| Summary: | 碩士 === 國立臺灣大學 === 電信工程學研究所 === 104 === Crowdsourcing has been an emerging technology in social and network science. Existing researches, however, often neglect the underlying tasks selection process of crowd workers, and how it effects the efficiency and tasks quality of a crowdsourcing system. In this thesis, we consider and formulate this fundamental phenomenon as random walks over graphs; Hence, the efficiency and quality are connected to the mean cover time and occupation measure on graphs respectively. We propose the optimal tasks management and recommendation problem, which could be viewed as a resource allocation problem over graphs, and aim at reducing the mean cover time while satisfying some quality constraints {it via} adding edges efficiently among tasks. Exploiting graph spectrum and submodular set function optimization, elegant and computationally efficient methodologies are proposed to implement such systems, together with illustrative verification from numerical results and data of real crowdsourcing systems. This methodology could be applied to numerous applications in other fields, including network searching and navigation, resource harvesting, sensor distributing problems, etc..
|
|---|
