베이지언 네트워크에 기반한 인지진단모형의 영어 읽기 시험에 대한 적용 (Application of bayesian network-based cognitive diagnostic modeling to small sample English reading comprehension test data.)

서지정보

· 발행기관 : 대한언어학회
· 수록지 정보 : 언어학 / 31권 / 4호 / 83 ~ 111페이지
· 저자명 : 최세일

초록

Cognitive diagnostic models (CDMs), a family of classification models developed to provide fine-grained diagnostic information for learning and teaching in education, have increasingly been used in language testing. However, most of the previous CDM studies in language testing have mainly been conducted based on large samples from professional testing agencies.
This trend makes it difficult for practitioners to apply the models in classroom assessment contexts for which the models were originally developed. Realizing this limitation, researchers working in CDMs have recently begun to turn their attention to the conditions in which CDMs can work for classroom assessments, especially with small sample sizes. Bayesian networks (BN) provide an efficient and intuitive framework for modeling complex systems of observable or latent variables and have been extensively employed in the data science as well as in intelligent tutoring systems for modeling students’ learning progress. The framework has also huge potential to be well suited for diagnostic modeling of students’ learning in classroom contexts. This study was to examine whether BN can be applied in cognitive diagnostic modeling for classroom assessments. After constructing a set of small test data (N=100, 150, 200) from a large real test, the study applied a BN-based CDM model to the data sets and compared with conventional CDMs its item parameter and attribute classification recovery. The results show that the BN-based CDMs yielded uniformly better estimates in all testing conditions than the conventional methods. The study then discusses its implications for the CDM applications in language testing.

영어초록

Cognitive diagnostic models (CDMs), a family of classification models developed to provide fine-grained diagnostic information for learning and teaching in education, have increasingly been used in language testing. However, most of the previous CDM studies in language testing have mainly been conducted based on large samples from professional testing agencies.
This trend makes it difficult for practitioners to apply the models in classroom assessment contexts for which the models were originally developed. Realizing this limitation, researchers working in CDMs have recently begun to turn their attention to the conditions in which CDMs can work for classroom assessments, especially with small sample sizes. Bayesian networks (BN) provide an efficient and intuitive framework for modeling complex systems of observable or latent variables and have been extensively employed in the data science as well as in intelligent tutoring systems for modeling students’ learning progress. The framework has also huge potential to be well suited for diagnostic modeling of students’ learning in classroom contexts. This study was to examine whether BN can be applied in cognitive diagnostic modeling for classroom assessments. After constructing a set of small test data (N=100, 150, 200) from a large real test, the study applied a BN-based CDM model to the data sets and compared with conventional CDMs its item parameter and attribute classification recovery. The results show that the BN-based CDMs yielded uniformly better estimates in all testing conditions than the conventional methods. The study then discusses its implications for the CDM applications in language testing.

참고자료

· 없음