Learning Holistic Interactions in LBSNs with High-order, Dynamic, and Multi-role Contexts

Location-based social networks (LBSNs) have emerged over the past few years. Their exponential network effects depend on fact that each user can share her daily digital footprints with different communities, in places, and at times (for example form of check-in activities). Unlike other types networks, activities an LBSN potentially be performed by several users a collaborative way. Existing studies representation learning for LBSNs often consider them as regular graphs ignore these high-order, dynamic, multi-role contexts, since their holistic interactions are quite difficult to capture. In this paper, we propose model which learned transferred into node embeddings derived from hypergraph persona decomposition process. More specifically, learns friendship edges, hyperedges, personas same time, devises multiple presentations reflects roles context. The embedding process also exploits useful patterns such co-location sequential through carefully designed point-of-interest splitting step. Extensive experiments real synthetic datasets show our outperforms alternative state-of-the-art methods location prediction tasks.