基于认知诊断和神经网络的试题得分预测-《计算机技术与发展》

文章信息/Info

Title:: Prediction of Question Score Based on Cognitive Diagnosis Model andNeural Network

作者:: 史浩杰¹ ; 李幸² ; 贾俊铖¹ ; 匡健¹ ; 那幸仪¹; 1. 苏州大学计算机科学与技术学院,江苏苏州 215006;
2. Momenta-初速度(苏州)科技有限公司,江苏苏州 215100

Author(s):: SHI Hao-jie1 ; LI Xing2 ; JIA Jun-cheng1 ; KUANG Jian1 ; NA Xing-yi1; 1. School of Computer Science and Technology,Soochow University,Suzhou 215006,China;
2. Momenta ( Suzhou) Technology Company Limited,Suzhou 215100,China

Keywords:: score prediction; objective questions; subjective questions; cognitive diagnosis; neural network

摘要:: 随着教育越来越走向信息化,大量的教育数据会被保存,在海量教育数据中挖掘出学生的潜在信息是智能教育中非常值得研究的问题之一。针对目前大多数得分预测都是预测一个总分,无法具体到每一题得分预测的问题,对考试中存在的主要题型进行了研究,结合现有试题得分预测方法,指出其优势与不足,提出基于认知诊断和神经网络分别预测客观题和主观题得分的方法。该方法结合认知诊断理论计算出学生的知识状态,通过矩阵算法来预测学生在每道客观题上的得分,再将学生的知识状态作为特征,学生得分作为训练标签,使用卷积神经网络来训练并且预测学生在每道主观题上的得分。将两种方法分别与传统方法进行比较,结果表明两种方法分别在客观题和主观题上的效果比传统方法提高了很多。

Abstract:: As education becomes more and more informatized,a large amount of education data will be preserved. Mining the potentialinformation of students from the massive amount of education data is one of the issues worthy of research in intelligent education. In viewof most of the current score predictions that predict a total score,which cannot be specific to the score prediction? ?of each question,themain question types existing in the test are analyzed. Combined with the existing test question score prediction methods,we point outtheir advantages and disadvantages and put forward a method to predict the scores of objective and subjective questions respectively basedon cognitive diagnosis and neural network. This method combines the cognitive diagnosis theory to calculate the student’s knowledgestate,predicts the student’s score on each objective question through a matrix algorithm,trains and predicts student’s scores on eachsubjective question by convolutional neural network,with the student’s knowledge state as a feature and the student’s score as a traininglabel. Comparing the two methods with the traditional methods,the results show that the effects of the two methods on objective and subjective questions are much higher than the traditional methods.