Reasoning System Based On Expression Parsing And Matching

Over the years,artificial intelligence technology based on big data and deep neural networks has made great progress in computer vision,speech processing,natural language processing,game agents,and content generation.These successes continue to inspire the scientific research community to explore new applications scenes for artificial intelligence technology.Logical thinking ability is one of the most important human’s abilities to promote the development of science.It is of great significance to explore the application of new technologies of artificial intelligence in deductive reasoning.As one of the important topics of artificial intelligence,automatic reasoning technology mainly aims to realize the automatic theorem proving or test question solving,and its realization requires a higher level of abstraction ability and logical reasoning ability.In addition,in solving mathematical test questions,the questions are given in the form of a mixture of natural language and mathematical language,which puts forward higher requirements for the natural language understanding and symbol processing capabilities of the automatic reasoning system.This thesis mainly studies the extraction and formalization of topic information in the automatic solution system of elementary mathematics test questions,as well as the matching and substitution of theorems after formalization.Generally speaking,this thesis contains the following three parts:1.Design and implementation of a system for parsing and type recognition of elementary mathematical expressions.This thesis is based on relevant theories such as formal language,automata,and compiler principles.We have developed an easy-toextend and efficient system for parsing elementary mathematical expressions,including lexical analyzer,grammar analyzer,expression tree generator,and other components.We have also defined a grammar and corresponding semantic rules covering basic elementary mathematical expressions.2.Implementation of matching and substitution algorithms on expression trees.Based on the syntax analysis,this thesis has developed matching algorithms for expressions under different constraints.This algorithm can be used to achieve a series of basic reasoning functions,such as expression type determination,theorem substitution,and expression transformation.It can work with natural language processing modules to perform tasks such as named entity recognition.Additionally,it can serve as a basic component of the reasoning module to transform known information in the knowledge graph.3.This thesis investigates fundamental methods for entity recognition and relation extraction in natural language understanding of elementary mathematics based on deep learning.We have implemented an automatic reasoning system based on relation matching and expression matching on a relational graph.This thesis tested the expression module on a set of elementary mathematics test questions established by the project.The test results showed that the expression module can accurately and stably identify elementary mathematical expressions,and can achieve a success rate of over 80% in expression matching tasks.