Automatic Labeling Of Junior Middle School Mathematics Knowledge Points Integrating Domain Ontology

With the rapid development of artificial intelligence technology and the advent of the era of big data,in the face of the massive knowledge resources in the education field,how to skillfully combine related technologies in the field of natural language processing to better serve students,teachers and educational institutions is a current Research hotspots.Specific to the field of mathematics,intelligent test paper composition,automatic problem solving,etc.are the current research directions of the natural language processing technology and mathematics field.The process of intelligent test paper selection needs to select a suitable question type from the test question database for random combination according to the knowledge points examined,and the process of automatic problem solving often requires matching different types of test questions to the corresponding problem solving template.Therefore,this paper focuses on the basic task of automatic labeling of junior high school math knowledge points,and deeply studies the related theories and technologies of automatic labeling,aiming at the lack of inherent semantic correlation in junior high school math test questions,semantic sparseness and ambiguity caused by mathematical entities,and the characteristics of mathematical formulas Extraction and other difficulties,put forward the fusion of domain ontology and noise reduction autoencoder network to build an automatic knowledge point annotation model.This paper has achieved the following important results:(1)Domain ontology representation method based on template matchingAiming at the problem of data sparseness and ambiguity caused by a large number of mathematical entities derived from domain ontology in mathematical test questions,this paper proposes a recognition method based on template matching on the basis of studying the expression characteristics of mathematical test questions in junior high school,which can quickly identify the entities of mathematical test questions Make substitutions and retain the affiliation,so as to achieve a unified expression of the mathematical ontology and generate an intermediate representation of the mathematical test questions.Experiments show that by constructing an intermediate representation of the fusion domain ontology for mathematical test questions,it is possible to achieve a 2%to 4%improvement in multiple deep learning models.(2)Word vector representation method based on attention mechanismIn order to solve the problem that the existing math knowledge point labeling methods cannot fully exploit the characteristics of mathematics disciplines and ignore the inherent semantic correlation of the word vector representation process of mathematics questions,this paper proposes to use the noise reduction self-encoding network based on attention mechanism to learn the rich semantics of mathematics questions themselves Information,generating dynamic word vector representations based on context.In the process of word vector representation training,this paper proposes to use mathematical region formula loss,important concept labeling and other strategies to optimize the learning process of the model,which can obtain richer semantic information.Experiments show that,based on the intermediate representation of mathematical test questions,the use of word vector representation methods based on the attention mechanism can achieve an effect improvement of about 4%.(3)Based on M-Bert model of junior high school mathematics knowledge points automatic marking systemBased on the proposed M-Bert model,this paper builds an end-to-end junior high school math knowledge points automatic annotation system,designs and implements text normalization,intermediate representation,model training and other important modules,which can realize the automatic annotation of math test knowledge points And from the perspective of optimizing the training data set,a human-computer collaboration strategy is proposed to continuously optimize the recognition effect of the knowledge point automatic system.Finally,the M-Bert model proposed in this paper is significantly improved compared to the deep learning model based on recurrent neural network.It can basically achieve 82%accuracy on the junior high school math text data set built by itself,which can be better.Meet the needs of the current junior middle school mathematics text knowledge points automatically marked.