Learning Linguistic Descriptors of User Roles in Online Communities

Abstract

Understanding the ways in which users interact with different online communities is crucial to social network analysis and community maintenance. We present an unsupervised neural model to learn linguistic descriptors for a user's behavior over time within an online community. We show that the descriptors learned by our model capture the functional roles that users occupy in communities, in contrast to those learned via a standard topic-modeling algorithm, which simply reflect topical content. Experiments on the social media forum Reddit show how the model can provide interpretable insights into user behavior. Our model uncovers linguistic differences that correlate with user activity levels and community clustering.