Using the Bayesian Network’s Structural Learning Algorithm to Estimate the Q-Matrix in Cognitive Diagnosis Models

In cognitive diagnostic assessment, a Q matrix that associates each item with cognitive attributes in a test is necessary to derive the knowledge states of students. Defining the Q matrix is the most fundamental step in cognitive diagnosis. The conventional method for calibrating cognitive attributes relies mainly on the subjective judgments of experts. Herein, a method that uses a structural learning algorithm in a Bayesian network (BN) to estimate the Q matrix is proposed. Both the viability and efficacy of the suggested approach are examined by running simulations and conducting analyses on real data. The outcomes of the simulations indicate that the proposed approach mostly exceeds the performance of the available methods, and its advantages stem from not only better estimation accuracy but also its computational efficiency. The efficiency of the proposed approach is also verified via real data analysis. Compared with the stepwise method, the model data fits the estimated Q matrix via the BN method for the empirical data is more satisfactory, and the Root Mean Square Error of Approximation (RMSEA) is lower. Consequently, this Q-matrix estimation method based on a BN can improve accuracy and computational efficiency and promote the practice of cognitive diagnosis assessment.