Research On Automatic Solution Model Of Word Problem Based On Digital Correlation Features

In recent years,with the rapid development of computer science,natural language processing and artificial intelligence,intelligent question answering system has attracted more and more attention.Online auxiliary education,network automatic question answering and other applications have emerged and been widely used.Among them,the automatic solution of math word problems is the most basic intelligent problem-solving research in personalized auxiliary education,which can help students to learn independently and free from geographical and time constraints.Extracting effective features and constructing automatic solution model is one of the hot issues in the field of automatic solution of word problems.Based on the analysis of the features related to numbers,this paper proposes three kinds of extraction methods of features related to numbers,including single digital features,digital pair features and equation features.According to the accuracy of automatic problem solving model for algebra word problems,a algebra word problem solving model WAPNet(Word Algebra Problem Neural Network)based on deep neural network is constructed.The main contributions of this paper are as follows: first,identify the numbers in the algebra word problems and form the sequence of numbers,and extract the contextual information of each number in the sequence of numbers.The single digital feature,two adjacent digital pairs and equation features in digital sequence are extracted by context-free grammar;secondly,the algebra word problem solving model WAPNet,is constructed by using deep neural network,and the numerical feature is taken as input.Taking the equation template as the output,and using PSO(Particle Swarm Optimization,PSO)to optimize the weights and biases in the neural network model,the optimal weights and biases are obtained as the initial values of the network,and the final model is obtained by training the network;finally,Based on the accuracy of automatic solution of word problems,the automatic solution model of word problems based on digital correlation features is determined.In this paper,Alg514 and Single are used as data sets,including univariate first-order equation and binary first-order equation system of linear algebra word problems,completed in the word problems related to the number of feature extraction,feature representation and network model construction and training.The experimental results show that the method in this paper is trained and tested on Alg514 dataset,and cross-verified with 50% discount,and the accuracy of automatic problem solving is 88.9%.In the same case,compared with the method based on log linear model,it is improved by 9.2%.In this paper,the method is tested on the SingleEQ dataset,and the accuracy of 89.6% is obtained,which verifies the effectiveness of the proposed method.This paper has 11 pictures,18 tables and 52 references.