Analysis of Large Multi-Modal Social Networks: Patterns and a Generator

Abstract

On-line social networking sites often involve multiple relations simultaneously. While people can build an explicit social network by adding each other as friends, they can also form several implicit social networks through their daily interactions like commenting on people’s posts, or tagging people’s photos. So given a real social networking system which changes over time, what can we say about people’s social behaviors ? Do their daily interactions follow any pattern ? The majority of earlier work mainly mimics the patterns and properties of a single type of network. Here, we model the formation and co-evolution of multi-modal networks emerging from different social relations such as ” who-adds-whom-as-friend ” and ” who-comments-on-whose-post ” simultaneously. The contributions are the following : (a) we propose a new approach called EigenNetwork Analysis for analyzing time-evolving networks, and use it to discover temporal patterns with people’s social interactions; (b) we report inherent correlation between friendship and co-occurrence in on-line settings; (c) we design the first multi-modal graph generator xSocial that is capable of producing multiple weighted time-evolving networks, which match most of the observed patterns so far. Our study was performed on two real datasets (Nokia FriendView and Flickr) with 100,000 and 50,000,000 records respectively, each of which corresponds to a different social service, and spans up to two years of activity.