This study explores personalization of household cleanup by robots, using language-based planning and few-shot summarization with LLMs to infer generalized user preferences, achieving 91.2% accuracy on unseen objects and 85.0% success in real-world test scenarios.