Recent advancements in large language models (LLMs) have been significantly propelled by pre-trained language models, enabling them to excel in various natural language processing tasks, including complex reasoning and mathematical problem-solving. Despite their impressive capabilities, LLMs still encounter challenges with factual knowledge judgment, often leading to inaccuracies or “hallucinations”. To address this, integration of knowledge graphs with LLMs has been explored to enhance Knowledge Base Question Answering (KBQA). In the context of person attribute knowledge graph question answering, which often involves numerical data and complex relationship reasoning, we introduce an iterative reasoning framework that leverages LLMs to interact with the knowledge graph effectively. Additionally, we propose a dynamic demonstrations retrieval method to augment LLMs’ reasoning capabilities with high-quality prompts. Experimental results demonstrate that our approach not only achieves superior performance but also won the first prize in the CCKS2024 Person Attribute KBQA task leaderboard.