Inferring individual physical locations with social friendships

Abstract

Physical location is an important characteristic for digital individuals, as it is widely used in location based services, such as navigation, advertisements, and recommendations. This paper focuses on the problem of inferring individual physical locations from their friendships in a social network. We represent individual locations with a few high frequency places to eliminate the noise influence. By using of interactions between users, a spatial based inferring model is developed to directly estimate individual physical locations. The spatial weighted clustering method is used by considering the structure of interactions between friends. Data from Tencent, the biggest social network service provider in China, is used to conduct an experiment to validate the performance of the proposed inferring framework. Results indicate the framework can predict individual locations within 15 km in distance error with the accuracy of 68%.