Collective Geographical Embedding for Geolocating Social Network Users

Abstract

Inferring the physical locations of social network users is one of the core tasks in many online services, such as targeted advertisement, recommending local events, and urban computing. In this paper, we introduce the Collective Geographical Embedding (CGE) algorithm to embed multiple information sources into a low dimensional space, such that the distance in the embedding space reflects the physical distance in the real world. To achieve this, we introduced an embedding method with a location affinity matrix as a constraint for heterogeneous user network. The experiments demonstrate that the proposed algorithm not only outperforms traditional user geolocation prediction algorithms by collectively extracting relations hidden in the heterogeneous user network, but also outperforms state-of-the-art embedding algorithms by appropriately casting geographical information of check-in.