AI Wizards at CheckThat! 2025: Enhancing Transformer-Based Embeddings with Sentiment for Subjectivity Detection in News Articles
The episode discusses a paper enhancing Transformer-Based Embeddings with sentiment for subjectivity detection in news articles. Hosts discuss fusing sentiment scores with sentence representations, using threshold calibration to manage class imbalance, and the results showing performance gains in languages like English and Italian. They conclude that combining feature engineering and post-processing leads to more nuanced, context-aware AI systems.