Research And Implementation Of Automatic Scoring System For Subjective Questions

Relying on the development of Internet technology,more and more people are paying attention to online education.In the online examination of Internet education,automatic scoring of objective questions is very mature.However,the subjective question is still scored manually,This approach greatly limits the development of online education.Therefore,it has important significance to research and implement an automatic scoring system of subjective questions for improving the teaching efficiency of online education.This thesis focuses on the problem of automatic scoring of the simple answer,this thesis extracts text features by language and semantics,and does some improvement on text feature extraction algorithms.a sample set with the text eigenvalue as the input value and the manual scoring result as the output value is constructed.Subjective Question Automatic Scoring Model by BP Neural Network is constructed.Based on the text feature extraction and scoring model construction,the prototype of the subjective title automatic scoring system is built.the research works completed are as follows:This thesis firstly conducts corpus collection and manual scoring,the 2018 autumn final exam thesiss of the Network Education Introduction Course of Xidian University is collected,the simple answer is selected,the simple answer is scored by professional scorers and is stored in a database according to a certain format.Secondly,the N-Gram model is studied,the BLEU theory is applied to the simple answer scoring,the similarity of words is calculated by sentence and text respectively.The experimental results show that the calculation of the word form similarity in units of text is more relevant than the calculation of the word form similarity score in terms of sentences.The N-Gram model is susceptible to common words and short text interference,an improved N-Gram Model is proposed.The experiment result shows that the proposed algorithm is effectively imporving the Quadratic Weight Kappa value.Thirdly,word vector and text vector model are studied,a word vector training corpus through course knowledge points,short answer question bank,and Wikipedia corpus is constructed.The word vector is trained using the Skip-Gram+NS model,and the text vector is trained using the PV-DM model.The text similarity feature is extracted according to various text vector representations.Due to the deficiencies of the PV-DM model,an improved PV-DM Model is proposed.The experimental results show that the improved PV-DM model extracts the text vector similarity feature and its score correlation is higher than that of the pre-improvement model.Then,automated scoring model of subjective questions is constructed by BP neural network.Because the BP algorithm converges slowly and falls into local minimum values easily,the PSO algorithm is added for optimization.The experimental results show that the results based on the PSO-BP neural network model predictive scores are closer to the results of manual scoring,and are more suitable for the construction of subjective auto-scoring models.Finally,the prototype of the subjective title automatic scoring system is constructed,and the modules of the system are designed and implemented,including data collection module,text feature extraction module and prediction implementation module.