Individual and Collective User Behavior in Crowdsourcing Services

Dominic DiPalantino, Thomas Karagiannis, and Milan Vojnovic

Abstract

We study profiles of user behavior in selecting tasks and the resulting profits in online crowdsourcing services. Specifically, we focus on (1) understanding the individual user behavior as well as the underlying collective behavior, (2) understanding the effects of competition among users on the resulting profits and (3) the evolution of user behavior with experience. Our analysis is based on data from a popular crowdsourcing service covering thousands of workers and jobs posted over a period of more than a year.

We found two distinct characteristics when looking at individual worker behavior versus the collective behavior of the community. On the one hand, we show evidence of a market segmentation where individual workers tend to direct their effort to tasks from specific ranges of rewards, reflecting their level of skill. On the other hand, the contribution of the community as a whole spreads more evenly across the entire range of rewards and this contribution exhibits a diminishing increase with the value of the reward. Furthermore, we found significant correlations between different measures of competition among users and the resulting profit and characterized how the user's performance improves with experience.

Our results would provide valuable insights to the designers of crowdsourcing services and may inform the design of novel features such as task recommendation based on user skill.

Details

Publication typeTechReport
NumberMSR-TR-2010-59
PublisherMicrosoft Research
> Publications > Individual and Collective User Behavior in Crowdsourcing Services