Enhancing Data Privacy in Large Language Models through Private Association Editing