一种弥补群组成员差异的可解释性组推荐方法

Interpretable Group Recommendation Method to Make up for Differences Among Group Members

现有的组推荐方法侧重于寻找组内成员之间的相似性,而忽略了成员之间存在的差异.在群组成员差异性显著的情况下,丢弃群组成员间的差异性,竭力寻求相似性,这样会大大降低群组成员对推荐结果的满意度.针对该问题,本文提出一种弥补群组成员差异的可解释性组推荐方法,通过群组成员差异度计算将群组分为差异性群组和相似性群组.对于相似性群组,采用基于决策权重的群组推荐方法生成推荐列表;对于差异性群组,本文构建了一种考虑成员信任值的群组LDA模型,基于此模型生成推荐列表和解释.实验结果表明,该推荐方法能根据群组成员差异性情况采用不同的推荐策略,提高群组推荐的满意度和推荐精度.

The traditional group recommendation method focuses on finding the similarity between members in the group,while ignoring the differences between members.In the case of significant differences among group members,discard the differences among group members and strive to find similarity,which will greatly reduce the satisfaction of the group members with the recommendation results.In response to this problem,this paper proposes an interpretable group recommendation method to make up for the differences of group members.The group is divided into different groups and similar groups through the calculation of the difference of group members.For similar groups,a group recommendation method based on decision weights is used to generate the recommendation list;for differential groups,this paper builds a group LDA model that considers the trust value of members,and generates recommendation lists and interpretations based on this model.Experimental results show that this recommendation method can adopt different recommendation strategies according to the differences of group members,and improve the satisfaction and recommendation accuracy of group recommendations.