CD-CAT选题策略测验精度和题库使用均匀性的关系研究

发布时间：2018-05-08 13:28

本文选题：认知诊断计算机自适应测验 + 选题策略　；参考：《江西师范大学》2017年硕士论文

【摘要】：选题策略是CD-CAT中非常关键的一个环节,需要在保证效率的同时尽可能提高测量精度。这也是CD-CAT中选题策略一直以来的发展主题。国内外学者对测验精度和题库利用率的关系探讨停留在理论层面。本文从实证层面探讨测验精度和题库利用率的关系,为选择合适的选题策略提供参考,并为今后的相关研究和认知诊断计算机自适应测验的发展提供指导。为了解决上述问题,本文采用蒙特卡洛(Monte Carlo)模拟方法并以DINA模型为例展开一系列的研究,包括一个预研究和四个正式实验:预研究采用文献分析法,选取了多篇同时含有测量精度和题库使用情况等数据的论文,探讨两者之间的关系。正式研究采用模拟研究法,考察KL、SHE、PWKL、PPWKL、MPWKL、HKL、PHKL七种选题策略的题库使用均匀性和测验精度的关系,并考察不同测验长度、属性数量、被试的分布、被试数量等因素对测验精度以及题库使用均匀性的影响。本文主要结论如下:(1)MPWKL和SHE的测量精度较高,如果综合测验精度和题库的利用率,PHKL、PPWKL比较好,其次是MPWKL。(2)总体上,综合测验精度和题库的使用均匀性指标,PHKL和PPWKL处在同一水平,为第一梯队,测验精度中等,题库使用率最高,综合表现最好;HKL和PWKL处在同一水平,为第二梯队,测验精度中等,题库使用率一般;SHE和MPWKL为第三梯队,测验精度最高,题库使用率最差;KL为第四梯队,测验精度低,题库使用率一般。(3)测验长度与测验精度成正比,与题库使用均匀性成反比。测验长度为16时,各种选题策略的综合表现最好。(4)被试数量对选题策略的测验精度和题库利用率影响较小。随着被试数量的增加,选题策略的测验精度先上升,被试数量达到1000后,变化平缓,题库使用均匀性综合指标先下降,被试数量为1000后,变化平缓。被试数量为1000时综合表现最好。(5)属性个数与选题策略的测验精度成反比。属性个数越少,各选题策略综合表现越好。(6)被试知识状态分布形态为负偏态分布时,测验精度最高。被试为正态分布时,SHE、PWKL、MPWKL、PHKL等选题策略的题库使用率最高。
[Abstract]:The topic selection strategy is a key link in CD-CAT. It is necessary to improve the measurement accuracy while ensuring the efficiency. This is also a topic of development in the CD-CAT topic selection strategy. Scholars at home and abroad discuss the relationship between test accuracy and question bank utilization rate at the theoretical level. This paper discusses the relationship between test accuracy and question bank utilization from the empirical level, and provides a reference for selecting appropriate selection strategies, and provides guidance for future research and the development of computer adaptive test for cognitive diagnosis. In order to solve the above problems, this paper uses Monte Carlo Monte Carlo (Monte Carlo) simulation method and takes DINA model as an example to carry out a series of studies, including a pre-study and four formal experiments. The relationship between measurement accuracy and question bank data is discussed in this paper. The formal study uses the simulation method to investigate the relationship between the uniformity of the use of questions and the test accuracy in the question bank of the seven strategies for selecting the topic, including the KLX / PWKL / PPWKL / MPWKL / MPWKL / HKLHKL, and the different test lengths, the number of attributes, and the distribution of the subjects, The effect of factors such as the number of subjects on the accuracy of the test and the uniformity of the use of the question bank. The main conclusions of this paper are as follows: the accuracy of SHE and MPWKL are higher. If the accuracy of comprehensive test and the utilization ratio of item bank are better than that of MPWKL.2) on the whole, the accuracy of comprehensive test and the index of evenness of use of test bank are at the same level as PPWKL. For the first echelon, the test accuracy is medium, the use rate of question bank is the highest, the comprehensive performance of HKL and PWKL is at the same level, is the second echelon, the test precision is medium, the rate of question bank utilization is generally the third echelon of she and MPWKL, and the test precision is the highest. The lowest utilization rate of question bank was in the fourth echelon. The accuracy of test was low, and the test length was directly proportional to the test accuracy and inversely proportional to the uniformity of question bank use. The length of the test is 16:00, and the comprehensive performance of the various selection strategies is the best. 4) the number of subjects has little effect on the test accuracy and the utilization ratio of the question bank. With the increase of the number of subjects, the test accuracy of the topic selection strategy increased first. After the number of subjects reached 1000, the change was gentle, the comprehensive index of using uniformity of question bank first decreased, and after the number of subjects reached 1000, the change was gentle. The number of attributes is inversely proportional to the test accuracy of the strategy when the number of subjects is 1000. The smaller the number of attributes, the better the comprehensive performance of each topic selection strategy.) when the distribution of knowledge state is negative skewness distribution, the test accuracy is the highest. The question bank utilization rate was the highest when the subjects were normal distribution, such as SHEPWKL / MPWKL / PHKL.
【学位授予单位】：江西师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：B842

【相似文献】