融合注意力机制的群组推荐研究

现有群组推荐方法在偏好融合时大多采用预定义策略，这种静态方法忽略了群组间用户的交互，难以对复杂的决策过程进行建模，从而影响推荐效果。针对该问题，提出了一种基于注意力机制的群组推荐方法，使用注意力机制获取群组中每个用户对其他用户的注意力权重，为群组选出一个决策者，以此来模拟群组中用户的交互，再根据用户的加权偏好为群组推荐项目。通过在CAMRa2011和MovieLens1M数据集上与基线方法的对比可知，该方法在命中率和归一化折扣累计增益方面都有较大提高。

Research on group recommendation based on attention mechanism

Existing group recommendation methods mostly use pre-defined strategies in preference fusion. This static method ignores the interaction of users between groups, and it is difficult to model the complex decision-making process, thereby affecting the recommendation effect. In response to this problem, this paper proposes a group recommendation method based on attention mechanism. This method uses the attention mechanism to obtain the weight of each user in the group to other users, select a decision maker for the group to simulate the interaction of the users in the group, and then recommend the group according to the weighted preference of the user project. Compared with the baseline method on the CAMRa2011 and MovieLens1M data sets, this method has an improvement of more than 4% in terms of hit rate and normalized discount cumulative gain.