Abstract: Considering the sparsity and cold-start problems of traditional collaborative filtering recommendation algorithms, a new collaborative filtering algorithm based on trust and matrix factorization is presented. Firstly, an implicit trust calculation method based on user’s rating is proposed. This method considers the user’s similarity and interaction experience comprehensively and uses the trust propagation method to make users who do not have direct trust obtain indirect trust. Secondly, the user’s explicit trust information and the implicit trust information are integrated into the SVD++ algorithm through a dynamic factor. Experiments under the FilmTrust dataset show that the proposed algorithm outperforms other matrix decomposition recommendation algorithms on rating prediction and it also performs well on the cold start users’ rating prediction.

Key words: implicit trust, explicit trust, matrix factorization, collaborative filtering

摘要： 针对传统协同过滤算法普遍存在的稀疏性和冷启动问题，提出一种基于信任和矩阵分解的协同过滤推荐算法。提出一种基于用户评分值的隐式信任计算方法，该方法综合考虑用户的相似性和交互经验，运用信任传播方法使不存在直接信任的用户获得间接信任；通过动态因子将显式信任和隐式信任融入到SVD++算法当中。FilmTrust数据集下的实验表明，与其他矩阵分解推荐算法相比，该方法具有更好的预测效果，在冷启动用户的评分预测上也有很好的表现。

关键词: 隐式信任, 显式信任, 矩阵分解, 协同过滤