The rapid development of digital music industry promotes the in-depth application of artificial intelligence (AI) in the field of music creation. Aiming at the problems of insufficient emotion expression and limited human-computer interaction in AI music generation, this paper constructs a topological sort-guided optimization model for AI music generation and human-composition interaction. Methodologically, a topological network structure characterization is used to establish a guitar chord generation mechanism, the quality of music generation is optimized by Deep Convolutional Generative Adversarial Network (DCGAN) combined with unilateral label smoothing and feature matching, emotion-driven music creation is realized based on the emotion-guided diffusion model, and a hierarchical attention mechanism is designed to enhance the rhyme and emotional expression of the lyrics. The experimental results show that the model achieves an excellent performance of 4.5698, 0.2485, 0.0455, 0.0198 on seven objective evaluation indexes such as PR, PE, PH, SC, etc., and the total subjective evaluation score is 4.3485, with a mean value of 7.7419 and 8.3089 on the Lakh MIDI and MUT MIDI datasets, respectively. The study verifies that the effectiveness of the topological ordering guidance mechanism in improving the quality of AI music generation and human-computer interaction experience, which provides a new technical path for intelligent music creation and promotes the development and application of AI music generation technology.
