Analysis and Detection of Defamatory Posts against Athletes on Social Media

Defamation against athletes on social media has become a significant issue. Athletes who are subjected to defamation may not be able to perform at their best in the upcoming event. Therefore, it is necessary to develop a method to automatically detect defamation against athletes on social media at an early stage. In this paper, we analyze Japanese defamatory posts against athletes on Twitter (currently X) and propose methods to detect defamatory posts using feature-based models and pre-trained language models. Moreover, we investigate the effectiveness of using the source post of defamatory posts in defamation detection task. From the experimental results, we found that the method based on BERT achieved the best F1-score of 0.838 among all the methods. We also confirmed that the use of both source and reply posts was effective in detecting defamatory tweets.