在线理解以自然语言编写的数字内容对于生活的许多方面都至关重要,包括学习,专业任务,和决策。然而,面对理解困难会对学习成果产生负面影响,批判性思维能力,决策,错误率,和生产力。本文介绍了一种创新的方法来预测本地内容级别的理解困难(例如,段落)。使用负担得起的可穿戴设备,我们从自主神经系统非侵入性地获得生理反应,特别是脉搏率变异性,和皮肤电活动。此外,我们整合了来自经济高效的眼动仪的数据。我们的机器学习算法识别对应于高认知负荷的内容和区域内的“热点”。这些热点代表了理解困难的实时预测因素。通过将生理数据与上下文信息(例如个人的经验水平)集成,我们的方法的准确率为72.11%±2.21,准确率为0.77,召回率为0.70,f1得分为0.73.这项研究为开发智能,认知意识接口。这样的接口可以提供即时的上下文支持,减轻内容中的理解挑战。无论是通过翻译,内容生成,或使用可用的大型语言模型进行内容摘要,这种方法有可能增强语言理解。
Comprehending digital content written in natural language online is vital for many aspects of life, including learning, professional tasks, and decision-making. However, facing comprehension difficulties can have negative consequences for learning outcomes, critical thinking skills, decision-making, error rate, and productivity. This paper introduces an innovative approach to predict comprehension difficulties at the local content level (e.g., paragraphs). Using affordable wearable devices, we acquire physiological responses non-intrusively from the autonomous nervous system, specifically pulse rate variability, and electrodermal activity. Additionally, we integrate data from a cost-effective eye-tracker. Our machine learning algorithms identify \'hotspots\' within the content and regions corresponding to a high cognitive load. These hotspots represent real-time predictors of comprehension difficulties. By integrating physiological data with contextual information (such as the levels of experience of individuals), our approach achieves an accuracy of 72.11% ± 2.21, a precision of 0.77, a recall of 0.70, and an f1 score of 0.73. This study opens possibilities for developing intelligent, cognitive-aware interfaces. Such interfaces can provide immediate contextual support, mitigating comprehension challenges within content. Whether through translation, content generation, or content summarization using available Large Language Models, this approach has the potential to enhance language comprehension.