In dance, we must understand the essential meaning of dance movements from the inside and express them on the basis of dance. Therefore, in the process of developing new dance teaching methods, it is necessary to improve the basic education of dance students, so that they can express the emotions conveyed by dance through body language and movements, and improve dance expression ability. In this context, we made the research and reached the following conclusions: (1) the number of frames of different dance types is also different, and the number of frames to be learned is also increasing. The dance with the highest number of frames is Latin2, which has 3635 frames, and the dance with the highest number of frames that need to be learned is also Latin2, which requires 2519 frames to learn. (2) The data mining method is still the highest among the three methods, and the accuracy of the complete teaching method is 82%, which is the lowest among the three methods, and the accuracy of the decentralized teaching method is 87%. No matter in the test set or the mixed test set, the curve values of deep mining are very stable. First of all, human movements emphasize that in dance, the essential meaning of dance movements needs to be understood from the inside and expressed through the foundation of dance. Therefore, when developing new dance teaching methods, it is necessary to strengthen the basic dance training of students so that students can express the emotions conveyed by dance through body language and movements and improve their dance expression ability. We conduct research in this ecological environment. Different types of dance learning process using different frames, different types of dance in the algorithm transport have different recognition methods, using better and different algorithms can achieve the best performance. Both groups in the Hip Hop dance had a shorter average learning time than both groups in the Latin dance.