Advancements in Text Subjectivity Analysis: From Simple Approaches to BERT-Based Models and Generalization Assessments

Text subjectivity is an important research topic due to its applications in various domains such as sentiment analysis, opinion mining, social media monitoring, clinical research and patient feedback analysis. While rule-based approaches dominated this field at the beginning of the 21st century, contemporary works rely on transformers, a specific neural network architecture designed for language modeling. This paper explores the performance of various BERT-based models, including our fine-tuned BERT (Bidirectional Encoder Representations from Transformer) model, and compares them with pre-built models. To assess the generalization abilities of the models, we evaluated the models on benchmark datasets. Additionally, the models underwent evaluation on two synthetic datasets created using large language models. To ensure reproducibility, we have made our implementation publicly available at https://github.com/margitantal68/TextSubjectivity .