Towards unsupervised home location inference from online social media

Abstract

Users' home location is important information for many advanced information services in big data applications (e.g., localized recommendation, target ads of local business and urban planning). In this paper, we study the problem of accurately inferring the home locations of people from the noisy and sparse data they voluntarily share on online social media. Previous studies have developed supervised learning approaches to predict a person's home location in a city. However, the accuracy of these techniques largely depends on a high quality training dataset, which is difficult and expensive to obtain in practice. In this study, we propose a new analytical framework, Unsupervised Home Location Inference (UHLI), to accurately infer the home locations of people using a set of principle approaches. In particular, the UHLI scheme addresses the critical challenges of using sparse and noisy online social media data and derives an optimal solution to the home location inference problem. We evaluated the performance of our scheme and compared it to the state-of-the-art baselines using three real world data traces collected from Foursquare. The results showed that our scheme can accurately infer the home location of people and significantly outperform the state-of-the-art baselines.