Multilingual Toxic comments Classification using Bert

Abstract

The swift expansion of online platforms has led to a surge in toxic comments, disrupting digital communities and adversely affecting users. Tackling this pervasive issue presents significant challenges, particularly in a multilingual context, as most available solutions tend to focus primarily on English. This project presents a multilingual toxic comments classification system harnessing Multilingual BERT (mBERT) capabilities. By utilizing mBERT's proficiency in various languages, the system can proficiently detect and classify toxic content—ranging from hate speech to abusive language—in real time. Fine-tuned on a diverse multilingual dataset, it promotes inclusivity by catering to less-resourced languages and providing a toxicity score for each comment to facilitate moderation. This innovative solution offers a robust and scalable method for cultivating healthier and more respectful online communities worldwide.