Research On Automatic Evaluation Of Subjective Questions In High School Mathematics

At present,the vast majority of automatic examination evaluation only better realize objective questions such as judgment questions,blank filling questions and multiple-choice questions,while the automatic evaluation of subjective questions,which occupy the largest workload of teachers' evaluation,is difficult to realize,especially in mathematics.Although there are some automatic scoring methods for mathematical application problems,the degree of intelligence is not high,and there are some problems,such as ignoring the logical order of sentences,unable to effectively identify the formulas in mathematical texts and difficult to deal with the situation of multiple solutions to one problem.In view of this,based on the deep learning algorithm of dictionary and text similarity,this paper studies the theory and method of automatic scoring for high school mathematics triangle application problems,and tries to solve the problem of automatic scoring for multiple solutions of one problem.The specific research work is as follows:(1)High school mathematics texts contain special symbols and formulas in many fields,and these symbols and formulas play a vital role in the whole mathematics evaluation.At the same time,the mathematical answers are complex.Generally,a complete solution requires multiple steps,and the problem-solving process is rigorous and logical.In view of this characteristic,this paper will solve the problem step by step,extract the formula,create a special formula dictionary in the field of trigonometric function,and comprehensively consider the characteristics of key steps,formula and sentence logic,and construct a multi feature fusion step-by-step scoring algorithm based on Bleu principle.The algorithm integrates various features of mathematical text,making the final score more reasonable and perfect.In addition,aiming at the situation of multiple solutions to one mathematical problem,an algorithm of fuzzy full matching between students' answers and various standard answers is also proposed,and the score with the highest matching degree is taken as the final score of students.(2)Aiming at the characteristics of high school mathematics text,such as strong logic and close causality,an overall similarity scoring model based on the symmetrical twin network structure of combined convolution network(CNN)and bidirectional long-term and short-term memory network(bilstm)is proposed.The local mathematical text feature matrix extracted by CNN and the global text feature matrix extracted by bilstm are modeled at different levels of similarity to obtain the deep-seated language features of sentences.Then the two similarity matrices are fused and spliced,which can effectively maximize the retention of mathematical text features.(3)In view of the particularity of the data set in the field of trigonometric function,this paper is based on the real data of Tianjin high school joint examination and the content of high school trigonometric function.The final trigonometric function data set is obtained through a series of processes such as asking the teacher to obtain,offline sorting and collection,cleaning and weight removal,scanning and correction,and manual marking scores.The data set is applied to the multi feature fusion step-by-step scoring algorithm based on Bleu and the overall similarity scoring model based on CNN bilstm,and good experimental results are obtained,which verifies the effectiveness of the two models in this paper.