Research And Implementation Of Key Technology Of Elementary Mathematical Word Problems Solving Automatically

There are two difficulties in the study of the automatic solution of word problems:the semantic comprehension and the derivation.This thesis takes the natural language processing and automatic reasoning technology as the theoretical basis of the research,with word problems of probability and statistics in Senior High School as the research object,in order to achieve the automatic solution as the research goal.It discusses the realization of knowledge representation,natural language processing,automatic derivation and methods for the overall implementation of the system about word problems.The main research of this thesis includes the following points:(1)The knowledge representation of word problem.On the basis of kintsch's one-step application knowledge representation,this thesis realizes the extended knowledge representation that can store the information of probabilistic statistical word problem.With this knowledge representation as the basis for understanding and deriving derivation,the problem is automatically solved.Experiments show that the extended framework knowledge representation can cover the information of probability and statistics,and effectively organize automatic derivation.(2)The research and construction of natural language processing.The natural language processing of word problems can be divided into two modules: automatic classification of question types and automatic extraction of semantics.This article classifies the question type classification into text classification.This thesis uses a support vector machine to use the vector space model obtained by feature engineering as a training input and trains the classifier out.Automatic classification is so important that provides a basis for the use of specific strategies for semantic extraction and automatic calculus.For the automatic extraction of semantics,this thesis starts with the perspective of linguistics,analyzes the linguistic features of word problems,and presents the organization forms of propositions and propositions in word problems.Combining the concept of proposition sets with the syntactic technique in natural language processing,a propositional sentence model method suitable for automatic extraction of application title semantics is proposed.The method of constructing the propositional sentence model is provided from two aspects: the organization form of the sentence model and the matching strategy,so as to achieve the complete extraction ofthe abstract quantitative relationship in the word problem from the complicated scene information.(3)Research and construction of automatic reasoning.In order to automate the derivation of word problems,this thesis adopts the method of inference engine that maps between the condition set and the problem to be solved is maintained by rules in the inference engine.In the inference engine,the condition set maps to the facts in the fact library,the problem serves as the equivalent condition of the downtime,and the inference method only uses forward inference.The inference engine is responsible for the iterative derivation of the conditions.Manpower only needs to build a rule base and an outage condition,which simplifies the process of searching for sub-questions when solving multi-step word problems.This thesis studies and constructs the modules of knowledge representation,natural language processing and automatic derivation.The solution and concrete implementation of each module are given.Finally,the automatic problem solving system for word problems is realized.The system automatically solves the problems of probability statistics.The success rate of the automatic solution reached 61%.