EDM
Temperature and Persona Shape LLM Agent Consensus With Minimal Accuracy Gains in Qualitative Coding
Wed Jul 1, 2:25 PM–2:50 PM · North 202
Part of AI Methods & Metrics
★ Notable speakers
Conrad Borchers
★
— AI-supported self-regulated learning; intelligent tutoring systems; educational data mining; ordered network analysis
Finds that varying LLM agent temperature and persona shapes consensus outcomes in qualitative coding tasks, but produces only minimal accuracy gains.
Authors
Conrad Borchers, Bahar Shahrokhian, Francesco Balzan, Elham Tajik, Sreecharan Sankaranarayanan, Sebastian Simon