Q-matrix learning and DINA model parameter estimation

The DINA model is one of the most widely used models in cognitive and skills diagnosis, and several algorithms have been developed for estimating the model parameters. However, since the parameter space is very large and has a mix of binary variables, even medium-sized testing is extremely challenging. To make the model practical, a fast optimization algorithm for parameter estimation is needed. In this study, we converted the deterministic Q-matrix learning problem into a Boolean matrix factorization (BMF) problem and developed a recursive algorithm to find an approximate solution while solving the uncertainty parameters analytically using maximum likelihood estimation (MLE). We proved that the MLE is equivalent to the minimum information entropy of the DINA model. Simulation results demonstrated that our proposed algorithm converges rapidly to the optimal solution under suitable initial values of skill - item association and is insensitive to the initial values of the uncertainty parameters.

[1]  Kenneth R. Koedinger,et al.  A Machine Learning Approach for Automatic Student Model Discovery , 2011, EDM.

[2]  Yi Sun,et al.  Alternating Recursive Method for Q-matrix Learning , 2014, EDM.

[3]  Edward H. Haertel.  Using restricted latent class models to map the skill structure of achievement items , 1989 .

[4]  Václav Snásel,et al.  Implementing Boolean Matrix Factorization , 2008, ICANN.

[5]  Yi Sun,et al.  Maximum likelihood estimation based DINA model and Q-matrix learning , 2014, 2014 International Conference on Behavioral, Economic, and Socio-Cultural Computing (BESC2014).

[6]  Michel C. Desmarais,et al.  Mapping question items to skills with non-negative matrix factorization , 2012, SKDD.

[7]  Kenneth R. Koedinger,et al.  Automated Student Model Improvement , 2012, EDM.

[8]  J. Templin,et al.  The Effects of Q-Matrix Misspecification on Parameter Estimates and Classification Accuracy in the DINA Model , 2008 .

[9]  Vilém Vychodil,et al.  Discovery of optimal factors in binary data via a novel method of matrix decomposition , 2010, J. Comput. Syst. Sci..

[10]  Jaideep Vaidya,et al.  Boolean Matrix Decomposition Problem: Theory, Variations and Applications to Data Engineering , 2012, 2012 IEEE 28th International Conference on Data Engineering.

[11]  Pauli Miettinen,et al.  The Discrete Basis Problem , 2006, IEEE Transactions on Knowledge and Data Engineering.

[12]  Jingchen Liu,et al.  Data-Driven Learning of Q-Matrix , 2012, Applied psychological measurement.

[13]  Yi Sun,et al.  Improved algorithms for exact and approximate boolean matrix decomposition , 2015, 2015 IEEE International Conference on Data Science and Advanced Analytics (DSAA).

[14]  Jonathan Templin,et al.  Diagnostic Measurement: Theory, Methods, and Applications , 2010 .

[15]  Jimmy de la Torre,et al.  An Empirically Based Method of Q‐Matrix Validation for the DINA Model: Development and Applications , 2008 .