Location-Based Correlation Estimation in Social Network via Collaborative Learning

In social network analysis, correlation estimation is a critical part for various applications. With the prevalence of location-based services, geographic information is incorporated as a new perspective to refer the interpersonal correlation. In this paper, we propose a novel multi-scale multi-feature collaborative learning model for robust location-based correlation estimation. Geographic attributes are explored from multiple scales, and in the meantime, depicted by multiple features. Using the observed interactions as labeled data and the unobserved ones with high predictive confidence as recommended unlabeled data, the global correlation can be estimated in a collaborative way.