| Cognitive diagnostic models have received research attention from numerous national and foreign scholars in recent years,which aims to classify examinees into different latent classes,and each of these categories has a unique pattern of attribute mastery that indicates the examinee’s mastery of the knowledge attribute or skill.The existing research literature shows that numerous cognitive diagnostic models have been developed for application to diagnostic tests in different measurement contexts.Cognitive diagnostic models can be divided into two categories based on the scoring of the items that can be handled: 0-1 scoring and polytomous scoring cognitive diagnostic models.The Bayesian networks consists of a directed acyclic graph and a corresponding set of joint conditional probability distributions,providing a convenient and intuitive framework structure for representing causal relationships and are therefore well suited for modeling the content of educational assessments in diagnostic tests.In recent years,Bayesian networks have been widely used in the field of artificial intelligence,but have received relatively little attention in the field of psychology,and most of the existing studies have applied Bayesian networks to 0-1 scoring cognitive diagnostic tests.This study combines Bayesian networks with polytomous scoring diagnostic tests,which has important theoretical and practical implications.To verify the effectiveness of Bayesian networks classification in polytomous scoring cognitive diagnostic tests,three studies were conducted in this research.Study 1 compared the classification performance of the polytomous scoring diagnostic model S-GDINA with the Bayesian network model in terms of the accuracy of correct classification of examinees when exploring the existence of Q-matrices correctly defined by experts.The results of the study show that the classification performance of the Naive Bayesian Classifier is comparable to that of the S-GDINA model,which can achieve equally good classification performance,and the classification performance of the Tree Augmented Naive Bayesian Classifier can also achieve good classification performance.Study 2 explores the classification performance of the polytomous scoring diagnostic model S-GDINA compared with the Bayesian networks classification model for examinees when the expert-defined Q matrix contains partial errors.The results showed that the classification accuracy of both the S-GDINA model and the Bayesian network decreased when partial error q-vectors were randomly added to the correct Q-matrix.However,under such experimental conditions,the classification of the Naive Bayesian Classifier outperformed that of the S-GDINA model,especially when there was large noise in the Q matrix(g,s=0.15,r=0.3);the difference between the correct classification rate of the Tree Augmented Naive Bayesian Classifier and that of the S-DINA model became smaller as the sample size of examinees increased.This indicates that the Bayesian Network Classifier is a better choice to deal with when the test data is a poor fit for the chosen diagnostic model,the Q matrix of the test contains errors,or the test data contains a high amount of noise.Study 3 used Bayesian Network Classification model to classify and analyze the empirical data to demonstrate the classification process of Bayesian Network Classification model in empirical data.The results of the study showed that both the Naive Bayesian Classifier and the Tree Augmented Naive Bayesian Classifier had better classification results in the polytomous scoring cognitive diagnostic test.The classification consistency of the Naive Bayesian Classifier reached 81.8%,and the accuracy of the individual attributes of the examinees reached about 73%;the classification consistency of the Tree Augmented Naive Bayesian Classifier reached63.3%,and the accuracy of the attributes measured by the questions reached about 60%,indicating that Bayesian Networks can be well applied to polytomous scoring cognitive test data. |
PDF Full Text Request