Font Size: a A A

Research On Automatic Labeling Of Multiple Knowledge Points And Cognitive Verbs In Test Questions Based On Machine Learning

Posted on:2020-08-10Degree:MasterType:Thesis
Country:ChinaCandidate:Z D LvFull Text:PDF
GTID:2428330599964239Subject:Management Science and Engineering
Abstract/Summary:PDF Full Text Request
In the era of internet education,the informatization of education has brought about major changes in education field,meanwhile,many education resources are stored in the form of data.The test data resources are particularly important.However,the prevailing situation is that these test data only contain the content and answers of the test questions,and do not directly reveal the objectives examined by the test questions,which means missing cognitive verbs and knowledge points.Therefore,it is a pretty meaningful research content to find a way to automatically label the test points and the cognitive verb level for the test questions in order to play the role of test data resources better.Based on this purpose,the research task of this paper is mainly divided into two parts,which are the knowledge point annotation and the cognitive verb annotation for the mathematics test data.(1)This paper proposes a multi-knowledge point labeling method based on integrated learning.In the aspect of the test point annotation,this paper firstly defines the problem of question point labeling in a formal way,and transforms the problem of test question multi-knowledge point labeling into a multi-label classification problem;constructs the knowledge graph of the knowledge point by means of the teaching material catalogue and domain knowledge,which is used as the knowledge point labeling system of the test questions,and the original test knowledge point data is replaced by the knowledge graph of knowledge point;The ensemble learning method based on SVM for multi-knowledge point labeling is proposed,by setting the sub-set accuracy threshold,some excellent base classifiers are selected for achieving better labeling results.(2)This paper proposes a cognitive verb annotation method based on data enhancement for mathematics questions.First is to construct a deep text generation model using the test text data to generate a large amount of data for different cognitive verbs,so as to make up for the imbalance problem in the original data.Second is based on the active learning sample query strategy,using the original data to train an SVM classifier based on OVR strategy,which is used for predict the test questions after the preliminary screening.Then the distances from the hyperplane to the six cognitive categories of each test are obtained,and the difference between the maximum and the second largest value of the distances are calculated.In the end,the top m samples are selected in ascending order.Third is to use the selected data and 80% of the original data as the training set,to train the TextCNN model which also used to the prediction of the remaining 20% of the original data.In this paper,the researches about the mathematical knowledge points and cognitive verb labeling are fully studied,and two different labeling methods are proposed,which are better than traditional machine learning labeling methods to some extent.
Keywords/Search Tags:Educational Data Mining, Knowledge Points Labeling, Cognitive Verbs Labeling, Ensemble Learning, Deep Learning
PDF Full Text Request
Related items