Social Computing and User Generated Content: A Game-theoretic Approach



Arpita Ghosh

Monday, November 12, 2012
4:00pm 5130 Upson Hall



Social computing is now ubiquitous on the Web, with user-generated contributions on sites like Amazon and Yelp, Q&A forums like Y! Answers or StackOverflow, blogs and YouTube forming a growing fraction of the content consumed by Web users. But while there is a large amount of user-generated content online, not all of it is of the same quality. What can we understand, using an economic approach, about what incentive schemes elicit high quality contributions, as well as adequate participation, in such systems?

We first provide a game-theoretic model within which the design and performance of mechanisms for incentivizing high quality UGC can be analyzed. Our model consists of strategic contributors motivated by exposure, and has the feature that the quality {\em as well as} the number of contributions are endogenously determined in a free-entry Nash equilibrium--- it is crucial in UGC to not just incentivize high quality, but also to encourage the production of content. An ideal mechanism in this context should be easily implementable in practice, and elicit both high quality and high participation in equilibrium, with near-optimal quality as the available attention diverges (corresponding to large viewership). We first demonstrate that a very simple elimination mechanism can achieve quality that tends to optimal, along with diverging participation, as the number of viewers diverges. Next we analyze equilibria in the widely used {\em rank-order} mechanism, where contributions are allocated positions on the page in decreasing order of their ratings, and show that the rank-order mechanism also elicits high quality contributions--- the {\em lowest} quality that can arise in a mixed strategy equilibrium of the rank-order mechanism becomes optimal as the amount of available attention diverges. Finally, we compare the rank-order mechanism against the more equitable proportional mechanism, which distributes attention in proportion to the number of positive ratings--- here we show that the rank-order mechanism, while less `fair', almost always incentivizes higher quality contributions in equilibrium than the proportional mechanism.

Based on joint work with Preston McAfee (WWW 2011) and Patrick Hummel (EC 2011).