大规模开放课程数据中因果关系挖掘方法研究

The focus of MOOC data mining is the correlation between learning effects and learning behaviors. In contrast to causality, the conclusion reached by the correlation analysis of MOOC data is that it is difficult to provide decision support and intervention in a given context. Therefore, the proposal intends to adopt some novel technology to carry out causality extraction and empirical testing methods from MOOC data. The research contents include: .1..Three key problems of causality mining from MOOC data, which have high-dimensional, massive and sparse features, will be solved such as causal diagram construction, consistency identification and hidden variable discovery; .2..Tools and software of causality mining and analysis will be developed. Then verify the effective of our proposed approach in this proposal on MOOC platform...This study extends the research on correlation analysis of educational data to causality mining and benefit to promote the development of educational data mining. The research results of the proposal help to reveal the valuable rules and patterns of learning in large-scale online learning...Expected results include: publishing more than 15 high-level academic papers; applying for 3 to 5 national invention patents; and developing software of causality mining for MOOC data.

当前，MOOC数据挖掘关注的重点是学习效果与学习行为间的相关性。与因果关系相比，相关性分析得到的结论很难提供决策支持和干预手段。为此，本项目拟围绕MOOC数据，对因果关系挖掘开展理论方法研究与实证测试。内容包括：一、针对MOOC数据高维、海量、稀疏特点，重点解决因果图构建、因果图一致性检验以及隐变量发现三项关键问题；二、研制面向MOOC数据的因果关系挖掘与分析工具软件，依托Coursera等MOOC平台数据，对所提方法进行测试、验证与完善。本项目将教育数据相关性分析的研究延伸到因果关系挖掘，有助于促进教育数据挖掘领域的发展，揭示大规模在线学习中有价值的学习规律和模式，为在MOOC平台上构建智能导学提供理论与关键技术支撑。. 预期成果：发表高水平学术论文15篇以上；申请国家发明专利3-5项；研制面向MOOC数据的因果关系挖掘与分析软件。

学习者在MOOC学习过程中产生了大量行为数据，同时MOOC平台也提供了大量学习资源数据以及学习者基本属性数据。对这些海量MOOC数据进行统计分析，将有助于揭示人类更深层次的认知机理，挖掘有价值的学习规律与模式。本项目重点围绕MOOC数据，对因果关系挖掘开展理论方法研究与实证测试。内容包括：一、针对MOOC数据高维、海量、稀疏特点，重点解决MOOC数据辍学行为因果图构建方法研究；二、研制面向MOOC数据的辍学行为因果关系挖掘与分析工具软件。提出了MOOC数据中辍学行为因果关系挖掘研究框架。提出两阶段的因果图构建方法，最后将本研究提出MOOC日志的因果关系挖掘方法整合到智慧教育示范应用系统中，验证了模型的有效性。通过分析自动构建的MOOC数据因果图以及专家的指导，我们找到了两个成因的学习行为：1）课程论坛寻求帮助行为；2）自主做课后习题行为。围绕上述研究成果，我们在主流期刊会议上发表20余篇论文，申请国家发明专利3项，软著2项，培养研究生10人。研究成果还将对智能导学、个性化推荐和学习评价提供辅助决策的依据。