Aplicação de técnicas de inteligência artificial para análise de tweets sobre Institutos Federais Brasileiros

Abstract

The increasing use of social media as communication platforms has generated large volumes of textual data, offering valuable potential for analysis and insights. In the context of public institutions, particularly Brazilian Federal Institutes (IFs), the analysis of this data can offer a detailed understanding of public opinion, allowing for example the monitoring of the general sentiment of the community towards IFs, the anticipation of communication crises and the assessment of the support or disapproval of measures adopted by management. To address these issues, this research explores Artificial Intelligence (AI) techniques, specifically Sentiment Analysis and Topic Modeling, on social media texts related to IFs. Eleven different techniques were investigated, including lexical dictionaries, machine learning methods, and transformer-based approaches. Results reveal that transformer-based methods, especially BERTimbau, significantly outperform other techniques in terms of Sentiment Analysis effectiveness, achieving an effectiveness of up to 86%. This suggests that these techniques are promising for future applications in analyzing sentiments related to IFs. Moreover, Topic Modeling with the BERTTopic technique identified relevant global and regional topics, including topics about campus communication team notices, students' fear and anxiety about assessments, regional concerns such as food, and other relevant topics. These findings highlight the potential of AI techniques for sentiment and topic analysis in social media texts related to IFs, providing valuable insights for management and decision-making.