Automatic Evaluation for Cyberbullying Detection Method based on Statistical Scale

Abstract

Various methods have been proposed to detect harmful information automatically as a support of cyberbullying countermeasures. However, there remain various problems such as detection performance deterioration over time and detection of information including private information.In this paper, we propose an automatic evaluation method for harmful expression extraction method using BLEU, an automatic evaluation method typically used in evaluation of machine translation methods. In this method, for as a detected as a harmful by the harmful expression extraction method, the evaluation is performed based on word N-gram Precision using the harmful entry gold standard dataset. As a result of the evaluation, the proposed allowed assigning a high score to harmful entries, which confirmed the effectiveness of the method.