User classification in online communities

In this thesis, we explore the identification of expert users in online communities. We are interested in modeling user behavior and providing deeper insights into how a user evolves and interacts in a given community. Our goal in this research is to help community administrators in identifying top contributors, and so help them in retaining these users and sustaining the communities. We have focused on three key aspects of user identification. First, we have identified and modeled several key attributes that can be useful for characterizing user behavior in these communities and how these attributes impact the perceptions of online users. Secondly, We have developed models to identify the top users, also called as experts or authorities, in the communities. We show that along with top users, we can also identify users who are expected to become top users in future. We also show how bias due to name value impacts the perception of quality of Twitter authors. Third, We have examined how users evolve over time to understand different kinds of phases that users undergo. The results of this work show that the top users in the community can be distinguished by their activity patterns. These users can be distinguished: (i) efficiently, (ii) accurately, and (iii) early on. We show that expert users evolve in three distinctive patterns: (a) consistently active pattern, (b) initially active but later passive pattern, (c) initially passive but later active pattern. The practical applications of their identification range from: (a) estimation of bias while preparing a reading list for a user, (b) recommending similar users to each other, and (c) providing training to the users and improving their participation in these communities.