Location based Social Network analysis using Tensors and Signal Processing tools

Abstract

With the rise of online social networks and smartphones that record the user's location, a new type of online social network has gained popularity during the last few years, the so called Location-based Social Networks (LBSNs). In such networks, users voluntarily share their location with their friends via a “check-in”. In exchange they get recommendations tailored to their particular location as well as special deals that businesses offer when users check-in frequently. LBSNs started as specialized platforms such as Gowalla and Foursquare, however their immense popularity has led online social networking giants like Facebook to adopt this functionality. The spatial aspect of LBSNs directly ties the physical with the online world, creating a very rich ecosystem where users interact with their friends both online as well as declare their physical (co-)presence in various locations. Such a rich environment calls for novel analytic tools that can model the aforementioned types of interactions. In this work, we propose to model and analyze LBSNs using Tensors and Tensor Decompositions, powerful analytical tools that have enjoyed great growth and success in fields like Machine Learning, Data Mining, and Signal Processing alike. By doing so, we identify tightly knit, hidden communities of users and locations which they frequent. In addition to Tensor Decompositions, we use Signal Processing tools that have been previously used in Direction of Arrival (DOA) estimations, in order to study the temporal dynamics of hidden communities in LBSNs.