A Comparative Analysis of Various Machine Learning and Transformer Models for Harmful Tweet Detection
Tithi Rani Das, Shima Chakraborty, Abu Nowshed Chy · 2024
The COVID-19 pandemic's effects have remained for more than 2 years. During this time, people use social media to collaborate and acquire news, but these platforms can also be used for undesirable activities including distributing false information, making false claims, and posting harmful content. Harmful content can also take many different forms ranging from hate speech and profanity to cyberbullying. It has a subliminal influence over our thoughts, psychology, and feelings, which results in increasing the spreading of harmful, hateful, and rumored news among users, thus leading to an increase in aversions to taking the COVID-19 vaccine. In this paper, our study's objective was to identify harmful tweets. We present a comparative performance analysis on different machine-learning algorithms and the transformer models. Among them, RoBERTa transformer based model achieves the highest binary F1-score (positive class) of 42% and surpasses the prior state-of-the-art models by a large margin.