STGRN: A Spatial-Temporal Graph Resonance Network for Social Connection Inference

Social Connection Inference (SCI), as one of the most important social network analysis techniques, has been a popular research direction for decades. A lot of studies have been utilizing spatial-temporal co-occurrence events for mining latent social relations. The previous studies mainly look into the characteristics of the point-to-point meetup events. Moreover. some methods employ the social neighborhood of two target objects for better prediction. However, we have rarely seen research utilizing the network structure formed by meetup events for better SCI performance. In this paper, we propose a graph neural network framework, namely Spatial-Temporal Graph Resonance Network (STGRN), for social connection inference enhancement. Firstly, we calculate co-occurrence events' characteristics, including two novel features we designed: Ceo-Spread and ActM-Rate. Secondly, we build a network embedding based on meetup relations. Thirdly, we combine the meetup events, social network, and co-occurrence network for the final prediction. In the end, we conduct extensive experiments on public datasets to prove our method's effectiveness.