基于项目加工过程的多级评分认知诊断模型开发及其CD-CAT研究

Cognitive diagnosis (CD) is a method and theory of diagnostic assessment integrating the characteristics of cognitive psychology and psychometrics. Its primary purposes are to reveal the internal processing routines of individual, diagnose on its process, and provide fine-grained information feedback about learner’s test performance such that individual learners’ strengths and weaknesses can be identified to facilitate instruction on users. To realize the purposes, an efficient and appropriate model, the tool to realize diagnosis, is very important and indispensible. . Recently, the data set, analyzed by most of current cognitive diagnostic models(CDMs), are dichotomous, it reflects the right or wrong responses to items of individuals. Certainly, it might provide the information about the knowledge states of the individuals a lot. However, a lot of important information is also lost. For example, when a wrong response of one item is reported, does it mean that some or all the skills that the item measures are not mastered by the individual? Or does it mean that the skills are mastered by him/her mildly? Very few information on such conditions could be inferring from the responses. Thus polytomous responses are necessary and more valuable to be reported. .However, only few existing models are supposed to assess the polytomous data. This project focuses on polytomous data assessment under cognitive diagnosis framework, which will be done integrating many subjects such as psychometric, statistic modeling and mathematic interfering. Its purposes are as follows:. (1) To develop some new polytomous cognitive diagnostic models (CDMs) which are available with polytomous data based on item response processing. The works are expected to fill the gap of the conventional dichotomous CDMs. .(2) To construct statistic algorithms and improve the design for Q-matrix based on polytomous CDMs, which would provide theoretical guidance and technology support for the compiling of polytomous cognitive diagnostic test and classify all subjects. .(3) To build the framework of the polytomous computerized adaptive testing for cognitive diagnosis (CD-CAT), which could make assessment adaptively and provide more abundant and diagnostic information efficiently and effectively. . In this project, the MCMC algorithm is used to realize the parameters estimation, and Monte Carlo simulation method and real data analysis are both used to testify the rationality and feasibility of new models, algorithms, methods and technologies. All these are expected to provide the method supports on promoting the efficiency and classify accuracy and expanding the application fields.

认知诊断是认知心理学与心理计量学相结合的产物，其目的是探讨个体内部的心理加工机制，诊断个体的认知优势与不足，为个体的因材施教及有针对性地开展补救提供服务。本项目拟对当前具有重要理论及应用价值的多级评分认知诊断开展研究，采用心理测量、统计建模、数学论证多学科交叉研究范式：一方面开发出可以处理多级评分的基于项目加工过程的认知诊断模型（即多级评分CDMs），以弥补传统认知诊断模型的不足；另一方面构建并完善多级评分认知诊断Q矩阵设计及其算法，以充分实现对各类被试的诊断分类，并为多级评分的认知诊断测验编制提供理论指导及技术支持；最后构建全新的基于多级评分的CD-CAT，从而实现高效准确快速地对被试诊断分类并提供更为丰富和更具诊断价值的信息。总之，本项目拟弥补当前认知诊断及其CD-CAT研究的不足，为提高认知诊断的准确率、诊断效率以及为拓展认知诊断在实际中应用提供方法学支持。

认知诊断是认知心理学与心理计量学相结合的产物，其目的是探讨个体内部的心理加工机制，诊断个体的认知优势与不足，为个体的因材施教及有针对性地开展补救提供服务。本项目对当前具有重要理论及应用价值的多级评分认知诊断开展研究，采用心理测量、统计建模、数学论证多学科交叉研究范式：一方面开发出可以处理多级评分的基于项目加工过程的认知诊断模型（即多级评分CDMs），以弥补传统认知诊断模型的不足；另一方面构建并完善多级评分认知诊断Q矩阵设计及其算法，以充分实现对各类被试的诊断分类，并为多级评分的认知诊断测验编制提供理论指导及技术支持。. 本项目共发表SCI/SSCI以及CSSCI期刊论文30篇，其中SCI/SSCI期刊论文11篇，《心理学报》3篇，《心理科学》和《心理科学进展》16篇。发表的学术论文获得江西省教育科学优秀成果奖一等奖和三等奖各1项。在国际学术会议（如 NCME）上发表并报告学术论文 2 篇；国内会议（如中国心理学大会）发表并报告学术论文 10 余篇。培养研究生共20余名，其中博士研究生4人（2人已顺利毕业），硕士研究生17人（其中10人顺利毕业）。近4年研究中，项目主持人及项目组成员较好地完成了项目原定任务，另有一些研究突破了原申报书原定计划。