Phans, Stans and Cishets: Self-Presentation Effects on Content Propagation in Tumblr

Abstract

Research on content propagation in social media has largely focused on features from the content of posts and the network structure of users. However, social media platforms are also spaces where users present their identities in particular ways. How do the ways users present themselves affect how content they produce is propagated? In this paper, we address this question with an empirical study of interaction and self-presentation data from Tumblr. We use a pairwise learning-to-rank framework to predict whether a given user will reblog (share) another user's post from features comparing self-presented textual and visual identity information. We find evidence that alignment in identity presentation is associated with content propagation, as these features increase performance over a baseline of content features. Interpreting learned feature weights on self-presented text identity labels, we find that users who present labels that match or indicate shared interests and values are generally more likely to propagate each other's content.