东篱科研大数据发现系统（DRDS）

位置：成果数据库 > 期刊 > 期刊详情页

神经网络和支持向量机在认知诊断中的应用

ISSN号：1671-6981
期刊名称：《心理科学》
时间：0
分类：TP18[自动化与计算机技术—控制科学与工程;自动化与计算机技术—控制理论与控制工程]
作者机构：[1]江西师范大学计算机信皂工程学院,南昌330022, [2]江西师范大学初等教育学院,南昌330022, [3]江西省地质矿产勘查开发局,南昌330002, [4]江西省赣州市信丰县劳动局,信丰341600
相关基金：国家自然科学基金（31500909,31360237,31160203,30860084）、全国教育科学规划教育部重点课题（DHA150285）、教育部人文社科项目（13YJC880060）、江西省社会科学研究“十二五”（2012年）规划项目（12JY07）、江西省教育科学2013年度一般课题（13YB032）、江西省教育厅科技计划项目（GJJ13207）、国家留学基金委资助项目（201509470001）和江西师范大学青年成长基金和博士启动基金的资助.

关键词：认知诊断评估, 概率神经网络, 支持向量机, 诊断分类, 项目属性标定, cognitive diagnosis, probabilistic neural network, support vector machines, diagnosis classification, item attribute identification

中文摘要：

认知诊断评估的主要问题是如何准确进行被试分类和项目属性标定。本文使用概率神经网络（PNN）和支持向量机（SVM）进行被试分类和属性标定，重点讨论PNN用于诊断的理论根据。模拟研究表明：PNN方法表现最好，训练速度快且具有很好判准率和标定准确率；PNN与GDD方法在分类上表现相当，在独立结构下PNN更好；线性SVM具有较好判准率和标定准确率。软计算中此类方法可非常方便推广至多级评分测验数据分析。

英文摘要：

Soft computing is a term used in computer science to refer to problems whose solution is unpredictable and uncertain. With the development of society, people are not only satisfied with the achievement test that only gives a total score, but also hope to get more information about the examinees＇ cognitive states from the examinees＇ item responses, and to make inferences .about students＇ cognitive strengths and weaknesses to provide remedy. How to collect specification of item attributes and classification of student＇s states is critical to cognitive diagnosis. Gierl ＆ Cui （2007） applied Multi-Layer perception （MLP） neural network in the classification of cognitive diagnosis. Shu et al. （2013） also used MLP result as the precursor to the attribute profile to increase the accuracy of item parameter estimates. This article not only employs the other artificial neural networks （probabilistic neural networks, PNN）, but also employs SVM （linear and non-linear SVM） for the classification of attribute profile and aiding specification of Q-matrix. It mainly discusses the reasonability of PNN for the classification of attribute profile. These methods are applied to the 0, 1 scoring test, and then comparisons of the classification results are made with some of the typical methods of attribute hierarchy model： method A （Leighton, Gierl, Hunka, 2004） and generalized distance discrimination （GDD）. Furthermore, the Monte Carlo simulation study is used to examine the performance of these methods. In the simulation study, the pattern match ratio and average attribute match ratio are used as criteria to evaluate the classification accuracy of different approaches. Under four attribute hierarchical structures （linear, convergent, divergent, unstructured） used in the attribute hierarchy method of Leighton et al. （2004） and another structure with independent attributes, five kinds of Q-matrix with 6 attributes were simulated individually. Under each type of Q-matrix, we used four kind

同期刊论文项目