基于眼动大数据的学生学习状态画像及其应用研究

Physiological data based learning state analysis is a challenge to learning analytics. Eye movement data is an important type of physiological data and can objectively reflect cognitive process of human being. Commercial eye trackers are expensive and hard to be widely used. Recently by only using webcam, convolutional neural network based eye tracking algorithm can obtain eye movement data with considerable precision. .In this project, we plan to collect eye movement data from real education environment in long term at large scale. We will do research on learning state profiling based on big eye movement data, obtain customized data representations of students that reflect their learning behavior, and finally achieve learning monitoring, learning diagnosis and learning intervening. This project helps to improve education quality in multimedia classroom, smart classroom and online learning. The main research contents of this project are as follows (1) Data representations, clustering and time series analysis methods for individual and group eye movement data; (2) learner real time learning state profile, time dependent learning state profile, discipline dependent learning state profile, location dependent learning state profile based on real time and historical eye movement data ; (3) Eye movement data based learning state visualization models and diagnosis tools that help teachers to identify absent minded learners, monitor and warn learners, find learning interests, and mine learning habits.

基于生理数据的学习状态研究是当前学习分析的现实问题和挑战。眼动数据是重要的生理数据，客观反映了人的认知活动。商用眼动仪价格高昂，难以规模性使用。近年来基于卷积神经网络的眼球跟踪算法取得突破，仅使用普通摄像头可获得精度可观的眼动数据。.本项目在真实教学环境中长时间大规模采集学习者眼动数据。构建基于眼动大数据的学习状态画像，研究学习者的个性化数据表征、分析其学习行为，实现学习监控、学习诊断、学习干预，在多媒体教室、智慧教室、在线学习等场景提升教学质量。本项目研究内容（1）学习者个体、群体眼动数据的数据表征、聚类、时序分析方法；（2）研究实时、历史眼动数据，构建学习者实时学习状态画像、不同时间段学习状态画像、不同学科学习状态画像、不同座位学习状态画像；（3）提出基于眼动大数据的学习状态可视化模型、诊断工具，帮助教学者发现学习者信息迷航、走神、实时监控预警学习者、分析学习兴趣、挖掘学习习惯等。

本项目基于眼动数据研究慕课、测试两个学习场景，开发了GazeMOOC和GazeQuiz可视分析系统。GazeMOOC通过聚类、时序分析等方法根据实时、历史眼动数据生成个体、群体在不同时间段、不同学科中的学习状态画像，实现学习监控。GazeMOOC通过散点图、堆叠区域图、条带图、堆叠条形图等可视化组件直观展现个体、群体的学习状态画像。GazeQuiz可通过配置动态载入测试内容，通过凝视时间、路径、转换关系等构建个体眼动行为模型，通过平均凝视时间、平均回视次数等构建群体眼动行为模型。GazeQuiz通过转换图、箱体图等可视化方法展现在测试过程中个体和群体的认知过程，实现学习诊断。为提高基于眼动交互的UI界面用户体验，提出了GazeGrid多尺度多层级网格界面的交互新方法，并给出最佳参数选择方案。通过注意力机制和数据增强技术提高视线估计精度，提出了GazeAttentionNet等方法。基于面部特征点的区域级数据增强方法在MPIIFaceGaze视线估计任务中超越了现有SOTA算法。

内容获取失败，请点击重试