RESCOT: Reliable Scheduling of Social Computing Tasks

Scalable processing of Human Computing tasks by large pools of community members is one of the key opportunities of the modern Social Web. However, the self-organizing nature of Web communities does not allow for guarantees of user commitment and output quality. This is a serious drawback, especially for applications with continuously arriving input data (e.g. reviews and assessments in social streams). On the other hand, greedy replication of all tasks results in a high demand for human resources and - in case of monetary models - in rapidly growing costs. This paper aims to exploit the tradeoff between redundancy of task assignments and guarantees of successful task completion in the context of human computing. We introduce the probabilistic predictive model that allows for flexible tuning of induced redundancy of assignments, with respect to the estimated failure rates of invoked users and the desired global quality of service.