Information fusion methods for the automatic creation of Twitter lists

Abstract

Many microblogging services provide tools that allow users to organise the people they follow into groups for easier information access and filtering. However, the uptake of these tools is low with a likely reason being that curating groups of followees is a time consuming task. This paper proposes methods for automatically clustering followees into groups so that users can use these groups as their user lists. As social microblogging services contain both textual content posted by users and directed followee relationships between users, members in the same list usually share common interests and/or have dense followee relationships. Under this assumption, this paper first applies separate content- and graph-based methods to cluster users. Next, we propose several novel information fusion configurations that combine textual and network features. We evaluate these approaches using both an offline evaluation and a user evaluation on datasets crawled using the Twitter API.