An Iterative Utility Judgment Framework Inspired by Philosophical Relevance via LLMs