基于潜在标签挖掘和细粒度偏好的个性化标签推荐

Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-11-29 Cooperative journals: 《计算机应用研究》

李红梅 刁兴春 曹建军 张磊 冯钦

Abstract： To further improve the performance of personalized tag recommendation, this paper argued that traditional methods ignore the potential and informative tags hidden in the context of users and items. Aimed at this, this paper proposed a novel personalized tag recommendation method BPR-PITF-P based on potential tag mining and fine-grained preference. Firstly, BPR-PITF-P leverages the context information of both users and items to mine potential and useful tags, and gets three kinds of tags: positive tags, potential tags, and negative tags. Based on the above, it translates the traditional pairwise preference into fine-grained preference relationship among user-item post and tags. This kind of treatment helps alleviate the sparse problem of tagging data. Second, combined with pairwise interaction tensor factorization method to predict preference value, BPR-PITF-P models the preference relationship based on the optimization criteria of Bayesian personalized ranking, and develops a personalized tag recommendation model followed by optimization algorithm. The comparison results show that our proposed method could improve tag recommendation performance in the premise of guarantee convergence speed.

 Hits 1940  Downloads 1149  Comment