面向慕课学习者评估的认知反应模型  

Cognitive and Response Model for Evaluation of MOOC Learners

作　　者：王炼红[1] 罗志辉 刘畅 WANG Lian-hong;LUO Zhi-hui;LIU Chang(School of Electrical and Information Engineering,Hunan University,Changsha,Hunan 410082,China;R&D Center,Agricultural Bank of China,Tianjin 300392,China)

机构地区：[1]湖南大学电气与信息工程学院,湖南长沙410082 [2]中国农业银行研发中心,天津300392

出　　处：《电子学报》2023年第1期18-25,共8页Acta Electronica Sinica

基　　金：国家重点研发计划(No.2019YFE0105300);中国高等教育学会数字化课程资源专项研究课题(No.21SZYB15)。

摘　　要：认知诊断模型从学习者的认知结构出发,建模学习者与试题之间的潜在关系,结合智能算法并根据试题作答结果可评估学习者的知识水平.大多数认知诊断模型是将学习者的高阶能力特征视为单维,忽视了后天努力的影响.为此,本文提出了一种考虑能力特征与努力特征相互补偿的具有二维高阶特征的新认知诊断模型——认知反应模型(Cognitive and Response Model,C&RM).该模型通过设置能力特征与努力特征相互补偿机制来融合两高阶特征参数以精准建模学习者的知识水平.同时,还构建了知识点弱项特征参数,以综合考虑学习者的知识水平与不同知识点对作答试题的影响,进一步提高模型的解释性和预测精度.采用自建的HNU_SYS数据集和Math1,Math2,FrcSub公共数据集,通过实验对比分析了C&RM模型、最新的认知诊断模型和经典诊断模型.当数据训练集为70%最佳比例时,C&RM在4个数据集上分别比次优方法提升了6.3%,4.3%,3.3%,5.2%,其预测性能最佳,验证了本文模型的可行性和有效性.The cognitive diagnosis model starts from the learner’s cognitive structure, models the potential relationship between the learner and the test questions, and combines intelligent algorithms to evaluate the learner’s knowledge level according to the results of the test questions. Most cognitive diagnostic models treat learners’ higher-order ability characteristics as a single dimension, ignoring the effect of acquired effort. To this end, this paper proposes a cognitive diagnostic model with two-dimensional high-order features that considers the mutual compensation of ability and effort features—cognitive and response model(C&RM). The model integrates two high-order feature parameters by setting the mutual compensation mechanism of ability feature and effort feature to accurately model the knowledge level of the learner. At the same time, the characteristic parameters of knowledge point weaknesses are also constructed to comprehensively consider the knowledge point level of learners and the influence of different knowledge points on answering questions, and further improve the interpretability and prediction accuracy of the model. Using the self-built HNU_SYS data set and the Math1,Math2, FrcSub public data sets, the C&RM model, the latest cognitive diagnostic model and the classic diagnostic model are compared and analyzed through experiments. When the data training set is at the best ratio of 70%, C&RM is improved by 6.3%, 4.3%, 3.3%, and 5.2% on the four data sets, respectively, and its prediction performance is the best, which verifies the feasibility of the model in this paper.

关 键 词：认知诊断 认知反应模型 评估 慕课 补偿机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]