Research And Application Of Core Technology For Automatic Solving Of Elementary Mathematical Application Questions

In recent years,with the continuous development of artificial intelligence and the continuous advancement in the field of natural language processing,more and more online intelligent education platforms centered on automatic answering and human-like counseling are becoming more and more important for students' learning effect.In order to realize the online intelligent education system,the key is how the title of the Chinese text form is recognized by the computer and stored in a corresponding reasonable data structure for knowledge deduction.The automatic answering function of the online education system provides students with a more human-like and more effective learning method.Among them,the elementary mathematics application questions is the difficulty of the automatic answering function in the mathematics field.In this paper,the natural language processing technology and Google neural network machine translation and other related technologies are taken as the theoretical basis,and the automatic answering of application questions is taken as the research goal.The core contents such as knowledge representation,semantic understanding and automatic deduction of elementary mathematics application questions are discussed.On this basis,an automatic solution system is implemented.The main research of this thesis is as follows:Firstly,the application knowledge representation of this paper is based on Kintsch's single-step application knowledge representation framework,and proposes a new information framework,including Kintsch's knowledge representation framework,and expands some new content,such as attribute slots,unit slots,and quantity relationship slots.It is proved by experiments that the proposed information framework can represent the topic information of elementary mathematics application questions and can be effectively used in automatic solution.Secondly,semantic understanding,the semantic understanding of elementary mathematics application questions refers to extracting information from the application text and storing it.In the information framework,the entities,attributes,quantities,units,and quantity relationships are extracted from the text and filled in the slots corresponding to the information frame.For the extraction of entities and attributes,this paper uses the conditional random field as the theoretical basis,uses the named entity recognition method,and uses the CRF++ toolkit for named entity recognition.For the extraction of quantities and units,this paper uses the feature template matching method to extract the quantity relationship.This paper takes machine translation as the theoretical basis,uses Google Neural Network Machine Translation(GNMT),prepares training corpus,and uses GNMT to realize the mapping of application language to mathematical language.In summary,a complete information framework can be constructed and used in the automatic solution of the application questions.Finally,it is an automatic deduction.This paper proposes a relational framework to extract the hidden relationship in the application title,and uses Maple symbol calculation to realize the automatic solution of single-step or multi-step application questions,and uses the self-built rule base to solve the problem.The relational framework relies on the common sense relational library and the dynamic entity library.The common sense relational library is based on the common sense knowledge base and knowledge map,and is built on the database.The dynamic entity library is an entity and attribute library automatically generated for each application title text.Based on the above research,this paper constructs an automatic answering system for elementary mathematics application questions,and elaborates the various modules and implementation methods of this system.In the self-built 300 test questions,the automatic answer rate is 74%.