Modeling and Exploration of Crowdsourcing Micro-Tasks Execution

Abstract

There are two key issues which are critical for a person crowdsourcing work: quality control and timeliness. While the first topic is extensively covered in the literature, the timeliness is still not very well analyzed. In our past experiments on CrowdFlower we noticed that 90% of task instances were completed in only 10 minutes, the rest 10% could take hours to be completed. There were also extra peaks in performance on the 30th and 60th minutes since the task publication. From the observation of these patterns we wanted to understand better how the crowd workers execute tasks, to try to describe this behavior mathematically and to identify ways to speed up the tasks execution.