Assigning heterogeneous tasks to workers is an
important challenge of crowdsourcing platforms. Current approaches
to task assignment have primarily focused on contentbased
approaches, qualifications, or work history. We propose an
alternative and complementary approach that focuses on what
capabilities workers employ to perform tasks. First, we model
various tasks according to the human capabilities required to
perform them. Second, we capture the capability traces of the
crowd workers performance on existing tasks. Third, we predict
performance of workers on new tasks to make task routing
decisions, with the help of capability traces. We evaluate the effectiveness
of our approach on three different tasks including fact
verification, image comparison, and information extraction. The
results demonstrate that we can predict worker’s performance
based on worker capabilities. We also highlight limitations and
extensions of the proposed approach.