Twitter Session Analytics: Profiling Users’ Short-Term Behavioral Changes

Authors: Farshad Kooti, Esteban Moro, and Kristina Lerman
Journal: Proceedings of SocInfo 2016 LINK

Abstract:  Human behavior shows strong daily, weekly, and monthly patterns. In this work, we demonstrate online behavioral changes that occur on a much smaller time scale: minutes, rather than days or weeks. Specifically, we study how people distribute their effort over different tasks during periods of activity on the Twitter social platform. We demonstrate that later in a session on Twitter, people prefer to per- form simpler tasks, such as replying and retweeting others’ posts, rather than composing original messages, and they also tend to post shorter messages. We measure the strength of this effect empirically and statis- tically using mixed-effects models, and find that the first post of a session is up to 25% more likely to be a composed message, and 10–20% less likely to be a reply or retweet. Qualitatively, our results hold for different populations of Twitter users segmented by how active and well-connected they are. Although our work does not resolve the mechanisms responsi- ble for these behavioral changes, our results offer insights for improving user experience and engagement on online social platforms.