Construction of online classroom instructional quality assessment system of university music based on BP neural network

In view of the strong subjectivity of traditional instructional quality assessment methods and the difficulty in quantifying nonlinear teaching elements, this article proposes an online classroom assessment model of college music based on improved BP neural network (BPNN). The model integrates multi-dimensional data such as the frequency of teacher-student interaction and the utilization rate of teaching resources, and constructs a nonlinear mapping structure with double hidden layers. At the same time, the optimization strategy of dynamic learning rate (initial value 0.01, attenuation coefficient 0.9) and momentum factor (coefficient 0.9) is adopted. After cleaning and standardization, the real data is divided into training set and test set according to the ratio of 7: 3. Through cross-validation, the optimal number of hidden layer nodes is 25, and the normality of error distribution is verified (p > 0.05). The experimental results show that the prediction accuracy of this model reaches 95.2%(95% confidence interval is [93.7%, 96.4%]), which is 20.69% higher than the traditional ID3 algorithm, and the mean absolute error (MAE) is reduced to 0.032. The model is excellent in capturing complex indicators such as emotional interaction between teachers and students and innovation of teaching content (F1 value is 0.93), and it has stable generalization ability for data of different teaching platforms (standard deviation < 1.5). The dynamic learning rate strategy improves the training efficiency by 37% and effectively avoids the local optimization problem. This study confirms the effectiveness of neural network in education assessment and provides practical reference for the digital transformation of music education. In the future, this achievement is expected to be extended to interdisciplinary online teaching scenarios, thus promoting the development of education in the direction of fairness and individuality.