From the perspective of how to understand students better, the study provides a feasible path for teachers’ differentiated teaching and role reconstruction by constructing learner profiles and clustering and subclustering. The “learning-learner” feature layer is constructed, and the multimodal data of learners are integrated from six dimensions, such as intrinsic layer, learning layer, and expectation layer, and transformed into six dimensions, such as theory grade, performance grade, cooperation frequency, independent practice, and so on. It is transformed into seven quantifiable music learning effect indicators, such as theory score, performance score, cooperation frequency, and independent practice time. Canopy-Kmeans clustering algorithm was used to identify different types of students, and real data from 124 music performance majors were selected for analysis. The students’ overall theoretical and practical scores amounted to 83.02 and 76.75, and the average expected match was 89.83%, which was a good overall performance but with significant individual differences. Correlation analysis further revealed that peer evaluation and expectation fulfillment were highly correlated with r=0.919. Through Canopy-Kmeans clustering, the students were clearly classified into three categories, with 25 students of the leading type who performed comprehensively and were significantly excellent in all the indicators, especially in theory (95.8), practice (93.24), and frequency of collaboration (52.84 times). The 85 students of the intermediate type were in the middle to upper range of all indicators and had the greatest improvement in skills (2.67). In contrast, 14 of the lagging type were significantly lower in theory, practice, practice engagement and collaborative participation.
