknown ratings. The proposed method is evaluated using real-life datasets including the Epinions and Douban datasets. The experimental results and comparisons demonstrate that the proposed approach is superior to the existing social networks-based recommendation algorithms.
51
Learning the parts of objects by non-negative matrix factorization
Daniel D. Lee, H. Sebastian Seung · Nature · 1999 · 14.1K citations
Probabilistic Matrix Factorization
Andriy Mnih, Ruslan Salakhutdinov · Neural Information Processing Systems · 2007 · 3.6K citations
Performance of recommender algorithms on top-n recommendation tasks
Paolo Cremonesi, Yehuda Koren, Roberto Turrin · 2010 · 1.4K citations
Hao Ma, Haixuan Yang, Michael R. Lyu et al. · 2008 · 1.4K citations