Graph embedding based real-time social event matching for EBSNs recommendation

Event-based social networks (EBSNs) are platforms for users to publish, organize, or choose to participate in events through social networks. In recent years, the number of users and events on EBSNs has increased dramatically, and interactions among them have become more complicated. This makes it difficult to model EBSN networks for effective analysis and mining. Moreover, the requirement of real-time matching between users and impromptu events according to impromptu spatio-temporal constraints is becoming urgent because of the significant dynamics brought by the widespread use of mobile devices on EBSNs. Therefore, graph embedding based real-time social event matching technologies for EBSNs are studied in this paper. We first model an EBSN as a heterogeneous information network, and perform graph embedding to represent the nodes in it, which can more effectively reflect the hidden features of nodes and contribute to mining users and events preferences. Then heuristic social event matching methods are employed to effectively find overall optimal recommendations between users and events under spatio-temporal constraints in real-time. This forms a two-stage framework, i.e., the representation learning stage and the real-time matching stage. We conducted experiments on the Meetup dataset to verify the effectiveness of the framework by combining different graph embedding methods and heuristic matching algorithms. The results show that the proposed framework yields improvements in the matching success rate, user satisfaction, and user waiting time.