Subjectivity Detection with Machine Learning & Transformers

The project investigated how language models can distinguish between subjective and objective sentences. This is relevant to text classification, content analysis, news analysis and automated sorting of large volumes of text.

An encoder-only Transformer/BERT-based model was fine-tuned to classify sentences as subjective or objective. The work included data processing, training, evaluation and comparison against simpler baselines.

The Transformer model was evaluated against simpler baselines on an imbalanced dataset, using evaluation measures that account for the class imbalance. The project shows practical experience with Transformer-based NLP models and evaluating classification on imbalanced data.