Integration of Systemic Functional Grammar to improve automated sentiment analysis with GPT-4o

This study first examines the uncertainties and ambiguities underlying the practical and theoretical foundations of SA, particularly the challenges associated with the discretization of linguistic categories, a well-documented issue in linguistics. It then evaluates the impact of integrating SFG into GPT-based automatic SA detection by comparing human annotations, transformer-based machine learning annotations, GPT-4o predictions without SFG, and GPT-4o predictions with SFG. The use case is a sample of Italian Twitter posts from late 2013 to early 2015, drawn from the xLiMe Twitter Corpus (Rei, Mladenić, and Krek 2016).