Title: Design of deep learning-assisted practical teaching system based on multi-level semantic feature extraction and text matching modelling

Authors: Li Peng; Zhenglong Wang; Jing Xiao; Huan Ning

Addresses: College of Economics and Management, Hunan Applied Technology University, Changde, 415100, China ' College of Economics and Management, Hunan Applied Technology University, Changde, 415100, China ' College of Economics and Management, Hunan Applied Technology University, Changde, 415100, China ' College of Economics and Management, Hunan Applied Technology University, Changde, 415100, China

Abstract: This study presents the A-BRUNet model for multi-level feature extraction, designed to achieve high-precision text matching and classification by seamlessly integrating bidirectional encoder representations from transformers (BERT), bidirectional gated recurrent unit (Bi-GRU), convolutional neural networks (CNN), and an attention mechanism. The model first utilises BERT to extract word vectors enriched with contextual semantic depth, followed by Bi-GRU to capture global semantic relationships. CNN, equipped with multi-scale convolutional kernels, identifies local salient features, while the attention mechanism assigns adaptive weights to different feature layers, refining the overall semantic representation. Experimental results on the Quora question pairs (QQP) and Microsoft Research Paraphrase Corpus (MRPC) datasets demonstrate that A-BRUNet significantly outperforms existing models in both accuracy and F1-score for text matching tasks. Furthermore, in a limited-sample test using randomly selected datasets, the proposed model consistently exhibits robust performance, highlighting its adaptability and generalisability in small-sample scenarios. These findings establish A-BRUNet as a technical foundation and research benchmark for optimising future intelligent teaching frameworks.

Keywords: multi-level semantic analysis; text matching; construction of teaching system.

DOI: 10.1504/IJCAT.2026.151388

International Journal of Computer Applications in Technology, 2026 Vol.78 No.1, pp.83 - 93

Received: 13 Jan 2025
Accepted: 04 Jun 2025
Published online: 26 Jan 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article