摘要
规则空间模型是一种高效的知识结构诊断模型,但较高的规则空间构造代价阻碍了在小规模、实时认知诊断中的应用.为了提高规则空间模型的可扩展性,提出使用近似子图生成理想属性模式集进而压缩规则空间的方法.近似子图能够通过忽略和测试项目无关的属性降低子图规模量级,从而有效缩减理想属性模式集规模,达到压缩规则空间的目的;同时通过构建顶点间的虚拟边模拟领域知识图上的传递依赖关系,使近似子图在不引入额外属性的前提下保持领域知识图上的依赖关系,实现对不合理属性模式的有效过滤.在此基础上,给出了构造规则空间所需的近似子图构造算法以及由近似子图生成理想属性模式集的方法.最后在标准测试集上开展了近似子图与依赖保持子图和顶点导出子图两种方法的性能对比实验,并将近似子图应用于实际教学认知诊断中验证其诊断准确率,实验结果表明近似子图能够在不损失诊断结果准确率的前提下显著压缩规则空间,降低规则空间模型应用于小规模、实时诊断的门槛.
Rule-space model is an effective method to diagnose knowledge structure of subjects. But high time cost of rule-space construction is a major obstacle to its application in the small and real time cognitive diagnosis. In order to improve the scalability of rule-space model, an algorithm is proposed to compress rule space by constructing an ideal attribute pattern set using the approximate subgraph. Approximate subgraph can decrease the order of subgraph by ignoring attributes unrelated to testing items so as to reduce the scale of ideal attribute set effectively, then the rule space can be compressed accordingly. At the same time, virtual edges between vertexes are constructed to simulate transitive dependencies on the domain knowledge graph, and these virtual edges preserve indirect dependences on the domain knowledge graph without introducing additional attributes, which ensures effective filteration of unreasonable attribute pattern. On this basis, the approximate graph construction algorithms and ideal attribute pattern generation algorithm supported by approximate subgraph are given for rule space construction. Finally, experiments for comparing the performance among approximate subgraph, dependency preserving subgraph and vertex derived subgraph on the benchmark dataset were carried out, followed by an application in practical teaching cognitive to verify the diagnosis accuracy of approximate subgraph based method. It is concluded that approximate subgraph compresses rule space remarkably without loss of diagnosis accuracy and makes it possible to apply rule space model to small and realtime cognitive diagnosis.
作者
黄宏涛
梁存良
李大鹏
叶海智
HUANG Hong-Tao;LIANG Cun-Liang;LI Da-Peng;YE Hai-Zhi(Henan Provincial Engineering Research Center for Analysis and Application of Educational Big Data, Henan Normal University, Xinxiang 453007;College of Telecommunications and Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210003)
出处
《自动化学报》
EI
CSCD
北大核心
2019年第8期1586-1598,共13页
Acta Automatica Sinica
基金
教育部人文社会科学研究项目(16YJC880017)
河南省哲学社会科学规划项目(2016BJY010)
江苏省自然科学基金(BK20161518)
河南省高等教育教学改革研究与实践重点项目(2017SJGLX042)
河南省教育厅科学技术研究重点项目(15A880010)资助~~
关键词
规则空间
近似子图
依赖保持
压缩
理想属性模式
Rule space
approximate subgraph
dependency preserving
compression
ideal attribute pattern
