【作者】 卫芳娜； 甘登文； 丁树良；

【Author】 WEI Fangna Gan Dengwen Ding Shuliang Computer Information Engineering College,Jiangxi Normal University,Nanchang,330022,

【机构】 江西师范大学计算机信息工程学院；

【摘要】 在传统认知诊断计算机化自适应测验(CD_CAT)的研究中,高测量精度一直是选题策略的追求目标,而多少有一点忽视选题速度和题库的安全。香农熵(SHE)方法的优点是模式判准率高,缺点是计算复杂,选题慢,且诊断分类使用DINA模型时,趋向于选择s+g值小的项目,使题库的使用很不均匀。本文针对此提出将题库按项目属性模式分层,并提出了新的选题方法作为对SHE方法的替代。在选题时,先依据最大期望判准率选择"最佳"项目类,在层内再依据s+g函数选择"最佳"项目。模拟采用定长测试,产生线型、收敛型、发散型、无结构型四种属性层级结构的题库。模拟研究显示,新方法在选题速度和题库使用均匀程度上均表现优异,并保有高测量精度。更多还原

【Abstract】 In traditional computerized adaptive testing for cognitive diagnosis,high accuracy is the goal of studying item selection methods.However,the speed of item selection and the security of item pool are more or less neglected.Shannon’s Entropy strategy (SHE) has an advantage in pattern classification accuracy,but with considerable calculations and more skewed item exposure rates. SHE tends to choose the items that have small values in sum of slipping parameter and guessing parameter when DINA model is employed to classify the examinees,thus the security of item pool is difficult to promise.In order to address this problem,this paper attempts to partition an item pool into some sub-pools according to the attribute pattern(briefly,p-STR) contains in items,and proposes a new item selection strategy as an alternative to SHE.The selection procedure contains two steps:first the "best" sub-pool of items based on the expected PMR(Pattern Match Rate)is chosen,and then the "best" item in this sub-pool according to the function based on s+g is selected.The four types of the attribute hierarchies of linear,convergent,divergent and unstructured structure are considered and each type corresponds to an item pool and tests are fixed-length.The results of Monte Carlo simulations indicate that compared to other methods,the new strategy is effective in improvement of the speed of item selection and the security of item pool,meanwhile with high measurement accuracy.更多还原