Simulating Wikipedia: Modeling peer production as networked social learning

Wikipedia is one of the most successful examples of how large-scale volunteer groups can self-organize to achieve a common goal. Editors work to maximize the quality of articles, subject to the consensus process which governs the project. Incremental improvements are common, but larger restructuring is sometimes necessary, suggesting that the quality of Wikipedia articles can be modeled as a rugged landscape, having many local maxima. Social learning literature has yielded important insights about how networked groups of agents can collaborate to maximize a rugged objective function, but has traditionally made assumptions incompatible with common collaborative settings such as Wikipedia.

This work evaluates the effectiveness of social learning using a novel network construction and consensus-based learning algorithm, both inspired by Wikipedia's collaborative process. The consensus-based algorithm outperforms traditional algorithms in both speed and quality, in part due to randomization which occurs as consensus is formed. The consensus algorithm also displays a dependence on network structure which is not seen in other methods, but which is similar to behavior observed in empirical observations of Wikipedia. This work aims to help bridge the gap between empirical studies of collaboration and simulation studies of networked social learning. A better understanding of social learning in more realistic settings could be used to better understand the performance of collaborative teams, establish best practices, and inform the design of online collaborative platforms.

PLATT_EDWARD_07281708.gif

Speaker: Edward L. Platt is a third-year PhD student at the University of Michigan School of Information. He studies collaboration in self-organized networked groups. Formerly, he worked as a staff researcher at the MIT Center for Civic Media, and as a freelance web developer and civic technologist. He also co-founded the i3 Detroit hackerspace and maintains the free/open-source Seltzer CRM. He holds bachelor's degrees in physics and computer science from MIT, and a master's degree in applied math from the University of Waterloo.

Information Analysis and Retrieval (IAR) seminars are held approximately weekly throughout the semester. Find out more by joining the IAR Seminar Group on MCommunity.