Attention Network for Group Recommendation

摘要

Group activities have become important in the current social networks. Recommending activities for a group has become an important task in many information systems. Group recommendation not only needs to aggregate the preferences of group members, but also needs to properly understand the group decision-making process. In addition to integrating the individual preferences of each member of the group, we also need to consider the interactive information among users, groups and activities, so as to assign a higher influence weight to the experts or leaders in the group. Based on the attention mechanism and the latest development of NCF, we contribute a novel solution, namely NCF-AGREE, which can learn aggregation strategies according to historical data, so as to better solve the problem of group preference aggregation. In order to adapt to group recommendation, different members of the group are assigned weights. Experiments on real data sets show that NCF-AGREE is better than baseline method in group recommendation.