In this paper, the dance teaching digitization process and virtual character dance movement dataset are designed, combined with the Vicon motion capture system, neural network fusion shape technology and dance movement recognition algorithm to realize dance teaching digitization. The traditional K-modes algorithm is improved from three aspects: determining the number of clustering families, the determination of the initial clustering center, and the clustering distance metric to construct a digital dance teaching quality evaluation model. After the application analysis, it was concluded that the difference between the students of the experimental group and the students of the conventional group of dance movements was P<0.05, and the learning situation of the experimental group showed an upward trend, the learning enthusiasm increased by 0.21, the learning efficiency increased by 0.23, and the clustering index were higher than 0.9. Dance teaching based on digital technology can improve the quality and effect of the teaching.