LLM-POTUS Score: A Framework of Analyzing Presidential Debates with Large Language Models