Abstract (EN)

Crowdsourcing has recently gained momentum as a means for outsourcing tasks to human workers with no specific expertise. This is a viable option for tasks that are difficult to achieve correctly using machines or that are too expensive to be carried out by expert professionals. This leads to the possibility of addressing large campaigns of jobs without facing extremely high cost. While this technique has proven to be effective for simple labeling tasks, such as text recognition, the quality of results it delivers for relatively sophisticated tasks such as scientific paper translation, may appear as deteriorated. The objective of our work is instead to define in a rich way the concept of community of workers, defining a way to properly match crowdsourcing tasks to the communities based on expertise of workers and field/topic of tasks. We describe a set of conceptual models for queries, communities and workers, a high level architecture of the approach, and outline a set of possible strategies for addressing various aspects of expert-targeted crowdsourcing.