178 0

Full metadata record

DC FieldValueLanguage
dc.contributor.advisor최용석-
dc.contributor.author나선웅-
dc.date.accessioned2020-03-17T16:42:04Z-
dc.date.available2020-03-17T16:42:04Z-
dc.date.issued2012-02-
dc.identifier.urihttps://repository.hanyang.ac.kr/handle/20.500.11754/137076-
dc.identifier.urihttp://hanyang.dcollection.net/common/orgView/200000419472en_US
dc.description.abstract본 연구는 학습자의 능력을 측정하는 평가 방법 중 컴퓨터를 활용한 능력측정방법인 CAT(Computerized Adaptive Testing), 그 중에서도 특히 베이지안 네트워크(Bayesian Network) 학습 모델을 도입한 CAT와 해당 CAT에서 사용되는 베이지안 네트워크의 생성에 필요한 네트워크 구조 학습에 대한 방법을 제시한다. 많은 학습 시스템에서 학습자의 수준에 맞는 맞춤학습을 위해서 학습자의 정확한 능력을 측정하는 평가 방법이 필요하다. CAT는 기존 지필고사 방식으로 이루어지던 평가를 컴퓨터로 대체한 것으로 지필고사방식의 문제점을 상당부분 해결하였다. 그러나 CAT에서 주류로 사용되는 문항반응이론식 CAT의 경우 문항 간 연관성에 대해서는 독립을 가정하였지만, 실제 세계에서는 출제되는 문항간의 연관성이 높은 경우가 적지 않아 이 점이 실제 평가시 불필요한 문항의 중복으로 인한 비효율성을 가져올 수 있다. 이에 대한 대안으로 문항간의 연관성을 그래프의 형태로 표현한 베이지안 네트워크를 도입한 CAT가 있다. 이는 평가 문항을 몇 가지의 세부 분야로 분류하여 베이지안 네트워크의 노드에 대입하고, 각 분야 간의 연관성을 노드 사이의 아크에 대입하여 학문의 위계를 베이지안 네트워크로 모델링하여 한 문항의 응답을 통하여 그 문항과 관련된 다른 문항의 이해여부를 추론할 수 있게 하여 결과적으로 문항 간의 연관성이 학습자의 능력 측정에 영향을 미치게 되는 구조이다. CAT에 베이지안 네트워크 모델을 도입하기 위해서는 먼저 문항들에 대응하는 베이지안 네트워크의 구조를 학습시킬 필요가 있다. 본 연구에서는 CAT에 활용하기 위한 베이지안 네트워크 모델의 구조 학습에 특화된 형태의 학습 방법을 제시한다. 각 노드 간의 독립여부를 테스트하여 의존도가 높은 노드 쌍을 무방향 아크로 서로 연결하여 1차적인 네트워크 구조를 형성하고, 아크의 방향을 결정하기 위하여 조건부 확률을 이용하여 상위 노드와 하위 노드의 판별을 수행한다. 상위 노드에서 하위 노드로 이어지도록 아크의 방향을 결정하고 나면 문항 응답 데이터를 훈련 데이터로 사용하여 EM(Expectation Maximization) 알고리즘을 통해 네트워크의 CPT(Conditional Probability Table)을 완성하여 베이지안 네트워크의 구조 학습을 완료하게 된다. 본 연구에서는 실험을 통해 기존의 기법을 사용했을 때에 비해 학습자의 실제 능력치에 수렴하는 속도가 증가함을 보인다. 또한 앞에서 설명한 것과 같은 방법을 사용하여 훈련 데이터로부터 베이지안 네트워크의 구조를 학습시켜, 실제로 존재하는 베이지안 네트워크의 구조와 비교하였으며 기존의 구조학습 방법을 통한 결과를 비교하여 성능향상을 확인하였다.| For the personalized learning according to the level of learners, a good proficiency estimation that measures exact learners' proficiencies is needed. To solve the problem of traditional paper-based exam, computerized adaptive test (CAT) that combines information theory and computer's computational power, has been studied extensively. CAT using Item Response Theory is a good solution to solve the above problem, but it assumes independence about the dependencies between items. If not, it may causes inefficiencies in testing. In this paper, we propose the CAT using Bayesian network that can solve the problem of the CAT using item response theory methods and improve the performance. It assigns question items to each node of Bayesian networks, and estimates a learners' proficiency from a response to question related abilities of the learners. In the experiment, we show that an increase in the rate of convergence by comparing to the conventional well-known approaches. And we present an efficient algorithms for learning Bayesian networks from databases. To measure the learner's ability in CAT (Computerized Adaptive Testing) using Bayesian networks (Bayesian Network), we need to learn Bayesian Network model that discribes the problem domain. In order to use the Bayesian network model in CAT, the learning of Bayesian network structure is necessary: In this study, we propose a method that make structure of the Bayesian network model for the CAT by explaining how specific forms of learning. The CI(Conditional Independence) test that asks whether each node are highly dependent pairs of nodes. And dependent pairs of nodes interconnected by undirected arcs to form the primary network structure. To determine the direction of the arc using the conditional probability, it performs identification to the parent and child nodes. Sub-nodes from the node to we use answer questions as the training data after determinant the direction of the arc. Using EM (Expectation Maximization) algorithm, the CPT (Conditional Probability Table) of the network will be filled, the structure of the Bayesian network will be completed. At the end of this study, we present the experimental results of Bayesian network structure that learned from the training data by using the same method as above, and were compared with a Bayesian network structure that actually exist. The result shows that this algorithm improve performance by comparing to the results that the existing method shows.; For the personalized learning according to the level of learners, a good proficiency estimation that measures exact learners' proficiencies is needed. To solve the problem of traditional paper-based exam, computerized adaptive test (CAT) that combines information theory and computer's computational power, has been studied extensively. CAT using Item Response Theory is a good solution to solve the above problem, but it assumes independence about the dependencies between items. If not, it may causes inefficiencies in testing. In this paper, we propose the CAT using Bayesian network that can solve the problem of the CAT using item response theory methods and improve the performance. It assigns question items to each node of Bayesian networks, and estimates a learners' proficiency from a response to question related abilities of the learners. In the experiment, we show that an increase in the rate of convergence by comparing to the conventional well-known approaches. And we present an efficient algorithms for learning Bayesian networks from databases. To measure the learner's ability in CAT (Computerized Adaptive Testing) using Bayesian networks (Bayesian Network), we need to learn Bayesian Network model that discribes the problem domain. In order to use the Bayesian network model in CAT, the learning of Bayesian network structure is necessary: In this study, we propose a method that make structure of the Bayesian network model for the CAT by explaining how specific forms of learning. The CI(Conditional Independence) test that asks whether each node are highly dependent pairs of nodes. And dependent pairs of nodes interconnected by undirected arcs to form the primary network structure. To determine the direction of the arc using the conditional probability, it performs identification to the parent and child nodes. Sub-nodes from the node to we use answer questions as the training data after determinant the direction of the arc. Using EM (Expectation Maximization) algorithm, the CPT (Conditional Probability Table) of the network will be filled, the structure of the Bayesian network will be completed. At the end of this study, we present the experimental results of Bayesian network structure that learned from the training data by using the same method as above, and were compared with a Bayesian network structure that actually exist. The result shows that this algorithm improve performance by comparing to the results that the existing method shows.-
dc.publisher한양대학교-
dc.title컴퓨터 적응적 검사를 위한 베이지안 네트워크 기법-
dc.title.alternativeBayesian Network Approach to Computerized Adaptive Testing-
dc.typeTheses-
dc.contributor.googleauthor나선웅-
dc.contributor.alternativeauthorNa, Sun Woong-
dc.sector.campusS-
dc.sector.daehak대학원-
dc.sector.department전자컴퓨터통신공학과-
dc.description.degreeMaster-
Appears in Collections:
GRADUATE SCHOOL[S](대학원) > ELECTRONICS AND COMPUTER ENGINEERING(전자컴퓨터통신공학과) > Theses (Master)
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE