In recent years, with the rapid development of the Internet, various virtual communities continue to emerge, and the phenomenon of user groups working together is gradually increasing. People begin to pay more attention to group-oriented recommendation. Most of existing group recommendation methods are improved on the memory-based collaborative filtering recommendation method, or think that the members of the group are independent of each other, ignoring the impact of association among the members of the group on the results of group recommendation. In this paper, a group recommendation method based on joint probability matrix decomposition is proposed to better model the group recommendation problem. Firstly, the user-plus-person group information is used to calculate the correlation between users. Secondly, the user correlation matrix is integrated into the process of probability matrix decomposition to get the individual prediction score. Finally, the group-to-item prediction score is obtained by using the common synthesis strategy in group-oriented recommendation problem. Furthermore, the proposed method is compared with existing group recommendation methods. Experiments on CiteULike dataset show that the proposed method achieves better recommendation results in accuracy, recall and other evaluation indicators.

近年来，随着Internet的快速发展，各种虚拟社区不断涌现，用户群体协同工作的现象逐渐增多。人们开始更加重视面向小组的推荐。现有的大多数组推荐方法都在基于内存的协同过滤推荐方法上进行了改进，或者认为组成员彼此独立，而忽略了组成员之间的关联对组推荐结果的影响。本文提出了一种基于联合概率矩阵分解的群体推荐方法，以更好地对群体推荐问题进行建模。首先，使用用户加人组信息来计算用户之间的相关性。其次，用户相关矩阵被集成到概率矩阵分解的过程中，以获得单独的预测分数。最后，采用面向群体推荐问题的通用综合策略，得到了群体对项目的预测分数。此外，将所提出的方法与现有的小组推荐方法进行了比较。在CiteULike数据集上的实验表明，该方法在准确性，召回率和其他评估指标上均取得了较好的推荐结果。