Human computation is a long-existing concept and has been practiced for centuries. Specifically, whenever a human serves to compute, human computation is observed. This leads to a history of Human Computation even longer than that of electronic computer. Now with the development of Internet web service, the workforce of human computation is broadened to a vast pool of crowds, e.g. Amazon Mechanic Turk, instead of designated exerts or employees. This type of outsourcing to crowds, a.k.a. crowdsourcing, ushers in the new computation paradigm of Crowdsourced Human Computation. Data-driven applications also benefit from the crowdsourcing power, where the crowds are utilized as a data processing module.

However, traditional crowd-powered task processing relies on centralized platfo...[ Read more ]

Human computation is a long-existing concept and has been practiced for centuries. Specifically, whenever a human serves to compute, human computation is observed. This leads to a history of Human Computation even longer than that of electronic computer. Now with the development of Internet web service, the workforce of human computation is broadened to a vast pool of crowds, e.g. Amazon Mechanic Turk, instead of designated exerts or employees. This type of outsourcing to crowds, a.k.a. crowdsourcing, ushers in the new computation paradigm of Crowdsourced Human Computation. Data-driven applications also benefit from the crowdsourcing power, where the crowds are utilized as a data processing module.

However, traditional crowd-powered task processing relies on centralized platforms. These markets are specially designed based on a labor market structure, which facilitates the task display and post-task payoff. But such mechanism also constrains the source of crowd workforce, which leads to difficulties in terms of quality control, cost management, as well as bias of worker demographics.

In this thesis, we elaborate the effort of employing online social users as another source of crowdsourcing workforce. We show that a most of data-driven applications can be decomposed into binary decision making or information elicitation tasks for human workforce. Then we illustrate the majority voting over the decision making as crowdsourced answer aggregator and discuss its properties. Moreover, there are three major challenges to establish high-performance crowdsourcing applications onto online social users as crowdsourcing workforce; therefore we present corresponding techniques as follows:

Quality: Jury-selection algorithms to solve “Whom to Ask” challenge to improve answer quality under majority voting[3];

Cost : WiseMarket as a new crowdsourcing paradigm to conduct payment with less cost and higher quality[4];

Authenticity: COPE as an approach to elicit opinion from online crowds with authenticity guarantee and cost control[5].

In the end, we show directions of future work in applying the data-driven crowdsourcing via online social users.