Using Co-Following for Personalized Out-of-Context Twitter Friend Recommendation

Abstract

We present two demos that give personalized `"out-of-context" recommendations of Twitter users to follow. By out-of-context we mean that a user wants to receive recommendation on, say, musicians to follow even though the user's tweets' contents and social links have no connection to the "context" of music. In this setting, where a user has never expressed interest in the context of music, many existing methods fail. Our approach exploits co-following information and hidden correlations where, say, a user's political preference might actually provide clues about their likely music preference. We implement this framework in two very distinct settings: one for recommending musicians and one for recommending political parties in Tunisia. Our framework is simple and similar to Amazon's "users who bought X also bought Y" and can be used not only for explainable out-of-context recommendations but also for social studies on, say, which music is "closest" to users of a particular political affiliation. It also helps to introduce and to "link" a user to an unknown domain, say, politics in Tunisia.