Wu Zihe, Zhang Ye
Faculty of Education, Universiti Malaya, 50603, Kuala Lumpur, Kuala Lumpur, Malaysia.
Sci Rep. 2025 Jul 2;15(1):22737. doi: 10.1038/s41598-025-08182-0.
This study aims to explore and construct an evaluation system of English language teaching effect combining data visualization analysis and intelligent text analysis technology. The study is based on the limitations of traditional English teaching evaluation methods, that is, relying too much on teachers' subjective judgment and lacking real-time monitoring, which cannot meet the needs of personalized teaching. In this study, data visualization technology is used to transform students' learning data into intuitive charts, and intelligent text analysis technology is used to deeply explore students' writing and oral performance in order to achieve accurate teaching decisions. The research methods include emotion analysis with Bidirectional Encoder Representations from Transformers (BERT) model, theme modeling with Latent Dirichlet Allocation (LDA) model, semantic analysis with Word2Vec model, and model optimization with ensemble learning methods such as Bagging, XGBoost and Stacking. The experimental results show that the accuracy of Stacking model in training set and testing set is 95.0% and 94.3% respectively, which is significantly better than other single models. In addition, the optimization model combining BERT, Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN) and self-attention mechanism also shows significant advantages in emotional analysis and teaching content analysis. This study not only provides a new idea for the development of educational evaluation methods, but also a powerful tool for realizing personalized and accurate teaching.
本研究旨在探索并构建一个结合数据可视化分析与智能文本分析技术的英语教学效果评估体系。该研究基于传统英语教学评价方法的局限性，即过于依赖教师的主观判断且缺乏实时监测，无法满足个性化教学的需求。在本研究中，数据可视化技术被用于将学生的学习数据转化为直观的图表，智能文本分析技术则被用于深入探究学生的写作和口语表现，以实现精准的教学决策。研究方法包括使用双向编码器表征变换器（BERT）模型进行情感分析、使用潜在狄利克雷分配（LDA）模型进行主题建模、使用词向量模型（Word2Vec）进行语义分析，以及使用Bagging、XGBoost和Stacking等集成学习方法进行模型优化。实验结果表明，Stacking模型在训练集和测试集上的准确率分别为95.0%和94.3%，显著优于其他单一模型。此外，结合BERT、长短期记忆网络（LSTM）、卷积神经网络（CNN）和自注意力机制的优化模型在情感分析和教学内容分析方面也显示出显著优势。本研究不仅为教育评价方法的发展提供了新思路，也为实现个性化精准教学提供了有力工具。
