Privacy-Preserving Crowdsourcing
洸 梶野 · Institutional Repositories DataBase (IRDB) · 2016
Crowdsourcing is an idea in which requesters outsource tasks to unspecified workers via the Web.A basic procedure can be described using the following three steps.In the assignment step, a crowdsourcing platform matches tasks and workers, employing either a push-type or pull-type assignment strategy.The push-type assignment strategy uses the platform to assign tasks to appropriate workers based on the features of the workers and tasks (e.g., skills, preferences of tasks, and minimum wages), while the pull-type assignment requires workers to choose tasks they like.In the request step, each requester sends a job instruction and task instances (e.g., audio files in case of an audio transcription task) to the allocated workers.Finally, in the delivery step, each worker, having processed the assigned task, sends the results back to the requester.Crowdsourcing provides requesters with easy access to a huge pool of workers and enables workers to work much more flexibly than in the traditional labor market.These unique advantages have led to a number of real applications and businesses as well as new research opportunities such as human computation.Despite its revolutionary power, it is often pointed out that using crowdsourcing entails several risks including the risk of poor quality task results.Among others, this thesis focuses on the privacy risks.Although the privacy risks in crowdsourcing have been pointed out in diverse domains, little has been investigated until now.Toward establishing a research basis for privacy-preserving crowdsourcing, this thesis addresses the following two research questions:• What types of privacy risks are present in crowdsourcing?• How can we measure and control the privacy risks in crowdsourcing?To answer the first research question, we carefully examine the three steps of crowdsourcing and discover that four types of data can lead to privacy issues: features (the assignment step), job instruction and task instances (the request step), and task results (the delivery step).Further, by analyzing the applicability of existing privacy preservation strategies, we find that some I would like to express my great gratitude to both of my PhD advisers, Professor