Scholarlink

<p>The English reading comprehension ability has long been measured using text responses and subjective assessment. Nevertheless, as multimodal Machine Learning (ML) develops, there is an increasing possibility to improve precision and depth of comprehension measurement by combining various data capture methods, including eye-tracking, voice, and facial expressions and interaction behavior. The research proposes a novel assessment framework that utilizes an ML model to process multimodal inputs and evaluate students’ reading comprehension in a more holistic and data-driven manner. The system captures user data through sensors and multimedia interaction platforms as learners engage with English reading materials. Key comprehension skills assessed include reading fluency, critical thinking, reading strategies, and vocabulary recognition. The collected data is preprocessed using a median filter to denoise both video and audio inputs. Features such as reading time, gaze fixation, vocal clarity, emotional response, and textual annotations are extracted and analyzed using a Convolutional Neural Network (CNN) to derive individual comprehension scores. A Multi-Model Beetle Antennae Search Weighted Random Forest (MBAS-WRF) algorithm is employed to effectively interpret the multimodal data and assess English reading ability. Results indicate that the proposed multimodal approach significantly improves assessment accuracy compared to traditional evaluation methods. It achieves 629.7 GFLOPs computational cost, 1224.4 ms latency, 82.5% accuracy, 0.88 Pearson correlation, and 0.48 RMSE, demonstrating that multimodal ML can effectively support objective and the assessment of English reading comprehension. It indicates that English teachers should adopt technology-based assessment plans that capture cognitive, emotional, and behavioral reactions of learners in reading exercises.</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

English reading comprehension ability assessment based on multimodal machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月28日

查看全文
  • Qian Liu

摘要

The English reading comprehension ability has long been measured using text responses and subjective assessment. Nevertheless, as multimodal Machine Learning (ML) develops, there is an increasing possibility to improve precision and depth of comprehension measurement by combining various data capture methods, including eye-tracking, voice, and facial expressions and interaction behavior. The research proposes a novel assessment framework that utilizes an ML model to process multimodal inputs and evaluate students’ reading comprehension in a more holistic and data-driven manner. The system captures user data through sensors and multimedia interaction platforms as learners engage with English reading materials. Key comprehension skills assessed include reading fluency, critical thinking, reading strategies, and vocabulary recognition. The collected data is preprocessed using a median filter to denoise both video and audio inputs. Features such as reading time, gaze fixation, vocal clarity, emotional response, and textual annotations are extracted and analyzed using a Convolutional Neural Network (CNN) to derive individual comprehension scores. A Multi-Model Beetle Antennae Search Weighted Random Forest (MBAS-WRF) algorithm is employed to effectively interpret the multimodal data and assess English reading ability. Results indicate that the proposed multimodal approach significantly improves assessment accuracy compared to traditional evaluation methods. It achieves 629.7 GFLOPs computational cost, 1224.4 ms latency, 82.5% accuracy, 0.88 Pearson correlation, and 0.48 RMSE, demonstrating that multimodal ML can effectively support objective and the assessment of English reading comprehension. It indicates that English teachers should adopt technology-based assessment plans that capture cognitive, emotional, and behavioral reactions of learners in reading exercises.