Social activity matching with graph neural network in event-based social networks

In recent years, event-based social networks (EBSNs) have become increasingly popular. Different from traditional online social networks, EBSNs consist of valuable online and offline social interactions, which can bring users more experience and entertainment. One of the crucial tasks for the EBSN platforms is to match users and social activities to help users participate in suitable activities. However, the existing matching methods either do not consider the influence of adjacent activities and users and the processing of newly added activities, or ignore the actual attribute constraints, e.g., budget and capacity. To address the limitations, we propose a novel graph neural network-based social activity matching method. Specifically, we model the historical records with a heterogeneous graph, and connect any new activity node to the user node who is the organizer. We then design a neural network-based affinity calculation model to predict the affinities between users and new activities. Moreover, we use a greedy-based heuristic method for social activity matching, considering the bilateral constraints extracted from the user and the activity attributes. Extensive experiments on three real event-based social service datasets show the effectiveness of the proposed method, which outperforms the state-of-the-art baselines in terms of affinity prediction and social activity matching.