Research And Design Of Automatic Scoring System For Virtual Experiment Platform

In recent years,with the application of the "Internet +" model in the field of education,the traditional education model has also changed.In the field of experimental education,many highly-corrected teaching methods use virtual experiments combined with traditional experiments.As a laboratory assistant system,the virtual experimental platform allows students to perform experimental operations before and after class,which helps improve students' learning ability.Among them,the evaluation score of experimental operation has become the research core and hot spot of the virtual experimental platform.Traditional scoring systems can only score multiple choice questions,they cannot achieve grading of operation questions,or they can only compare experimental data results with standard results to determine whether the student's operation is accurate.There is no evaluation of student's operation process.This thesis is based on the existing virtual experiment platform and combined with artificial intelligence technology to design a set of virtual experiment platform scoring system to achieve intelligent classification of the text data of the operation questions and obtain by judging each step in the text data final score.The main contents of this paper are as follows:(1)Study the overall design framework of the virtual experiment platform,analyze and summarize the scoring methods of the virtual experiment platform based on Lab VIEW,and study the advantages and disadvantages of other scoring system designs,such as:English composition scoring system,programming problem scoring System,etc.(2)In order to better characterize text content,a preprocessing algorithm based on natural language processing is designed to provide effective data for the next text classification and scoring system.In addition,when preprocessing the virtual test text data with more specialized terms,a custom dictionary needs to be added to improve processing efficiency.(3)Design corresponding text classification models by studying convolutional neural network(CNN)and long-short-term memory neural network(LSTM),the advantages of CNN in classification and LSTM in natural language processing are combined to design a hybrid neural network.By comparing the accuracy of the three in text classification,as a result,the classification effect of the hybrid neural network is better.(4)Study the scoring rules of virtual experiments,combine the model structure of decision tree algorithm and recurrent neural network(RNN)algorithm,and design a text scoring algorithm based on the model structure of both,At the same time use the neural network scoring model to re-score the data.When the score error is within the given range,the average value of the two scores is taken as the final result,otherwise the feedback is optimized.Finally,a graphical user interface was created using Python to display the scoring interface.This system gets rid of the shackles of manual scoring in the operation questions in the traditional scoring system.The intelligent scoring model is established through neural networks,which can realize the characteristics of scoring in virtual experiments in different disciplines.