Research on improving KGQA efficiency using self-enhancement of reasoning paths based on Large Language Models