Purpose: Selective crowdsourcing is an important type of crowdsourcing which has been popularly used in practice. However, because selective crowdsourcing uses a winner-takes-all mechanism, implying that the efforts of most participants except the final winner will be just in vain. The purpose of this paper is to explore why this costly mechanism can become a popularity during the past decade and which type of tasks can fit this mechanism well.Design/methodology/approach: The authors propose a game model between a sponsor and N participants. The sponsor is to determine its reward and the participants are to optimize their effort-spending strategy. In this model, each participant's ability is the private information, and thus, all roles in the system face incomplete information.Findings: The results of this paper demonstrate the following: whether the sponsor can obtain a positive expected payoff are determined by the type of tasks, while the complex tasks with a strong learning effect is more suitable to selective crowdsourcing, as for the other two types of task, the sponsor cannot obtain a positive payoff, or can just gain a rather low payoff; besides the task type, the sponsor's efficiency in using the solutions and the public's marginal cost also influence the result that whether the sponsor can obtain a positive surplus from the winner-takes-all mechanism.Originality/value: The model presented in this paper is innovative by containing the following characteristics. First, each participant's ability is private information, and thus, all roles in the system face incomplete information. Second, the winner-takes-all mechanism is used, implying that the sponsor's reward will be entirely given to the participant with the highest quality solution. Third, the sponsor's utility from the solutions, as well as the public's cost to complete the task, are both assumed as functions just satisfying general properties.
26 November 2019 in Journal of Computational Electronics
We develop a unified uncertainty propagation framework for computational models involving stochastic variability together with two distinct forms of epistemic uncertainty: vagueness and incomplete knowledge. In this framework, stochastic uncertainty is represented by random variables, while the epistemic components are modeled by random fuzzy sets with fuzzy mass functions generalized from Dempster-Shafer theory and fuzzy set theory. Using generalized polynomial chaos expansions and extension principles, the propagated uncertainty in output expectation is then quantified efficiently by a fuzzy mass function. We also introduce a distance measure between fuzzy mass functions and use it to analyze the error in the resulting numerical approximations. The proposed approach is demonstrated on several simple examples and applied to two biological systems: the Mitchell-Schaeffer model governed by a system of ordinary differential equations, and a computational model of olfaction governed by the Navier-Stokes equations.