Combining generative adversarial networks for emotion prediction in college students: An application to interactive musical interpretation

Traditional emotion prediction methods rely heavily on large amounts of labeled data and often struggle to capture subtle variations and individual differences in emotional expression. The goal of this paper is to enhance Generative Adversarial Networks (GANs) to improve emotion prediction accuracy, thereby providing college students with a more intelligent and personalized learning experience in interactive music interpretation. Firstly, a spatial channel attention mechanism is incorporated into the generator of the D2M-GAN multimodal generative adversarial network to improve the model’s ability to focus on important information. Additionally, the traditional large kernel convolutional layer is replaced by a convolutional layer with multi-scale convolution, enhancing the model’s ability to assess the authenticity of the generated data. To further optimize the model, the generative network is both rewarded and penalized using music theory rules, and the convergence speed is accelerated by optimizing the loss function. This improves the intelligence and personalization of interactive music interpretation. In this study, the accuracy and generalization ability of the proposed Deep Two-Modal Generative Adversarial Network with Spatial Channel Attention model (D2M-GAN-SCA) are evaluated using cross-validation and comparative validation. The experimental results demonstrate that the generator structure with the spatial channel attention mechanism, combined with the discriminator optimization strategy involving multi-scale convolutional layers, significantly enhances the accuracy of sentiment prediction. An accuracy of 97.03% is achieved after 1400 training iterations. Furthermore, the model shows notable improvements in loss function stability, convergence speed, and the quality of generated music. These advancements provide robust support for sentiment prediction and real-time interactive music generation, facilitating a more engaging and personalized online learning experience for college students in music interpretation.