Who Should I Follow? Recommending People in Directed Social Networks

Abstract

A variety of social networks feature a directed attention or “follower” network. In this paper, we compare several methods of recommending new people for users to follow. We analyzed structural patterns in a directed social network to evaluate the likelihood that they will predict a future connection, and use these observations to inform an intervention experiment where we offer users of this network new people to connect to. This paper compares a variety of features for recommending users and presents design implications for social networking services. Certain types of structural closures significantly outperform recommendations based on traditional collaborative filtering, behavioral, and similarity features. We find that sharing an audience with someone is a surprisingly compelling reason to follow them, and that similarity is much less persuasive. We also find evidence that organic network growth is very different from how users behave when they are prompted to connect to new people.