Point-of-interest recommendations by unifying multiple correlations

Abstract

In recent years, we have witnessed the development of location-based services which benefit users and businesses. This paper aims to provide a unified framework for location-aware recommender systems with the consideration of social influence, categorical influence and geographical influence for users’ preference. In the framework, we model the three types of information as functions following a powerlaw distribution, respectively. And then we unify different information in a framework and learn the exact function by using gradient descent methods. The experimental results on real-world data sets show that our recommendations are more effective than baseline methods.