Realizing Impact Sourcing by Adaptive Gold Questions: A Socially Responsible Measure for Workers’ Trustworthiness

TitleRealizing Impact Sourcing by Adaptive Gold Questions: A Socially Responsible Measure for Workers’ Trustworthiness
Publication TypeConference Paper
Year of Publication2015
AuthorsMaarry, K. E., U. Güntzer, and W. - T. Balke
Conference Name16th International Conference on Web-Age Information Management (WAIM)
Date Published06/2015
Conference LocationQingdao, Shandong, China

In recent years, crowd sourcing has emerged as a good solution for digitizing voluminous tasks. What’s more, it offers a social solution promising to extend economic opportunities to low-income countries, alleviating the welfare of poor, honest and yet uneducated labor. On the other hand, crowd sourcing’s virtual nature and anonymity encourages fraudulent workers to misuse the service for quick and easy monetary gain. This in turn compromises the quality of results, and forces task providers to employ strict control measures like gold questions or majority voting, which may gravely misjudge honest workers with lower skills, ultimately discarding them from the labor pool. Thus, the problem of fairly distinguishing between fraudulent and honest workers lacking educational skills be-comes vital for supporting the vision of Impact Sourcing and its social responsibility. We develop a technique with socially responsible gold questions as an objective measure of workers’ trustworthiness, rather than a mere discarding mechanism. Our statistical model aligns workers’ skill levels and questions’ difficulty levels, which then allows adapting the gold questions’ difficulty for a fair judgment. Moreover, we illustrate how low-skilled workers’ initial payloads, which are usually discarded along with the worker, can be partially recovered for an in-creased economic gain, and show how low-skilled workers can be seamlessly integrated into high-performing teams. Our experiments prove that about 75% of misjudged workers can be correctly identified and effectively be integrated into teams with high overall result correctness between 70-95%.

WAIM_141.pdf922.09 KB