Tag
This paper introduces a structured approach to extracting persona information for LLM-based digital twins, showing that structured representations improve predictive accuracy over raw transcripts. An automatic pipeline is proposed to adapt structures to different tasks.