A Method to Improve Crowdsourcing Outcome and to Reduce Calculation Costs Using Machine-Learning

Our study aims to reduce the calculation cost and improve the quality of crowdsourcing outcomes. While having all workers complete all tasks and taking a majority vote would obtain the most accurate outcome, it is impractical due to work time and monetary cost. We consider using a machine learning model that imitates work characteristics instead of real workers’ characteristics to fill in data that has not worked. Constructing a machine-learning model for each worker causes significant calculation costs, therefore we assume workers have similar work characteristics and cluster them into groups.By constructing machine-learning models for each group, we can reduce the number of models required, lowering calculation costs while maintaining accuracy. Then, we experiment to verify the effectiveness of our method.When the number of clusters is 4, the accuracy is 0.929, surpassing the performance of simple majority voting.Our method produces outcomes at lower calculation costs than methods that construct machine-learning models for all workers.