Article: Enhancing accuracy of pragmatic ability tests through multi-feature fusion based on graph neural networks Journal: International Journal of Information and Communication Technology (IJICT) 2025 Vol.26 No.50 pp.133 - 149 Abstract: Pragmatic ability assessment holds significant importance in language teaching and related fields, yet existing methods fail to capture and utilise the characteristics and information across different modalities. To address this, this paper optimises graph neural networks through multi-stage adaptive fusion. By decomposing the graph neural network into a multi-stage training format, higher-order features of graph data are progressively integrated into shallow models across multiple stages, thereby training a more robust shallow model. Subsequently, a pragmatic competence prediction model based on an improved graph neural network and multi-feature fusion is proposed. First, modal information is progressively integrated to ensure comprehensive fusion. Then, long-range pragmatic information is captured and incorporated into sentence-level information, enabling the model to better understand global features. Experimental results demonstrate that the proposed model achieves at least a 3.46% improvement in pragmatic competence test accuracy, facilitating more precise assessment of pragmatic competence levels. Inderscience Publishers - linking academia, business and industry through research

Title: Enhancing accuracy of pragmatic ability tests through multi-feature fusion based on graph neural networks

Authors: Teng Xie; Dingyu Liu; Wei Zhou

Addresses: College of Teacher Education, Longyan University, Fujian 364000, China ' College of Teacher Education, Longyan University, Fujian 364000, China ' College of Foreign Languages, Longyan University, Fujian 364000, China

Abstract: Pragmatic ability assessment holds significant importance in language teaching and related fields, yet existing methods fail to capture and utilise the characteristics and information across different modalities. To address this, this paper optimises graph neural networks through multi-stage adaptive fusion. By decomposing the graph neural network into a multi-stage training format, higher-order features of graph data are progressively integrated into shallow models across multiple stages, thereby training a more robust shallow model. Subsequently, a pragmatic competence prediction model based on an improved graph neural network and multi-feature fusion is proposed. First, modal information is progressively integrated to ensure comprehensive fusion. Then, long-range pragmatic information is captured and incorporated into sentence-level information, enabling the model to better understand global features. Experimental results demonstrate that the proposed model achieves at least a 3.46% improvement in pragmatic competence test accuracy, facilitating more precise assessment of pragmatic competence levels.

Keywords: pragmatic competency assessment; graph neural network; GNN; multimodal feature; multi-stage optimisation; adaptive fusion.

DOI: 10.1504/IJICT.2025.151066

International Journal of Information and Communication Technology, 2025 Vol.26 No.50, pp.133 - 149

Received: 22 Oct 2025
Accepted: 15 Nov 2025
Published online: 12 Jan 2026 *

Title: Enhancing accuracy of pragmatic ability tests through multi-feature fusion based on graph neural networks

Keep up-to-date