Protecting Mental Health by Detecting Toxic Comments Using Deep Learning Models

Toxic comments have become a problem in social media due it's side effects that gives bad environment to the space and giving trauma to everyone on using social media. Many researchers are trying to solve this problem, one of the methods is leveraging the capability of Deep learning. Compared to other methods, researchers are using Deep Learning aim to enhance efficiency by automating tasks and allowing machines to handle the work instead of humans. In this study, three deep learning models are used and compared it's performance on classifying toxic comments on Wikipedia's comments taken from Kaggle. The models are TextCNN, Bi-LSTM-GRV, and distilBert. The dataset is processed in the same way for all the models. From the three models, overall TextCNN has the best performance among the other models and made it to the first place with accuracy of 98.37% and 98.85% for ROC AUC score. In conclusion, TextCNN could be an option of detecting toxic comments using Deep learning.