Revisiting Automated Topic Model Evaluation with Large Language Models