Research On The Automatic Feedback Method Of Mathematics Short Answer Questions For Intelligent Marking

Intelligent marking is an important research content of intelligent education,including two sub tasks: automatic scoring and automatic feedback.Automatic scoring refers to giving scores for students' answers through the trained language model;automatic feedback refers to pointing out the wrong points of students' answers and putting forward suggestions for modification.The purpose of automatic feedback is to improve learning efficiency and reduce learning cost.At the same time,the feedback results are more rational and not affected by personal emotions.Based on the review of the history of automatic feedback at home and abroad,this paper analyzes the shortcomings of the existing technology.Aiming at the problem of automatic feedback of mathematical short answer questions,this paper studies the feedback from two perspectives of text classification and atlas construction.The main work and research results are as follows:(1)Research on automatic feedback method of mathematical short answer based on text classification.convolution neural network and capsule network are used to classify the answers.Convolutional neural network extracts students' answers through convolution layer and pooling layer,while capsule network replaces single neuron node in traditional CNN model with neuron vector,and trains this new neural network through dynamic routing and static routing.On this basis,a feedback template is set up by analyzing the characteristics of each type of answer.Different types of answers take different methods to fill in the feedback template.The experimental results show that the capsule network has a 3% to 8.3% improvement compared with CNN,and the overall accuracy can reach more than 83.3%.(2)Research on the automatic feedback method of mathematical short answer based on knowledge map.we construct a knowledge map of junior high school mathematical concepts,which extracts mathematical concepts from textbooks as entities through regular expressions,and uses crawler tools to supplement entities from Baidu Encyclopedia.At the same time,the relationship between four entities is defined and extracted according to the inherent logicality of the textbook.In the process of feedback,students' answers,standard answers and their relationship in the map are generated into a feedback path,which provides feedback for students.The experimental results show that the accuracy of entity extraction can reach more than 98%,the accuracy of concept relationship can reach more than 93.3%,and the generation ratio of feedback chain can reach more than 99%.(3)The design and implementation of the automatic feedback system of mathematics short answer questions.Based on the analysis of application requirements,this paper designs and implements an automatic feedback system of mathematical short answer questions based on B / s.The system encapsulates the trained automatic feedback language model based on capsule network to achieve the goal of automatically providing feedback for students' answers.