Research And Application Of Entity Introduction And Anaphora Resolution In Elementary Mathematical Problems

With the rapid development of computer science and technology and artificial intelligence,research directions such as human-like answers and machine proofs have be-come hotspots at home and abroad.The human-like answer and automatic reasoning of the computer means that the computer system takes mathematical text as input,extracts the knowledge in the text through natural language processing technology,and then re-alizes the iterative updating process of knowledge through knowledge reasoning.Entity extraction,as the basic task of natural language understanding,plays a vital role in the knowledge extraction of mathematical texts,and its effect will directly affect the meaning understanding of elementary mathematics.This article focuses on the research of entity extraction in the field of elementary mathematics.Based on the characteristics of elementary mathematics text and the overall process of natural language,the task of entity extraction is divided into two sub-tasks:Entity Introduction and Anaphora Resolution.Based on the characteristics of mathematical text representation and the uniqueness of knowledge representation,this paper defines a mathematical entity as a two-tuple containing both entity types and entity variables.However,the entity of text in mathematics is often not such a complete two-tuple.According to the different types of entity elements,the entity is divided into three types: missing entity type,missing entity variable,and missing entity two-tuple.For missing entity types,this article constructs named entity labeling based on the BERT-Bi-LSTM-CRF model of word sequences to complete and introduce the missing types of entities.For missing entity tuples,we expanded the entities by quantifiers and entity type and transformed the problem into the problem of missing entity variables.For the problem of missing entity variables,the method based on mapping rules is used to determine the number and corresponding relationships between the entities to be introduced and other entities,and the introduction of entity variables is performed based on these relationships.The appearance of anaphora makes the formulation of mathematical problems more difficult for computers to understand the relationship between entities in sentences,and it is particularly important to reduce the pronoun to its antecedent.This paper proposes that the anaphora resolution in elementary mathematics is divided into three steps:determine the anaphoras,determine the candidate antecedent list and evaluate the antecedent's ra-tionality.In the step of determining candidate antecedents,the anaphora is divided into inter-sentence anaphora and intra-sentence anaphora according to the positional relation-ship of the pronoun and the antecedent,and different candidate antecedent determination methods are proposed for the two anaphora phenomena.In the antecedent rationality as-sessment step,a rationality assessment method for entities in elementary mathematics is proposed.Based on the above methods,this article designs and implements an elementary math-ematical entity extraction system that combines the introduction of mathematical entities and the dissolution of references,and selects the mathematical problems with reference phenomena in the self-built question bank to comprehensively test the system,during the test,system runs stably and the average correct rate of natural language understanding reached 77%,which has good practical value.