Abstract
Crowdsourcing is an increasingly popular approach for utilizing the power of the crowd in performing tasks that cannot be solved sufficiently by machines. Text annotation and image labeling are two examples of crowdsourcing tasks that are difficult to automate and human knowledge is often required. However, the quality of the obtained outcome from the crowdsourcing is still problematic. To obtain high-quality results, different quality control mechanisms should be applied to evaluate the different type of tasks. In a previous work, we present a task ontology-based model that can be utilized to identify which quality mechanism is most appropriate based on the task type. In this paper, we complement our previous work by providing a categorization of crowdsourcing tasks. That is, we define the most common task types in the crowdsourcing context. Then, we show how machine learning algorithms can be used to infer automatically the type of the crowdsourced task.