COMPARATIVE SENTIMENT ANALYSIS FOR ONLINE HARASSMENT DETECTION USING DISTILBERT AND VADER

Abstract

Online harassment targeting women on social media platforms has turned into a major issue. It demands effective tech-based solutions right away. Our work here compares two main methods for sentiment classification. VADER works through rule-based scoring systems. DistilBERT depends on neural network models instead. We ran tests on both using the Jigsaw Toxic Comment Classification dataset. The results showed DistilBERT reaching 91.5 percent accuracy in classification. That figure stands much higher than VADER's 70.9 percent score. All this points to transformer models handling complex language and context in toxic online talks pretty well.