Intra- and inter-association attention network-enhanced policy learning for social group recommendation

Social Group Recommendation (SGR) is a critical task to recommend items to a group of users in social network platforms, such as Meetup, Douban, Mofengwo, etc. Recently, many state-of-the-art works have addressed the group decision making with pre-defined aggregation strategies or neural-based methods. The main challenge is how to capture the intra-interaction and inter-association among users, groups, and items. In term of this issue, we propose an Intra- and inter-association attention network with Policy learning for Social Group Recommendation (IP-SGR). Specifically, for intra-interaction attention model, we capture the preference of user pair agreement with the representation of their co-interaction items, while a gate filtering component is utilized to aggregate the group agreement with the member representations of the group. In addition, to capture the inter-association representation of groups and items, we present inter-group attention network and inter-item prototype learning model, respectively. Finally, we propose a reinforcement learning-based model to obtain the positive and negative reward for social group recommendation. Extensive experiments on three real-world datasets demonstrate our proposed IP-SGR model significantly outperforms several state-of-the-art methods in terms of HR and NDCG.