Toxic Comment Identification and Classification using BERT and SVM

Abstract

Bullying cases like toxic comments on many social media platforms cause a negative impact that occurs in every age circles. From those cases, we would like to make a system that can identify and classify toxic words from a comment before it is sent and seen by others. By utilizing a Machine Learning application, hopefully, the produced system can be useful in reducing bullying cases that are many in social media. Lot of experiments have been done to find the settlement for this problem, but various algorithms and models are used. In this research, we will be doing a comparison of two models, the BERT (Bidirectional Encoder Representations from Transformers) model which is usually used to solve NLP (Natural Language Processing) tasks, and SVM (Support Vector Machine) model which is great at classifying. Both models will be compared to find out which model is better in identifying and classifying toxic comments. The result that is gotten shows that BERT model is said to be superior compared to SVM model, with an accuracy of 98.3% including other metric evaluation scores that show a significant result compared to the result achieved by SVM model.