Abstract—Recently, a sparse linear method (SLIM) is developed for top-N recommender systems, which can produce high-quality recommendations for sparse data sets. SLIM provides a better performance than other existing methods. In this paper, we provide a novel user-item interest method (UIIM) based on bipartite network to improve the performance of SLIM. UIIM generates top-N recommendations by building the user-item interest matrix R with the bipartite network of users and items, calculating the item-item similarity matrix W with SLIM and predicting users’ ratings on items as a dot product of matrix R and W. And we also provide a parallel algorithm based on Spark to learn W. Our results indicate that UIIM provides better performance and recommendation quality than other existing methods and parallel algorithm of learning W outperforms serial algorithm on large-scale data sets.
Index Terms—Top-N recommender systems, bipartite network, UIIM, SLIM, parallel.
Zhixiong Jiang and Chunyang Lu are with CNPC Changping Data Center, Beijing 102206, China (e-mail: jiangzhixiong@cnpc.com.cn, luchunyang@cnpc.com.cn).
Siyuan Zheng and Juan Yang are with Beijing Key Laboratory of Intelligent Telecommunications Software and Multimedia, Beijing University of Posts and Telecommunications, Beijing 100876, China (e-mail: zsybupt@gmail.com, yangjuan@bupt.edu.cn).
[PDF]
Cite:Zhixiong Jiang, Chunyang Lu, Siyuan Zheng, and Juan Yang, "UIIM: A User-Item Interest Model Based on Bipartite Network for Top-N Recommender System," Journal of Advances in Computer Networks vol. 3, no. 3, pp. 251-254, 2015.