Design And Implementation Of Mathematical Natural Language Understanding System Based On Knowledge Graph

The research of artificial intelligence in education is a hot spot in recent years,how to make AI assistant teachers and students is the goal that people pursue.The importance of mathematics as a fundamental science of human science is self-evident.The purpose of the thesis is to establish a system that can extract semantic relations from the title text in a structured form of triples of entity relations.The main work and contributions of the thesis are as follows.1.In the thesis,a mathematical natural language understanding system is proposed,which is based on the features of multiple mathematical entities,short Chinese description,and multi-language mixture,a template matching method based on Bert is proposed,and good results of natural language extraction are obtained in the system.2.The thesis aims at the problems of semantic irrelevant words,the differentiation of synonymous word order,and the unequal distribution of word vectors in the space in the representation of sentence vector features in mathematical subject texts by using Bert,a fusion TF-IDF relation extraction model is proposed.3.In the thesis,the entity hierarchy structure based on the knowledge map makes up the defect of the template matching method.Based on the exploration of entity-level knowledge,a scheme of relation extraction based on syntax pattern is proposed,and the syntax pattern and template matching method are fused.4.With the improvement of data,the thesis uses relation extraction as a classification task based on the Bert-Bi GRU model and compares it with a classical model in two different scale mathematical data sets,recall and the f1-score are the three indicators that give the best performance.Based on the above work,the feasibility test and batch test of the system model is carried out at last.The test data show that the system has the perfect ability for relation extraction,it provides semantic support for the automatic solution of downstream tasks in the natural language understanding system of elementary mathematics.