Socially-driven multi-interaction attentive group representation learning for group recommendation

Group recommendation has attracted much attention since group activities information has become increasing available in many online applications. A fundamental challenge in group recommendation is how to aggregate individuals' preferences to infer the decision of a group. However, most existing group representation methods do not take into account the static and dynamic preferences of groups synchronously, leading to the suboptimal group recommendation performance. In this work, we propose a socially-driven multi-interaction group representation approach to learn static and dynamic group preference coherently. Specifically, we inject the social homophily and social influence into capturing static and dynamic preference of a group. Furthermore, we explore latent user-item and group-item multiple interactions with bipartite graphs for group representation. Extensive experimental results on two real-world datasets verify the effectiveness of our proposed approach. (c) 2021 Elsevier B.V. All rights reserved.

Automated summary

Group recommendation is a popular research field that focuses on aggregating individual preferences to infer group decisions. Existing methods fail to consider both static and dynamic preferences of groups simultaneously, and a socially-driven multi-interaction group representation method is proposed, exploring latent user-item and group-item multiple interactions with bipartite graphs. Extensive experimental results on two real-world datasets validate the effectiveness of the proposed approach.