Uma abordagem baseada em LLM para analisar sentimentos dos usuários do Twitter em relação ao impacto da IA no dia a dia considerando a evolução do tempo

Abstract

The growth of generative Artificial Intelligence tools has expanded discussions about their impacts and applications in everyday life. In this context, this work presents a sentiment analysis of Portuguese-language Twitter (X) posts related to GPT/ChatGPT, Gemini, and Copilot technologies. The study analyzes periods associated with the initial releases and the most recent versions of these tools, using a process composed of data collection, preprocessing, semantic filtering, and automatic sentiment classification with Large Language Models (LLMs) in a zero-shot approach. In addition to the classification performed by the main model, a strategy based on multiple LLMs was employed to support the validation of the generated classifications. The results show a predominance of neutral sentiments throughout most of the analyzed period, as well as differences in the evolution of positive and negative sentiments among the evaluated technologies. Overall, the results indicate that LLMs can be applied to sentiment analysis tasks on social media, even in scenarios involving short texts, informal language, and the absence of previously labeled datasets.