A Reproducibility Study of Detecting Cyberbullying Across Different Social Media Platforms Using Deep Learning

Cyberbullying is a serious issue in social media. Every day, people produce a huge amount of content over different platforms. Sometimes, it is offensive and aims at hurting others. There are already some mechanisms to report and cope with this phenomenon; however, most are ineffective. Therefore, the topic is analysed in the existing literature. Some approaches use traditional Machine Learning methods, but their performance is not satisfactory. In recent studies, Deep Learning models are examined and are claimed to have superior results. In this paper, we reproduced a study that used Deep Neural Network models for classifying offensive content. We compared its performance to the traditional machine learning models. We were able to reconduct the experiments; nevertheless, we observed that the experimental setup used in the study had some flaws. It led to overly optimistic results. We point out the limitations of the current setup and compare our results with the reported ones.