Task Selection for Bandit-Based Task Assignment in Heterogeneous Crowdsourcing

摘要

Task selection (picking an appropriate labeling task) and worker selection(assigning the labeling task to a suitable worker) are two major challenges intask assignment for crowdsourcing. Recently, worker selection has beensuccessfully addressed by the bandit-based task assignment (BBTA) method, whiletask selection has not been thoroughly investigated yet. In this paper, weexperimentally compare several task selection strategies borrowed from activelearning literature, and show that the least confidence strategy significantlyimproves the performance of task assignment in crowdsourcing.