用学习分析提升教学质量：基于大数据的形成性评价和动态学生画像研究

Considering China’s higher education characteristics, i.e. high student-teacher ratio and high capacity classes, this project develops and refines an efficient sampling tool for students’ learning activities data, which covers full learning cycle including activities pre-, during, and after class, based on cell phone and under partially flipped classroom or blended learning pedagogy. With the help of this tool, the formative teaching and learning assessment can be carried out with valid data sampled in the main teaching and learning place, i.e. classroom, in core courses, many curriculums, and various universities. In this way, students’ learning outputs can be understood in a fine-grained way. Therefore, the more intelligent decision making can be realized for learning, teaching, and management. Besides, the obtained learning behavior and input data of students can be combined with the learning output data to form the learning big data with the help of the previous mentioned tool. These educational big data can be mined dynamically. And therefore patterns can be recognized. So students time scheduling, in-and-out class activities, learning habits, learning efficiencies, learning methodologies, and learning effects can be depicted from various aspects, which means more profound dynamic student learning profiles can be acquired. The comprehensive dynamic student profiles help realize self-regulated learning and teaching-based-on-learning, and thus can facilitate a new way to promote higher education’s quality improvement with Chinese characteristics.

本项目结合我国高等教育生师比高、大班授课为主的国情，在部分翻转课堂或混合式学习场景下，基于手机这种自带设备，研制并完善覆盖课前、课上和课后的有效全周期学习数据采集工具。在量大面广的课程中、在各专业培养方案中、在各种类型学校中，在教育教学的主战场—课堂—中获取鲜活准确的学习成效数据，实现形成性评价，从更细的粒度了解学生的学习效果，为学、教、管三方面提供决策支撑。除学习评价外，还综合考虑前述工具中获取的学生学习行为和学习投入，构成学习大数据，开展动态数据挖掘和模式识别研究，从时间安排、学习习惯、学习效率、学习方法、学习效果等不同维度对学生学习投入产出进行描绘，刻画出丰满完整的动态学生学习画像，有助于实现学生的个性化学习和教师的因材施教，从根本上走出一条有中国特色的高等教育质量提升之路。

本项目依次完成了数据采集、数据特征量评估、数据挖掘分类算法的比较和选择、课程的形成性评价及其可视化等工作，并在不同课程、专业和学校中进行了推广。项目通过数据清洗整合了多源异构数据中的原始特征，结合专家知识设计高层特征，分析学生学习活动与学习成效的关联关系；将特征凝聚与域自适应相结合，提出了特征提取的新方法，在此基础上，提出了两种时间窗算法，实验结果表明，在学期过半时进行预测，已经可以实现一个相对较高的预测正确率；将模型应用到了不同教师、不同学科的课程上，结果表明在预测成绩靠后的学生时，算法可以进行有效的泛化，在不同课程上均能取得较好的预测效果；设计开发了学生成绩分析预测系统。该系统旨在快速、自动化地完成对影响学生成绩关键因素的分析以及对学生成绩的预测，已实现跨课程、跨专业、跨高校的应用。