Scholarlink

<p>Increased demand for scalable and objective assessment in education has fueled interest in AI-driven automatic scoring systems. This paper presents a new approach to enhancing Bidirectional Encoder Representations from Transformers (BERT) with modified version of Tailor Optimization Algorithm (MTOA) for automatic essay scoring in college English teaching. Traditional grading is subjective, time-consuming, and variable, posing enormous challenges in mass educational settings. In order to overcome these limitations, the improved contextual intelligence of BERT has been employed, and its performance has been enhanced with MTOA, which optimizes hyperparameters like learning rate, batch size, and model depth in an intelligent manner. The proposed system is evaluated on the ASAP Dataset, a standard student essay corpus, and achieves significant improvements over the evaluation metrics like Quadratic Weighted Kappa (QWK), Pearson Correlation, and Mean Absolute Error (MAE). Outcomes show that maximized BERT outperforms baseline models including vanilla BERT, LSTM, and GPT-2 by achieving a QWK score of 0.85 and minimizing MAE to 0.38. The system also delivers rich student feedback that focuses on areas for improvement in terms of grammar, coherence, and style. This research underscores the potential for deep learning combined with metaheuristic optimization to turn educational testing into a scalable, accurate, and interpretable automatic scoring model.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research on automatic scoring algorithm of college english teaching based BERT architecture optimized by modified tailor optimizer

  • 期刊论文

  • 发表日期：

    2026年7月24日

查看全文
  • Li Jing,
  • Xiaoyan Huang

摘要

Increased demand for scalable and objective assessment in education has fueled interest in AI-driven automatic scoring systems. This paper presents a new approach to enhancing Bidirectional Encoder Representations from Transformers (BERT) with modified version of Tailor Optimization Algorithm (MTOA) for automatic essay scoring in college English teaching. Traditional grading is subjective, time-consuming, and variable, posing enormous challenges in mass educational settings. In order to overcome these limitations, the improved contextual intelligence of BERT has been employed, and its performance has been enhanced with MTOA, which optimizes hyperparameters like learning rate, batch size, and model depth in an intelligent manner. The proposed system is evaluated on the ASAP Dataset, a standard student essay corpus, and achieves significant improvements over the evaluation metrics like Quadratic Weighted Kappa (QWK), Pearson Correlation, and Mean Absolute Error (MAE). Outcomes show that maximized BERT outperforms baseline models including vanilla BERT, LSTM, and GPT-2 by achieving a QWK score of 0.85 and minimizing MAE to 0.38. The system also delivers rich student feedback that focuses on areas for improvement in terms of grammar, coherence, and style. This research underscores the potential for deep learning combined with metaheuristic optimization to turn educational testing into a scalable, accurate, and interpretable automatic scoring model.