Is Cognition and Action Consistent or Not: Investigating Large Language Model's Personality