CAJ | 학술논문

根据现有的研究使用神经网络进行认知诊断的方法，改变神经网络的建模方式，实现基于神经网络的认知诊断计算机化自适应测验。使用 Monte Carlo 方法生成模拟数据，训练并测试神经网络，在此基础上将神经网络用于自适应测试中，知识状态估计采用极大似然估计，选题策略采用后验加权的 KL 信息量指标。模拟结果表明，基于 BP 神经网络的 CD-CAT 的判准率较为理想，并且因为神经网络的训练过程可以在小样本情况下实现，这种 CD-CAT 更适合在课堂评估中使用。
근거현유적연구사용신경망락진행인지진단적방법，개변신경망락적건모방식，실현기우신경망락적인지진단계산궤화자괄응측험。사용 Monte Carlo 방법생성모의수거，훈련병측시신경망락，재차기출상장신경망락용우자괄응측시중，지식상태고계채용겁대사연고계，선제책략채용후험가권적 KL 신식량지표。모의결과표명，기우 BP 신경망락적 CD-CAT 적판준솔교위이상，병차인위신경망락적훈련과정가이재소양본정황하실현，저충 CD-CAT 경괄합재과당평고중사용。
Current neural network-based cognitive diagnosis was altered to realize cognitive diagnostic computerized adaptive testing.Monte Carlo method was used to generate datasets.The BP network was tralned through tralning datasets.Prediction accuracy of tralned network is computed with testing datasets. Such tralned and tested BP networks were used to realize CD-CAT,with maximum likelihood method to estimate knowledge state and posterior weighted kullback leibler (PWKL)information as the item selection strategy.Classification accuracy for respondents was computed and the outcome was found to be desirable.