[1K3-J-4] Knowledge utilization and sharing 2

[1K3-J-4-02] TurkScanner: Predicting the Hourly Wage of Previously Unseen Microtasks

〇Susumu Saito¹, Chun-Wei Chiang², Saiph Savage², Teppei Nakano¹, Tetsunori Kobayashi¹, Jeffrey Bigham³ (1. Waseda University, 2. West Virginia University, 3. Carnegie Mellon University)

Keywords:Crowdsourcing, Amazon Mechanical Turk

In crowd markets, workers struggle to earn adequate wages by accurately gauging hourly wage of microtasks that they have not completed before. This paper explores how we might be able to predict the necessary working time (and thus hourly wage) of a previously unseen task based on data collected from prior workers completing other tasks. We collected 9,155 data records using a web browser extension, installed by 84 Amazon Mechanical Turk workers, and explore the challenge of accurately recording working time both automatically and by asking workers.
Our predictive model, TurkScanner, was created using ~150 derived features and was able to predict working time with high accuracy. Future directions include observing its effects on work practices, adapting it to a requester tool for better price setting, and predicting other elements of work (e.g., acceptance likelihood, worker task preference, etc.)