CRII: SaTC: Toward Secure, Privacy-Preserving, and Efficient Crowdsourcing Systems

University Of Tulsa, Tulsa OK

Investigators

Abstract

Researchers and the industry have widely used crowdsourcing systems in various disciplines for large-scale data collection and analysis to conduct user studies, improve machine learning performance, and accelerate product iteration. However, today's crowdsourcing systems cannot adequately support requesters and workers due to three main problems: inadequate quality or integrity of data, privacy violations, and low efficiency of task completion. This project aims to address these problems to enhance security, privacy, and efficiency in crowdsourcing systems. The project's novelties are developing an advanced quality control approach to increase data quality and integrity, a novel scheme to detect and prevent privacy violations, and an intelligent framework to improve efficiency. The project's broader significance includes: (1) addressing the critical challenges of building secure, privacy-preserving, and efficient crowdsourcing systems, (2) supporting crowd workers, including protecting their privacy and assisting them to work in a cost-efficient way, and (3) assisting job requesters in obtaining high-quality data that they can rely on to complete their important studies and make important decisions confidently. The project also contributes to increasing undergraduate involvement in research, including integrating research outcomes into the curriculum and mentoring undergraduate students in conducting research. The project enhances security, privacy, and efficiency in crowdsourcing systems. First, the project develops a subtask-aware and robust quality control approach to ensure data quality and integrity. Second, the project investigates privacy risks to workers and third parties in real-world crowdsourcing tasks and designs machine learning-based schemes to detect and prevent privacy violations. Third, the project designs and implements a framework based on artificial intelligence techniques to assist workers in identifying and prioritizing crowdsourcing tasks more efficiently. This project also integrates the abovementioned schemes into a client-side browser extension that workers can use and a server-side model that can be applied in crowdsourcing systems for large-scale deployment. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.