SentPT: A customized solution for multi-genre sentiment analysis of Portuguese-language texts

Sentiment analysis is a data -driven task, and the resources currently available mostly cover only a couple of text genres in specific contexts. Notably, sentiment analysis advancements have primarily centered on high-resource languages, whereas numerous languages and their speakers are overlooked. This paper introduces SentPT, a novel polarity classifier designed for sentiment analysis of Portuguese-language texts spanning various genres. The aim is to address the gap in multi -genre sentiment analysis by offering a customized solution. To this end, we curate a comprehensive dataset covering different contexts, such as news, literary texts, opinions, comments, social media, and more, followed by preprocessing for consistency. Our proposed classifier adopts a Transfer Learning approach, fine-tuning a BERT model, and is evaluated against a diverse set of texts, including product reviews, literary works, news articles, and game comments. The evaluation employs traditional metrics like precision, recall, and F1 -score, with SentPT demonstrating the best overall performance. Our classifier proves effective for formal and informal texts, outperforming existing systems.