Multilingual Obnoxious Message Classification using Bidirectional Encoder Representation from Transformers (BERT)

Everyday we witness incidents of social media bullying, harassment and various other toxic remarks. It has now become the part and parcel of social networking. Promotion of violence and hate speech, etc involve toxic comments in various forms on social media. The proposed model in this paper helps with the identification of all the obnoxious and questionable content on the internet by classifying the online comments as toxic or clean to tackle this problem. This is realised by using a pre-trained model - BERT. To enhance the performance, this model used RoBERTa (Robustly Optimized BERT Pre Training Approach), which has basically been formed by altering some hyperparameters which are tuned during the pre-training of BERT.