Anomaly Detection Support for Crowdworkers by Providing Anomaly Scores

In recent years, crime clearance rates have remained low. To improve them, current methods have used deep learning and crowdsourcing to automatically detect crimes. However, these methods suffer from false positives and false negatives and also fail to achieve high accuracy (AUC, correct answer rate). This paper proposes a method that supports the judgments of crowdworkers by providing them with anomaly-score output from deep learning models. The proposed method enhance the number of correct judgments made by crowdworkers. As a result, false positive and false negative rates are reduced and accuracy is improved.