计算机工程

• •    

融合用户信任度的概率矩阵分解组推荐系统

  

  • 发布日期:2021-01-14

Incorporating User Trust with Probabilistic Matrix Factorization for Group Recommender System

  • Published:2021-01-14

摘要: 随着推荐系统的应用越来越广泛,群组推荐引起了越来越多的关注。群组成员之间的交互关系往往对群组推荐结果 具有很大影响,而以往的群组推荐算法较少考虑用户信任度的重要性,使得社交关系信息的利用率较低。针对该问题提出一 种融合用户信任度的概率矩阵群组推荐算法,该方法首先利用概率矩阵分解法进行用户信任度计算,然后改进概率矩阵分解 的后验概率计算,加入用户间的信任度因素,极大化该后验概率以获得预测评分。最后在群组成员偏好融合过程中使用基于 用户信任度的权重策略。基于 Epinions 数据集和 FilmTrust 数据集上的实验结果表明,该方法在均方根误差和命中率这两项 评估指标上优于 NeuMF、RippleNet 等其他对比方法。

Abstract: 】With the increasing application of recommendation systems, group recommendation has attracted more and more attention. The interaction between group members often has a great influence on the group recommendation results, but the previous group recommendation algorithm rarely considers the importance of user trust, which wastes a lot of existing information. To address this problem, a probability matrix group recommendation algorithm is proposed, which firstly uses the probability matrix decomposition method to calculate user trustworthiness, and then improves the a posteriori probability of probability matrix decomposition, adds the trust factor between users, and maximizes the a posteriori probability to obtain the prediction score. Finally, a weighting strategy based on user trustworthiness is used in the group member preference fusion process. Based on experimental results on the Epinions dataset and the FilmTrust dataset, the method outperforms other comparison methods such as NeuMF and RippleNet on two evaluation metrics, root mean square error and hit rate.