Article: A dynamic multimodal evaluation model for college aesthetic education integrating graph attention network and reinforcement learning Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2026 Vol.18 No.16 pp.51 - 70 Abstract: This study innovatively proposes a cross-scene multimodal data dynamic evaluation model named cross-scenario multimodal fusion dynamic evaluation (CS-MFDE), which combines graph attention network (GAT) and reinforcement learning technology. The test results on Cora public dataset show that the CS-MFDE model has achieved excellent results. The test accuracy is as high as 0.94. The F1 score is 0.934. The recall rate is 0.936. On the dataset of aesthetic education in colleges and universities, the model also performs well, and the dynamic time warping (DTW) distance of dynamic trajectory fitting is only 6.23, which is 1.42 lower than the optimal baseline. The accuracy of cross-time prediction reaches 88.6%, which is 4.1% higher than the optimal baseline. The macro average F1 score is 0.867, which is 0.032 higher than the optimal baseline. The time consistency error has also reached the lowest level. Inderscience Publishers - linking academia, business and industry through research

Title: A dynamic multimodal evaluation model for college aesthetic education integrating graph attention network and reinforcement learning

Authors: Huili Nie

Addresses: Department of Primary Education, Jiaozuo Normal College, Jiaozuo, Henan, 454100, China

Abstract: This study innovatively proposes a cross-scene multimodal data dynamic evaluation model named cross-scenario multimodal fusion dynamic evaluation (CS-MFDE), which combines graph attention network (GAT) and reinforcement learning technology. The test results on Cora public dataset show that the CS-MFDE model has achieved excellent results. The test accuracy is as high as 0.94. The F1 score is 0.934. The recall rate is 0.936. On the dataset of aesthetic education in colleges and universities, the model also performs well, and the dynamic time warping (DTW) distance of dynamic trajectory fitting is only 6.23, which is 1.42 lower than the optimal baseline. The accuracy of cross-time prediction reaches 88.6%, which is 4.1% higher than the optimal baseline. The macro average F1 score is 0.867, which is 0.032 higher than the optimal baseline. The time consistency error has also reached the lowest level.

Keywords: core literacy of aesthetic education; cross-scenario fusion; multimodal learning; dynamic evaluation; graph attention network; GAT; reinforcement learning.

DOI: 10.1504/IJRIS.2026.154386

International Journal of Reasoning-based Intelligent Systems, 2026 Vol.18 No.16, pp.51 - 70

Received: 11 Mar 2026
Accepted: 10 May 2026
Published online: 25 Jun 2026 *

Title: A dynamic multimodal evaluation model for college aesthetic education integrating graph attention network and reinforcement learning

Keep up-to-date