A System For Automatic Classifying And Solving Math Word Problems Based On Procedural Knowledge

The automatic solution of math word problems is a very comprehensive research topic: it involves several key technologies such as text preprocessing,natural language understanding,knowledge extraction and representation,and automatic solution.This article proposes the automatic classification and automatic solution of word problems based on procedural knowledge.Based on the two core goals of automatic classifying and automatic solution of word problems,an word problem classification system is established,the text processing of word problems based on procedural knowledge,the automatic classifying,the automatic solution is studied.This article mainly researches from the following points:(1)Research and construction of word problems text processing based on procedural knowledge.Word question text processing aims to process the original text of the word problem into a more uniform and abstract text form,which will provide a lower complexity input form for automatic classifying and automatic solution of the word problem.Word problems text processing based on procedural knowledge abstracts the declarative knowledge of the word problem while retaining procedural knowledge,which mainly includes coreference resolution,word segmentation and pos tagging,morphological reduction,and text abstraction.(2)Research and construction of automatic classification algorithms of word problems.This section first establishes a classification system for word problems.This article starts from the textbook system for mathematics subjects in primary schools in China.Based on the ideas for solving word problems in this range and examining students' corresponding mathematical abilities,a classification standard for word problems is established.Variations in research topics for large categories are divided into small categories based on large categories to establish a more detailed classification system.Then this part studies the application of automatic text classification technology in the automatic classification of word problems.It mainly studies the features extraction and representation,classification model establishment,model fusion and other technologies.(3)Research and construction of automatic solution of word problems based on Transformer.Word problem auto-mapping maps the entire text of the text after processing to a custom structured formula description,which is finally converted into a standard formula representation and solved.This part mainly includes formula mapping based on Transformer and Maple symbol calculation.In this paper,by establishing a classification system for word problems,this article studies the techniques of text processing,automatic classification,and automatic solving of word problems based on procedural knowledge,and builds a system that can complete the automatic classification and automatic solving of word problems.The final system was tested on a data set consisting of the public data sets AI2,IL,CC and self-built 1000 questions.A classification accuracy of 60.9% and a solution accuracy of 81.0% were achieved.