Exploring the Effectiveness of Generative Models in Sentiment Analysis Tasks in Brazilian Portuguese

Abstract

Large language models (LLMs) have been successfully applied in various natural language processing (NLP) tasks. This paper investigates their effectiveness in sentiment analysis tasks in the context of Brazilian Portuguese, exploring the identification of opinionated sentences, polarity, and comparative sentences. The study evaluates the performance of models such as ChatGPT and Sabiá on different tasks and datasets, comparing them with methods from the literature. Furthermore, we explore the use of LLMs in automatic data annotation. The results demonstrate the potential of LLMs in sentiment analysis, especially in polarity identification, and discuss their limitations and applications in data annotation tasks.