COMPARISON OF FOUR PRE-TRAINED MODELS OF SENTIMENTAL ANALYSIS ON COVID 19 NEWS HEADLINES

Abstract

This research paper presents a comprehensive comparison of four pre-trained models for sentiment analysis applied to COVID-19 news headlines. The COVID-19 pandemic has triggered an unprecedented flow of information and public discourse, making it essential to understand the sentiments and opinions expressed in news headlines. The study evaluates the performance of RoBERTa, VADER, BERT, and TextBlob in this specific context. In this research, a diverse dataset of COVID-19 news headlines is collected and preprocessed. Each of the four models is employed to perform sentiment analysis on this dataset. The evaluation involves assessing the recall, accuracy, and F1 score of these models in capturing the nuances of sentiment and emotion within the headlines. The findings of this comparative analysis reveal important information. RoBERTa, a powerful transformer model, exhibits a robust performance in understanding the subtleties of sentiment in COVID-19 news headlines. VADER, a rule-based system, demonstrates its adaptability to this domain. BERT, a sibling of RoBERTa, showcases its ability to discern sentiment nuances, albeit with some complexities. TextBlob, a simplistic rule-based library, provides a user- friendly approach to sentiment analysis.