Choral works, as an important form of musical expression, have a complex compositional process and require a deep foundation in music theory. The rapid development of artificial intelligence technology provides new possibilities for music creation. In order to solve the problems of low efficiency and limited creative ideas of traditional choral works, this study constructs a choral works creation model based on the combination of deep learning and knowledge graph. Methodologically, the Transformer and BART models are adopted as the core architecture, and the deep mining and representation learning of semantic information of choral works is realized through keyword extended graph generation, TransE knowledge representation encoder and two-layer graph attention network. Specifically, the keyword expansion graph is constructed by utilizing the knowledge map of ancient poems, the keyword semantic representation is enhanced by the multi-head graph attention mechanism, and the dual cross attention mechanism is incorporated in the BART decoder to improve the quality of text generation. The experimental results show that the model achieves 75.4%, 76.3%, and 78.4% in terms of accuracy, recall, and F1 value, respectively, which significantly outperforms baseline models such as SVM, BiLSTM, and TextCNN. The model achieves convergence at about 460 training sessions, and the convergence speed is significantly faster than the comparison model. The application practice shows that the average score of students’ choral works in the experimental group reaches 86.24, which is 5.58 points higher than that of the control group. The study shows that the model can effectively support the creation of multi-topic choral works, which provides a new technical path for the intelligent development in the field of music creation.
