基于机器学习与眼动追踪的认知风格模型构建

Construction of Cognitive Style Model Based on Machine Learning and Eye Tracking

认知风格反映了学生接近、获取、组织、处理和解释信息的模式,可用来解释和指导学生的行为。将认知风格集成到智能系统中,有助于开发个性化的用户模型,推动智能教育发展。当前有关认知风格自动分类的研究较为匮乏,尚未将机器学习与眼动追踪技术联合起来进行应用。基于机器学习与眼动追踪的认知风格模型,选取注视时长、注视点数量、扫视时长、眼跳次数、眼跳距离与瞳孔直径等6个与认知有着密切关系的眼动指标,归一化处理后借助机器学习算法进行认知风格自动分类。实验结果表明:在进行同样时长的视频学习时,不同场认知风格的学习者表现出不同的视觉行为模式;场依存型学习者注视点较为分散,表现出有较多的扫视行为、较少的注视行为、较长的眼跳距离与较大的瞳孔直径变化,信息搜索效率较低;而场独立型学习者有着较为密集与定向的视觉注意模式,信息搜索效率更高。对5种机器学习算法进行性能对比后发现,逻辑回归算法的分类效果最好,准确率达到89.01%,Kappa值达到0.774。该认知风格自动化分类模型既可用于智能学习系统的课程资源优化设计,也可用于个性化学习路径的推荐。未来可整合更多生理数据,通过不同模态数据之间的信息互补,提升数据分析的准确性以及对学习者认知能力评估的可靠性。

Cognitive style reflects the patterns of students’approaching,acquiring,organizing,processing and interpreting information,and can be used to explain and guide students’behavior.Integrating cognitive style into intelligent systems can help develop personalized user models and promote the development of intelligent education.Currently,there is a lack of research on automatic classification of cognitive styles,and machine learning has not yet been combined with eye tracking technology for application in this research field.Based on machine learning and eye tracking,a cognitive style model is built.It selects six eye movement indicators closely related to cognition,including fixation duration,number of fixations,saccade duration,number of saccades,saccade distance,and pupil diameter.After normalization,machine learning algorithms are used to automatically classify cognitive styles.The experimental results show that learners with different field cognitive styles exhibit different visual behavior patterns when learning from videos of the same duration.Field-dependent learners have a more dispersed fixation points,showing more saccades,fewer fixations,longer saccade distances,and larger pupil diameter changes,resulting in lower information search efficiency.Field-independent learners have a more intensive and directional visual attention pattern,resulting in higher information search efficiency.After comparing the performance of five machine learning algorithms,it is found that the classification effect of logistic regression algorithm is the best,with an accuracy rate of 89.01%and a Kappa value of 0.774.This automatic classification model of cognitive styles can be used for optimizing the design of course resources in intelligent learning systems and for recommending personalized learning paths.In the future,more physiological data can be integrated to improve the accuracy of data analysis and the reliability of learner cognitive ability assessment through information complementarity between different modalities.