一种融合社会化信息的改进单类协同过滤方法研究

Study of improved one class collaborative filtering method merged with social information

文章在负例抽取阶段考虑用户的活跃度和项目间相似度,以及在概率矩阵分解时融合用户好友关系和项目标签社会化信息的基础上,提出了一种融合社会化信息的改进单类协同过滤(one class collaborative filtering with social information,OCCF-SI)方法,并在科研社交网络CiteULike的真实数据集上进行了实验。研究结果表明,与其他传统的推荐方法相比,该文所提出的方法取得了较好的推荐结果,具有良好的可扩展性。

In this paper, the improved one class collaborative filtering with social information（OCCF- SI） is proposed. On the one hand, the user＇s activity and the similarity between projects are consid- ered when extracting the negative cases; on the other hand, the social information of user＇s friends relations and project＇s labels is merged into the probability matrix factorization. The experiments on the real dataset in a scientific social network named CiteULike are conducted. The experimental re- sults show that compared to other traditional recommendation methods, the proposed method gets the best recommendation results and performs well in scalability.