Article: Attention mechanism improved ST-GCN model for character action recognition in virtual drama performance Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.84 pp.1 - 24 Abstract: This study tackles inefficiencies in virtual drama character action recognition. An improved skeleton extraction method employs sparse matrix operations and a novel activation function to reduce parameters and enhance nonlinearity. The core model uses an enhanced spatial-temporal graph convolutional network with a learnable matrix and multi-head graph attention to resolve topological rigidity. A twin network enables precise action quality assessment. Experiments demonstrate the superior performance. It achieves 97.8% average skeletal feature recognition accuracy across dramatic scenes and 95.85% accuracy in specific scenarios such as ballet jumps, outperforming comparable models. It also maintains robustness in noise, reaching 81.2% accuracy at a 0.25 noise level. For quality assessment, its evaluations align closely with expert judgements, achieving 98.6% average accuracy in rhythm matching tasks. In summary, this technique provides effective technical support for optimising virtual drama character action systems. Inderscience Publishers - linking academia, business and industry through research

Title: Attention mechanism improved ST-GCN model for character action recognition in virtual drama performance

Authors: Xiaobin Cai

Addresses: Broadcasting and Hosting Art Program, Sanming University, Sanming, 365004, China

Abstract: This study tackles inefficiencies in virtual drama character action recognition. An improved skeleton extraction method employs sparse matrix operations and a novel activation function to reduce parameters and enhance nonlinearity. The core model uses an enhanced spatial-temporal graph convolutional network with a learnable matrix and multi-head graph attention to resolve topological rigidity. A twin network enables precise action quality assessment. Experiments demonstrate the superior performance. It achieves 97.8% average skeletal feature recognition accuracy across dramatic scenes and 95.85% accuracy in specific scenarios such as ballet jumps, outperforming comparable models. It also maintains robustness in noise, reaching 81.2% accuracy at a 0.25 noise level. For quality assessment, its evaluations align closely with expert judgements, achieving 98.6% average accuracy in rhythm matching tasks. In summary, this technique provides effective technical support for optimising virtual drama character action systems.

Keywords: virtual drama; action recognition; ST-GCN model; attention mechanism; skeletal features.

DOI: 10.1504/IJICT.2026.155479

International Journal of Information and Communication Technology, 2026 Vol.27 No.84, pp.1 - 24

Received: 20 Jan 2026
Accepted: 30 Apr 2026
Published online: 03 Aug 2026 *

Title: Attention mechanism improved ST-GCN model for character action recognition in virtual drama performance

Keep up-to-date