A Multilayer Approach to Multiplexity and Link Prediction in Online Geo-Social Networks

TL;DR

This paper introduces a multilayer framework for analyzing and predicting links in multiplex online social networks, demonstrating that cross-network features improve link prediction accuracy and offer new insights into user interactions.

Contribution

It presents a novel multilayer approach combining data from Twitter and Foursquare, enhancing link prediction and understanding of multiplex social interactions.

Findings

Multiplex links correlate with higher interaction and neighborhood overlap.

Multilayer features outperform single network features in link prediction.

Users with multiplex links visit closer locations and show greater similarity.

Abstract

Online social systems are multiplex in nature as multiple links may exist between the same two users across different social networks. In this work, we introduce a framework for studying links and interactions between users beyond the individual social network. Exploring the cross-section of two popular online platforms - Twitter and location-based social network Foursquare - we represent the two together as a composite multilayer online social network. Through this paradigm we study the interactions of pairs of users differentiating between those with links on one or both networks. We find that users with multiplex links, who are connected on both networks, interact more and have greater neighbourhood overlap on both platforms, in comparison with pairs who are connected on just one of the social networks. In particular, the most frequented locations of users are considerably closer, and similarity is considerably greater among multiplex links. We present a number of structural and interaction features, such as the multilayer Adamic/Adar coefficient, which are based on the extension of the concept of the node neighbourhood beyond the single network. Our evaluation, which aims to shed light on the implications of multiplexity for the link generation process, shows that multilayer features, constructed from properties across social networks, perform better than their single network counterparts in predicting links across networks. We propose that combining information from multiple networks in a multilayer configuration can provide new insights into user interactions on online social networks, and can significantly improve link prediction overall with valuable applications to social bootstrapping and friend recommendations.