基于SPOC学习行为分析的学生成长监测及适应性学习支持研究

In MOOC related research field, many researchers analyzed students learning behavior based on the logging data and tried to predict students’ performance, improve the course design, and conduct the teaching intervention. Blended teaching mode is gaining attention in the SPOC, online learning management system (LMS) collects all the students' interactions with it. But how could we extract meaningful information from them? And how could we improve the learning outcomes of a SPOC?.This proposal aims at how to design the learning monitoring model based on students' learning and interaction behavior. Furthermore, we will focus on how to find the exercise related resources and build a personalized mechanism to recommend sequence of exercises in the SPOC..Existing Knowledge Tracing (KT) models need large scale training sets, and fail to pinpoint exactly which concepts a student is good at or unfamiliar with. We propose a KT model based on multi-block Memory-Augmented Neural Networks, which incorporates individual learning behavior. This improved model could automatically discover underlying concepts of exercises and depict the changing knowledge state of a student..On the other side, we will design an important resource discovery method which is driven by behavioral data, in order to promote the personalized teaching in different curriculum effectively. In addition, a personalized exercises will be sequenced to learners based on collaborative filtering and reinforcement learning, which will be constructed using learning knowledge level and correlation weight between exercises and concepts. We will conduct teaching interventions in our SPOC iteratively..The achievements of the study will provide theoretical basis and technical support for applying educational data mining method in SPOC personalized teaching.

教育数据挖掘技术被广泛应用于MOOC教学中的辍学预警、课程设计改进及教学干预等方面。随着更多校园课程转向混合化在线教学模式，课程设计者能够获取学生全周期在线学习行为数据，如何从丰富的数据中提取有价值的信息是提升SPOC教学质量的关键。本项目拟利用在线学习行为数据，设计学生成长监测模型，并实现问题相关重要资源及适应性答题序列的推荐机制。为解决现有知识追踪模型需要大规模训练样本，且隐藏在模型中的题目与知识点关联及知识点掌握程度信息无法被有效利用的问题，提出基于多区块记忆增强神经网络的模型设计，并将个体学习行为差异纳入模型考量。为在不同课程中应用个性化教学推荐，设计并实现由行为数据驱动的问题相关重要资源发现方法。此外，利用个体知识水平及知识点-题目关联信息构建融合协同过滤的强化学习答题序列推荐机制。项目研究成果将在校园教学中进行迭代检验，为在SPOC领域开展适应性学习支持提供理论基础和实践经验。

知识追踪（Knowledge Tracing，KT）模型应用于校园课程追踪学生知识掌握情况时，因SPOC课程难以获得大规模训练样本，传统RNN模型难以达到理想预测效果。而如何获取知识追踪模型所发现的关联关系及学习者知识掌握程度的变化轨迹，是探索面向SPOC学生成长监测和适应性学习支持机制的关键。.本项目为提升知识追踪模型准确率，引入学习行为量化特征及在线学习活动时间间隔信息；利用记忆增强网络（Memory Augmented Neural Network，MANN）的外部存储机制，对DKT模型进行扩展，将隐含信息转移至相应的外部记忆单元，以实现对网络隐含信息的读取与解释；在改进深度知识追踪模型的基础上，基于模型评估的学生知识状态，结合从外部记忆区块提取的关联信息，实现学生学习成长监测系统。.此外，将DKVMN模型融合学习行为建模，可在学习过程中实时评估学习者的知识状态。基于贪心算法、协同过滤和强化学习设计并实现了相应个性化学习路径推荐系统，评价依托于模拟实验及实际课程运行分组实验。模拟实验中共包含两个知识追踪模型，分别用于评估虚拟学生的知识状态，和模拟学生对于推荐题目的回答结果、完成题目后知识状态的更新。模拟实验结果表明，贪心算法和强化学习算法的效果较好，协同过滤算法的效果和固定题目顺序持平。通过模拟实验，发现使用题目前置关系图可以在维持答题路径推荐算法表现的情况下有效减少耗时。SPOC课程中进行了线上随机对照实验，结果显示采用贪心算法来推荐题目的学生平均知识掌握程度显著高于对照组，而采用协同过滤和强化学习进行题目推荐时学生的平均知识掌握程度和对照组没有显著差异。.项目研究成果在校园教学中进行了迭代检验，为在SPOC领域开展适应性学习支持提供理论基础和实践经验。