Abstract: This paper proposes a matrix probabilistic factorization model which integrates user dynamic tags and trust relationships to solve the problem of cold boot and data sparseness of collaborative filtering recommendation algorithm. Firstly, the model builds the dynamic contact of the user sets, tag sets and item sets, establishes user dynamic preference matrix. Secondly, it constructs user trust relationship matrix based on user social network information, the trust matrix uses the user trust feedback mechanism to update the trust value among users in real time. Finally, it integrates user dynamic tags and trust relationships, and carries out simulation experiments on MovieLens and Jester_Joke_data data sets. The results show that the proposed algorithm achieves better results in terms of absolute error mean, precision and recall rate, and effectively improves the performance of collaborative filtering recommendation algorithm to some extent.

Key words: collaborative filtering recommendation, cold boot, data sparseness, dynamic tags, trust relationships, matrix probabilistic factorization model

摘要： 针对协同过滤推荐算法中的冷启动以及数据稀疏问题，提出一种融合用户动态标签和用户信任关系的矩阵概率分解模型。该模型首先通过构建用户集、标签集和物品集三者间的动态联系，建立用户动态偏好矩阵；接着构建基于用户社会网络信息的用户信任关系矩阵，该信任关系矩阵使用用户信任反馈机制以实时更新用户间的信任值；最后提出融合用户动态标签和用户信任关系的矩阵概率分解模型，并在MovieLens与Jester_Joke_data数据集上进行仿真实验。实验结果表明，该算法在绝对误差均值、准确率与召回率方面获得了较好的效果，在一定程度上能有效提高了协同过滤推荐算法的性能。

关键词: 协同过滤推荐, 冷启动, 数据稀疏, 动态标签, 信任关系, 矩阵概率分解模型