Tag
A systematic study on detecting Schwartz values in political text, comparing context lengths, model sizes, and retrieval-augmented generation methods. Results show that full-document context improves supervised models but not zero-shot LLMs, while retrieved moral knowledge consistently helps via early fusion.