Abstract:Social networks, with tremendous popularity all over the world, have become the most important platform for many services in the past years. Location, as part of users' basic information, is always the key to many recommendation services in social networks. Most of the previous research works focus on inferring on the users' home locations. However, it is not enough as many people in social networks have multiple location tags, including home location, work location and on. In this paper, we propose a multiple location tags inference algorithm, i.e. MuLTI to build complete location profiles for users in social networks. We formulate the correlations between the users' location tags and their friendships, tweets, and then infer the users' locations in each of their friendships and tweets. It reflects the activity level of users to be in different locations. Apart from the activity level, we also consider the time span of users to be in different locations, so as to infer the users' long-term location tags better, as we find that users may also be active in their temporal locations. Experiments show that MuLTI improves the precision by about 15%, and the recall by about 25% compared with the state-of-the-art algorithms.

MuLTI: Multiple Location Tags Inference for Users in Social Networks