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

Researchers tested sentence, window, and full-document inputs, plus retrieval-augmented setups, for Schwartz value detection in political text.

Full-document context helped supervised DeBERTa models, while retrieved moral knowledge improved performance across both supervised and zero-shot models.

Larger models were not consistently better; the best results came from matching context length and external knowledge to the model type.

The findings suggest value classification improves more from the right setup than from simply using longer inputs or bigger models.