Mathematical Arithmetic Operators Recognition Based On The Character-level Convolutional Neural Network

In recent years,under the background of education informatization and with the rapid development of artificial intelligence,natural language processing,and the keen needs for more and more electronic documents of mathematic questions,exploring a new learning mode of students independent innovation under the environment of educational informatization,helping students to improve the ability of math problem solving,and offering the intelligent education service for learners become inevitable trend.Arithmetic plays a key role in mathematics.The realization of the arithmetic of intelligent solutions problems can provide the technical support for the students'autonomous learning under the environment of education informationization.Building the intelligent answering model for the arithmetic problems can provide technical support and reference for the mathematical operators,merging similar terms,solving equations,which lay the basis for machine automatic solving mathematic problems.However,the key of the intelligent solutions of the arithmetic problems is that the recognition the type of arithmetic operators.Put forward to the Deep Learning as the breakthrough point,this study adopt the manner of character-level encoding,to train the CNN and produce the solving model for the recognition of arithmetic operators,and verify that the CNN network can correctly recognize the arithmetic operators.This research mainly includes the following four aspects:First,building the algorithm structure and establishing the convolutional neural network.Convolutional Neural network needs to set the convolution,pooling,feature extraction method,and the number in the neural network of hidden layer and the corresponding node number of hidden layers,and so on.Second,design the encoding and the rules of characters,quantitation.Quantitation the characters by one-hot encoding method according the rules,which will make the input datasets to turn to a one-dimensional vector.Third,auto-generate the experimental data,the data sets of three types of mathematical arithmetic were generated from four dimensions of the addition,subtraction,multiplication,and division,including the training set,validation dataset and test dataset.Fourth,the processed datasets were input to the designed network,to train and study the "experience" for recognizing the mathematical operators,and using test dataset test the recognition result.Nowadays,a large number of scientific research work for automatic solving mathematic problems is based on the word text categorization,but this study depends on the encoding and recognition of character.Through the experiment verified the effectiveness of the network,the network model for the arithmetic operators recognition of recognition accuracy is close to 100%,compared with the traditional neural network,the deep convolution network recognition accuracy rate has achieved 100%.Besides,Deep convolution network is forecasted with a randomly generated forecast,achieved the exactly right's result,which realized the goal of recognition the mathematical arithmetic operators.