Comparative Evaluation of Large Language Models for Sentiment Analysis

Sentiment analysis is an important task in the processing of natural language which is necessary for understanding public opinion and emotions in the fields such as social media, feedback from customers and market analysis. Despite notable progress, a comprehensive evaluation of different models (LLMs) is needed to understand the strengths and weaknesses of these models in the tasks of sentiment analysis. This research performs a comparative analysis of two prominent LLMs: BERT, a wellestablished model best known for text classification, and LLaMA 3.2, a newly released model by Meta Technologies. While previous research has thoroughly used models like BERT, there is still a gap in the literature regarding the performance and efficiency of LLaMA 3.2, especially in the context of sentiment analysis. By fine-tuning these models on a dataset consisting of 12 distinct sentiment classes, this research seeks to fill this gap by providing new findings into the comparative evaluation of these LLMs. The outcomes of this study are significant as they offer a comprehensive analysis of LLaMA 3.2's capabilities relative to established models, helping researchers and practitioners select the most suitable model for sentiment analysis tasks. It is essential to solve this problem to strengthen the state-of-theart in sentiment analysis, resulting in more accurate and better applications across various industries.