Mining User Lifecycles from Online Community Platforms and their Application to Churn Prediction

Recent work has studied user development in the domains of both telecommunication and online community platforms, examining how users develop in terms of the company they keep (socially) [1] and the language they use (lexically) [2]. Such works afford key insights into user changes along individual dimensions, yet they do not examine how users develop relative to their prior behaviour along multiple dimensions. In this paper we examine how users develop along various properties (in-degree, out-degree, posted terms) in three online community platforms (Facebook, SAP Community Network, and Server Fault) and using three models of user development: (i) isolated lifecycle periods, (ii) historical contrasts, and (iii) community contrasts. We present an approach to mine the lifecycle trajectories of users as a means to characterise user development along the different properties and development models, and demonstrate the utility of such trajectories in predicting churners. We find consistent effects with past work: users tend to reflect the behaviour of the community in early portions of their lifecycles, before then diverging from the community towards the end. We also find that users form sub-communities with whom they communicate and remain within.