베이지안 네트워크를 적용한 컴퓨터 기반 적응적 평가

초록

For the personalized learning corresponding with level of learners, a good estimation method which can exactly measure proficiency of the learners is needed. The existing estimation methods, such as paper-based exam, are inefficient because they need many time and cost in order to estimate proficiency of the learner. To overcome such problem, computerized adaptive testing (CAT) that combines information theory and computational power of the computer has been extensively studied. Item Response Theory (IRT)-based CAT is a good solution which can relax the problem of traditional estimation methods; however it assumes that all items are mutually independent between the items. If not, it may cause inefficiencies in testing. In this paper, we propose a novel CAT using Bayesian networks that can solve the problem of IRT-based CAT and improve the performance of estimation. It assigns question items to each node of the bayesian networks, and estimates proficiency of the learner from a response to question related to abilities of the learner. In the experiment, we show that our novel CAT using bayesian networks can effectively improve the ratio of convergence required to estimate exact proficiency of the learner as compared with the conventionally well-known CAT approaches.

키워드