Esplorando l'utilizzo dei modelli Transformers per Sentiment Analysis su Twitter

This thesis discusses the importance of sentiment analysis on Twitter, where the vast amount of data generated can provide valuable insights into people’s thoughts and opinions about various topics, including the COVID-19 pan- demic. The study focuses on transformer-based methods, including BERT and its variants, for sentiment analysis on Twitter data. These models leverage self-attention mechanisms to capture contextual information and have achieved state-of-the-art results in various natural language process- ing tasks. The thesis results show that BERT outperforms other models in terms of accuracy, precision, recall, and F1 score. However, the use of trans- former models comes with a trade-off in terms of computational expenses, which may make them less suitable for applications with limited computa- tional resources. Overall, this study provides insights into the strengths and limitations of transformer-based methods in sentiment analysis on Twitter data. 2