Effectiveness of utilizing meta attributes in anomaly detection using crowdsourcing

Abstract

In recent years, the clearance rate for crimes has remained low. One of the reasons for this is that it is difficult to clear crimes based on after-the-fact reports from surveillance cameras. Increasing the clearance rate requires manpower to constantly monitor surveillance cameras. Crowdsourcing is one way to secure manpower, and  nomaly detection using crowdsourcing has already been proposed. However, anomaly detection using crowdsourcing has a problem in that the accuracy rate decreases in challenging tasks. Challenging tasks are tasks in which it is difficult for crowdworkers to determine whether there are anomalies. In this study, we evaluate the improvement of response accuracy in challenging tasks by using crowdworkers’ response time and confidence in their answers as meta-attributes.