JAM: Joint attention model for next event recommendation in event-based social networks

Abstract

In recent years, event-based social networks (EBSNs) have emerged as popular applications for people selecting social events such as concerts, hikes, and technical talks, and how to precisely promote events to specific users has become an essential topic in both academia and industry. However, the next event recommendation in EBSNs is facing new challenges compared to general recommender systems — events are usually new and have a short life cycle, and user preferences may change over time. This study proposes a next-event recommendation method called the joint attention model (JAM), where rich contextual information is integrated for event representation, and attention mechanism is employed to handle the dynamic preferences of users. In particular, the participants are also considered as context of the events. To capture the dynamic changing preferences of users, this study develops a signed multihead attention mechanism, which assigns positive and negative weights to historical events visited by users in the past and uses higher-order attention to simulate the weights of positive and negative effects. Empirical experiments are conducted with different datasets from Meetup, and the results show that the proposed model achieves better performance than the state-of-the-art methods.