The research and application of artificial intelligence(AI)in music have primarily focused on Western classical and popular music,while studies on traditional music from diverse global cultures are relatively insufficient.One main reason is the lack of high-quality datasets and annotation standards.In this paper,we explore the ongoing development of traditional music datasets from the perspectives of data content,annotation and application.Based on the in-depth analysis of the previous studies,we propose for the first time the construction principles for annotated music datasets aimed at artificial intelligence research,specifically addressing the characteristics of traditional music.According to this,we have constructed the first multimodal performance dataset of HuQin music (CCOM-HuQin),and take it as an example to demonstrate the research outcomes and application prospects of datasets in music AI.This paper emphasizes that the significance of applying AI to traditional music lies not only in assisting with composition and performance,but also in promoting the preservation,inheritance and development of traditional music.