Argument Quality Assessment in Brazilian political tweets

Abstract

Argumentation is an inherent skill in human communication, both in oral and written situations. Well-founded arguments are important to support decision-making and learning, as well as to reach widely accepted conclusions. As a research area, argumentation is a multidisciplinary field that studies the processes of debate and reasoning. In computational linguistics, investigations have been carried out to (i) identify arguments and their units and (ii) generate or (iii) evaluate the quality of arguments. However, most current work focuses on argument mining in formal English texts. In this article, we evaluated the quality of argumentation in political domain tweets, written in Brazilian Portuguese, using traditional machine learning algorithms – such as Logistic Regression, KNearest Neighbor, Decision Trees, Support Vector Machines (SVM), Random Forest and Naive Bayes – and also a fine-tuning of two neural models (BERTimbau and RobertaTwitterBR). In addition to bringing practical results for the assessment of argumentation quality in a challenging textual genre, such as Twitter, and in a controversial domain, such as Brazilian politics, this article also aims to fill in the lack of works that automatically assess the quality of arguments in Portuguese. Among the evaluated classification algorithms, the model obtained from the fine-tuning of BERTimbau presented the best results, with an accuracy of 69.65\% when all classes were considered and 100.00\% for messages with high quality of argumentation