Measuring Human Value Expression in Social Media Texts: Calibrated LLM Annotation and Encoder Transfer

The paper presents a methodology for annotating social media texts based on Schwartz's theory of basic human values, using calibrated LLMs to enhance the accuracy of value expression measurement. It evaluates the performance of various LLMs in terms of precision, recall, and F1 scores, while also addressing structural alignment and error analysis to improve annotation stability and reduce misattributions. The findings highlight the importance of theory-driven annotation procedures for practitioners, enabling more reliable transfer of value interpretations to encoder models through soft-label training, which is crucial for scalable applications in sentiment analysis and value-based content moderation.