A self-attention model with contrastive learning for online group recommendation in event-based social networks

Recently, there has been a surge in the popularity of online groups on event-based social networks (EBSNs) like Meetup and Douban Event. These groups cater to individuals who share common interests, provide comments, and engage in various activities. Our research focuses on online group recommendations, based on which users can conveniently join groups and participate in offline events organized by the groups. Traditional group recommendation methods do not work well in addressing this problem because they lack the ability to deal with the challenges posed by dynamic user interests, sparse supervision signals, and heterogeneous networks simultaneously. The self-attention model with contrastive learning for online group recommendation (SCL4GR) presented in this study exploits user-group sequential data, online and offline networks in a unified framework to predict user preferences for groups. First, a graph encoder is used to capture the high-order social interaction between users. Then, the pattern of dynamic interests is captured by sequence model Transformer. Furthermore, the contrastive learning is employed to derive self-supervision signals from both online and offline networks. We conduct experiments on three real-world datasets. Experimental results show that our SCL4GR consistently outperforms state-of-the-art methods for online group recommendation in EBSNs.