Aspect-oriented unsupervised social link inference on user trajectory data

Abstract

We focus on representation learning for social link inference based on user trajectory data. While traditional methods predict relations between users by considering hand-crafted features, recent studies first perform representation learning using network embedding or graph neural networks (GNNs) for downstream tasks such as node classification and link prediction. However, those approaches fail to capture behavioral patterns of individuals ingrained in periodical visits or long-distance movements. To better learn behavioral patterns, this paper proposes a novel scheme called ANES (Aspect-oriented Network Embedding for Social link inference). ANES learns aspect-oriented relations between users and Point-of-Interests (POIs) within their contexts. ANES is the first approach that extracts the complex behavioral pattern of users from both trajectory data and the structure of User-POI bipartite graphs. Our extensive experiments on several real-world datasets show that ANES outperforms state-of-the-art baselines.