Analysis And Research For The Behavior Model Of Online Workers In Crowdsourcing Platforms

Crowdsourcing is a new kind of business that outsources tasks to a large amount ofglobally distributed workers over Internet.It has advantages in cost and human resources.With the extensive popularity and the rapid development of Internet,now crowdsourcing has been widely used in politics,economy,production and life,but quality and schedule control are the cruxes to determine if a crowdsourcing project succeeds.Different configurations of tasks result in different quality and completion time.In order to design crowdsourcing tasks given target quality and completion time,this paper undertook regression analysis and Analysis of Variance(ANOVA)using SPSS and MATLAB,on the selected crowdsourcing platform,i.e.,the Amazon Mechanical Turk to analyze how the configuration of tasks affects the performance of tasks.Firstly,we design experiments for micro-tasks,which are to allow online workers to watch a parking lot video and record when a car entered or left the parking lot.Also,we identify the inputs and outputs of the selected crowdsourcing platform.Then we post the tasks after assigning the values of inputs in order to get the values of the outputs,and design a screening test so as to find the factors which have significant impacts on the online workers' behavior.Lastly,we design a full factor experiment for the statistically significant factors,and establish regression models considering the interaction effect.There are more than one dependent variables for the experiment,so this is a multi-objective optimization problem which can be shown by Pareto frontier illustrated by MATLAB.Through the analysis above we can find the factors which have significant impacts on the online workers' behavior.Those factors are Incentive(Reward),Video Speed,Country,Life Time,Assignment Duration.The regression analysis has figured out how the statistically significant factors affect the behavior,that is,how well and how long the online workers complete the task given the configuration parameters of tasks.This analysis helps to understand the dynamics of crowdsourcing platforms and lays the theoretical and practical foundation for a more efficient use of the crowdsourcing platform.