Article: Intelligent English composition scoring method based on transfer paragraph segmentation and Pearson detection Journal: International Journal of Continuing Engineering Education and Life-Long Learning (IJCEELL) 2026 Vol.36 No.13 pp.1 - 25 Abstract: With the growing demand for intelligent education, automated English composition scoring has gained significant research attention. Addressing limitations in cross-prompt generalisation and feature extraction, this study proposes a transfer learning-based method incorporating paragraph segmentation and Pearson detection. The approach combines BERT encoding, convolutional and bidirectional LSTM networks, augmented with attention mechanisms and cross-topic feature sharing. In performance evaluation, the proposed model achieved a weighted kappa coefficient of 0.866 and a Pearson correlation coefficient of 0.79 on the test set, significantly better than the comparison model in terms of consistency and correlation. It also showed lower prediction bias in error metrics, demonstrating higher fitting accuracy and rating stability. In three typical application scenarios, the model consistently outperformed the baseline method in terms of off topic detection accuracy and F1 score, and could more effectively capture semantic and structural features. In addition, in terms of resource consumption, the model parameter count was only 16.7M, the inference time was 28.9 ms, and the maximum processor utilisation rate was only 73.6%, demonstrating good deployment adaptability. The experimental results show that this method has good accuracy and generalisation ability in multi-scenario automatic scoring tasks, and has potential value for deployment and application in educational scenarios. Inderscience Publishers - linking academia, business and industry through research

Title: Intelligent English composition scoring method based on transfer paragraph segmentation and Pearson detection

Authors: Haibei Chen

Addresses: School of Culture and Tourism, Guangdong Engineering Polytechnic, Guangzhou, 510520, China

Abstract: With the growing demand for intelligent education, automated English composition scoring has gained significant research attention. Addressing limitations in cross-prompt generalisation and feature extraction, this study proposes a transfer learning-based method incorporating paragraph segmentation and Pearson detection. The approach combines BERT encoding, convolutional and bidirectional LSTM networks, augmented with attention mechanisms and cross-topic feature sharing. In performance evaluation, the proposed model achieved a weighted kappa coefficient of 0.866 and a Pearson correlation coefficient of 0.79 on the test set, significantly better than the comparison model in terms of consistency and correlation. It also showed lower prediction bias in error metrics, demonstrating higher fitting accuracy and rating stability. In three typical application scenarios, the model consistently outperformed the baseline method in terms of off topic detection accuracy and F1 score, and could more effectively capture semantic and structural features. In addition, in terms of resource consumption, the model parameter count was only 16.7M, the inference time was 28.9 ms, and the maximum processor utilisation rate was only 73.6%, demonstrating good deployment adaptability. The experimental results show that this method has good accuracy and generalisation ability in multi-scenario automatic scoring tasks, and has potential value for deployment and application in educational scenarios.

Keywords: English composition scoring; transfer paragraph segmentation; semantic modelling; off topic detection; feature sharing.

DOI: 10.1504/IJCEELL.2026.156733

International Journal of Continuing Engineering Education and Life-Long Learning, 2026 Vol.36 No.13, pp.1 - 25

Accepted: 30 Nov -0001
Published online: 01 Oct 2026 *

Title: Intelligent English composition scoring method based on transfer paragraph segmentation and Pearson detection

Keep up-to-date