Research And Implementation Of Knowledges Predictor Based On Graph Neural Network

In the field of education,people are increasingly realizing the importance of artificial intelligence in teaching activities.Therefore,applying AI technology to optimize teaching activities,mine educational data,and customize personalized teaching plans has positive significance for the development of education.The goal of this thesis is to obtain the mapping relationship between question stem text and knowledge points based on given data such as question stem text and corresponding knowledge points as labels,so as to automatically determine the corresponding knowledge points for input,which can be regarded as a text multi label automatic classification task in a vertical domain.The main research content is as follows:(1)Proposed a mathematical language text automatic annotation method based on regular expressionsDue to the disorganized knowledge points in the original data,standardization processing is needed.This thesis proposes an automatic annotation method based on regular expression matching to identify scattered knowledge points in text and classify them.(2)Proposed a Transformer based automatic word segmentation method for Chinese mathematical languageThis thesis proposes using the Transformer word segmentation algorithm to segment the input Chinese mathematical text.A word segmentation table is constructed for the unique symbols in the mathematical field,which is added to the word segmentation machine for pre training,enabling the word segmentation machine to effectively recognize these symbols.(3)A mathematical knowledge point multi label automatic classification model based on the combination of graph convolutional network and knowledge representation network is proposed.This thesis is based on the idea of representation learning and the word embedding method of deep learning,and trains a knowledge representation network to generate representation sentence vectors through question stem text.By utilizing the idea of graph convolutional networks and word vector embedding training,word vectors representing knowledge point categories and their interrelationships are trained.Use this vector to automatically classify knowledge points in the question stem text.Based on the above ideas,a mathematical knowledge point prediction system was designed and implemented,solving the problem of combining model representation learning methods with graph convolutional networks for training.Finally,the model was trained on samples containing 60000 basic mathematical problems and a comparative experiment was designed.The experimental results show that the basic mathematical problem knowledge point prediction system designed in this thesis can effectively integrate the text features and mathematical logic features of mathematical texts,improve classification accuracy,and better meet the needs of intelligent question banks and automatic problem-solving.It has good scalability and certain practical value.