Passerby Crowdsourcing: Workers' Behavior and Data Quality Management.

Worker recruitment is one of the important problems in crowdsourcing, and many proposals have been presented for placing equipment in physical spaces for recruiting workers. One of the essential challenges of the approach is how to keep people attracted because those who perform tasks at first gradually lose interest and do not access the equipment. This study uses a different approach to the worker recruitment problem. In our approach, we dive into people's personal spaces by projecting task images on the floor, thereby allowing the passersby to effortlessly access tasks while walking. The problem then changes from how to keep people engaged to how to manage data quality because many passersby unconsciously or intentionally walk through the task screen on the floor without doing the task, which produces unintended results. We explore a machine-learning approach to select only the intended results and manage the data quality. The system assesses the workers' intention from their behavior. We identify the features for classifiers based on our observations of the passersby. We then conduct extensive evaluations with real data. The results show that the features are effective in practice, and the classifiers improve the data quality.