Sentiment as a Predictor of Wikipedia Editor Activity
Wikipedia, the worlds largest encyclopedia, is created by millions of unpaid editors online. Every user can edit every article, and the project is protected against vandalism and low-quality contributions only through version control and a system of (again unpaid) reviewers. Somewhat hidden to most casual readers of the encyclopedia, Wikipedia also features a simple social network: every user has a personal user profile and a “user talk page” which acts as a publicly accessible guestbook where users can leave messages to each other. The messages exchanged in user talk pages are often related to a user’s editing behavior. For example, senior users may welcome new users, or congratulate them on their first edits. Administrators may officially warn culprits after transgressions of Wikipedias content guidelines or policies. Users may also thank one another for certain edits, and, of course, users engage in heated debates about what the ground truth reflected in a certain article should be. Not all such debates are pleasant, although the community as a whole has been noted for its considerable resilience against both anarchy and uncontrolled aggression [1]–[3]. Social feedback has long been known to be a strong influencer of intrinsic motivation [4], [5]. Observing praise and gratitude may be a strong incentive for Wikipedia editors to “keep up the good work,” whereas repeated unpleasant discussions, official warnings, or even personal insults may discourage further editing behavior. With this intuition in mind, we formulated our hypothesis: we ask if received message sentiment can help predict editor activity on Wikipedia. In so doing, we create the opportunity to engage with frustrated editors—for example, motivating emails could be sent to users who are expected to significantly reduce their * We thank the Wikimedia Foundation, and in particular Leila Zia of the Research and Data team, for their generous help and support of our project. This is a course project for Andrew Ng’s course CS 229 and (by approval of both teaching teams) for Sharad Goel’s course MS&E 331 at Stanford University. We are grateful to both courses for an inspiring quarter. 1 sergiomo@stanford.edu 2 deemeng@stanford.edu 3 roemheld@stanford.edu editing due to received message sentiment. If effective, this could increase overall editing activity (and editor happiness) on Wikipedia; for the scope of this paper we assume a high number of edits to be desirable, since it enables the encyclopedia to better reflect an everchanging world.
Research Paper Link: Download Paper