Leveraging Mixed Expertise in Crowdsourcing.

Abstract

Crowdsourcing systems promise to leverage the "wisdom of crowds" to help solve many kinds of problems that are difficult to solve using only computers. Although a crowd of people inherently represents a diversity of skill levels, knowledge, and opinions, crowdsourcing system designers typically view this diversity as noise and effectively cancel it out by aggregating responses. However, we believe that by embracing crowd workers' diverse expertise levels, system designers can better leverage that knowledge to increase the wisdom of crowds. In this thesis, we propose solutions to a limitation of current crowdsourcing approaches: not accounting for a range of expertise levels in the crowd. The current body of work in crowdsourcing does not systematically examine this, suggesting that researchers may not believe the benefits of using mixed expertise warrants the complexities of supporting it. This thesis presents two systems, Escalier and Kurator, to show that leveraging mixed expertise is a worthwhile endeavor because it materially benefits system performance, at scale, for various types of problems. We also demonstrate an effective technique, called expertise layering, to incorporate mixed expertise into crowdsourcing systems. Finally, we show that leveraging mixed expertise enables researchers to use crowdsourcing to address new types of problems.