Using the BERTimbau Model to Analyze Students’ Affective Subjectivity in a Virtual Learning Environment

This paper presents a sentiment analysis model based on the BERT (Bidirectional Encoder Representations from Transformers) model to identify students’ moods in textual information from a virtual learning environment (VLE). VLEs enable social interactions and exchanges of experiences and affectivity between teachers and students in online education. Thus, if teachers can access these subjective possibilities, they can consider them when acting in teaching and learning processes. The Affective Word Mining (AWM) framework, integrated into the ROODA VLE (Cooperative Learning Network, available at the Federal University of Rio Grande do Sul, Brazil), mines the student’s affective subjectivity recorded in communication tools. In the previous AWM version, the moods were mined based on the Affective Lexicon approach. In the current version, we decided to modify the algorithm for classifying moods using a deep learning approach. We therefore used BERTimbau as pretrained BERT model for Brazilian Portuguese. This study highlights the differences in the results between the Lexicon-based and BERT-based approaches, as well as discussing the limitations of the current version. The research methodology adopted is an applied approach, using case studies and integrating quantitative and qualitative evidence. We analyzed the texts of 20 graduate students and the results show significant differences between the Lexicon and BERT sentiment analysis models. This distinction emphasizes the need for continuous improvement of the model. Hence, this research provides evidence that the use of BERT-based sentiment analysis can be an indicator to infer students’ moods and can contribute to the improvement of pedagogical practices.