More Context, Larger Models, or Moral Knowledge? A Systematic Study of Schwartz Value Detection in Political Texts

The study investigates the detection of Schwartz values in political texts, analyzing the impact of context and moral knowledge on performance. Using supervised DeBERTa-v3 models and zero-shot LLMs ranging from 12B to 123B parameters, it finds that while full-document context improves performance for DeBERTa encoders by 3.8-4.8 macro-F1 points, it does not consistently benefit zero-shot models. The research highlights the importance of integrating context and moral knowledge in value-sensitive NLP, suggesting that larger models and longer inputs do not universally enhance performance.