Recently, methods investigating how to adapt large language models (LLMs) for specific scenarios have gained great attention. Particularly, the concept of \textit{persona}, originally adopted in dialogue literatu