关键词: 认知诊断, Q矩阵修正, DINA模型, g 法

Abstract: Recently more and more attentions were paid on cognitive diagnosis (CD). The recognition of Q matrix was the basis and proposition of CD, and it was the only information media about cognitive attributes and test items. Only if it was recognized correctly, then the following cognitive analysis might be reliable.
This paper developed a new modification method of Q matrix (called g method) based on DINA model. The method could detect and modify the mistakes on Q matrix that could make the Q matrix rationally and promote the correct match ratio of CD.
Monte Carlo simulation method was use here, and some comparisons with western similar studies were done. Findings showed:
(1) Under any response slip probability (5%, 10%, 15%) context, when the critical value of parameters s and g was set to be 0.2, 0.25 or 0.3, g method could modify the mistakes of Q matrix efficiently. When Q matrix was recognized correctly, no modification will be done by g method. These indicated that g method could work well on recognition and modification on Q matrix whether it has mistakes or not.
(2) Compared to the similar western studies, the correct modification ratio of g method was relatively great and similar with the consequences of d method suggested by de la Tarre (2008). But g method was simpler than d method.
(3) The g method could not only rectify the mistakes of Q matrix efficiently but also promote the correct match ratio of CD. Especially the correct match ratio of PMR increased greater than 40% which was a wonderful work.

Key words: cognitive diagnosis, Q-matrix validation, DINA model, the g method