Decoding Knowledge in Large Language Models: A Framework for Categorization and Comprehension