Some practical item selection algorithms in cognitive diagnostic computerized adaptive testing -- smart diagnosis for smart learning

Abstract

The current studies represent an effort to advance the feasibility of cognitive diagnostic computerized adaptive testing (CD-CAT), an intelligent educational measurement tool that was envisioned as enhancing individualized learning over twenty years ago. Several new selection algorithms are proposed for addressing two important issues in CD-CAT: measurement efficiency and item exposure control. The posterior-weighted CDM discrimination index (PWCDI) and posterior-weighted attribute-level CDM discrimination index (PWACDI) are computationally affordable and highly efficient alternatives to other information index-based algorithms. The binary stratification algorithm offers an elegant solution to item exposure control in both fixed-length and variable-length CD-CAT, compared with the restrictive stochastic methods for fixed-length CD-CAT and SHTVOR for variable-length CD-CAT.