| With the development of information technology and the speed of knowledge update in the era of big data,people have higher motivations for learning.Tests are not only used to select talent,but also to enhance students’ memory,deepen their understanding and improve their transfer learning.The development of learning-oriented assessments is in line with the needs of a new generation of test,as they are more likely to facilitate students’ learning and personal growth.At the same time,cognitive diagnostic assessments stand out from other assessment methods.The basic process of cognitive diagnostic assessment design includes the selection of test items and the selection or development of cognitive diagnostic models.Therefore,this paper focuses on the core of "learning-oriented" to conduct a study on model selection and item selection in cognitive diagnostic assessment.In Study 1,we conducted a comparative study of the generalization performance of longitudinal cognitive diagnostic models for learning-oriented assessment.Longitudinal cognitive diagnostic models for learning assessment have developed rapidly in recent years.To investigate the ability of longitudinal cognitive diagnostic models to adapt to fresh data,the generalization performance of three longitudinal cognitive diagnostic models on different types of longitudinal data is investigated.The first longitudinal cognitive diagnostic model is Patt-DINA,a model of latent transition analysis at the attribute pattern level.The second is the Att-DINA,a model of latent transition analysis at the attribute level.And the sLong-DINA model is based on higher-order latent structures.The performance of these three models was evaluated with the correct classification rates of attribute and pattern of students’ knowledge states,the absolute model fit index and the relative model fit index.The results of the simulation study showed that the Att-DINA model and the sLong-DINA model are more advantageous in most conditions,which means that their generalization performance is relatively better.The Patt-DINA model is less advantageous due to the larger number of parameters to be estimated,but the model still has advantages when the sample size is large and it can estimate transition probabilities of knowledge states with more space for variation.In addition,the study also introduces empirical data to validate the above findings.In Study 2,we conducted a study on learning-oriented test construction and focused on proposing a Bayesian extension for cognitive diagnostic test construction.Constructing high-quality tests has been a concern for researchers.Currently available test construction methods are mainly based on greedy algorithms to ensure the optimization of the objective function.For example,test construction methods based on Cognitive Diagnostic Index(CDI)and Attribute-level Discrimination Index(ADI),as well as their improvements MCDI(Modified Cognitive Diagnostic Index)Diagnostic Index)and MADI(Modified Attribute-Level Discrimination Index).However,the use of these methods tends to fall into local optimal solutions and fails to improve the quality of test construction in general.Maximize the minimum distance between classes(MMD)test construction method based on Mixed-integer linear programming(MILP)solves these problems,but it has not been applied to cognitive diagnostics with attribute structure of cognitive diagnosis.Therefore,in this paper,the MMD method is extended with Bayesian extensions while relaxing the constraint function in MILP to reduce the computation time,considering the priori information of attribute structure.This study is based on two cognitive diagnostic models with six attribute structures for the simulation study.In addition,this paper also conducts an empirical study using data from the Certificate of English Proficiency Examination(ECPE).The results show that the improved MMD method outperforms the existing method in terms of classification accuracy and Kullback-Leibler distance for shorter test lengths. |
PDF Full Text Request