Automatic Scoring Algorithm For Chinese Subjective Questions Based On BERT Pre-training Model

With the rapid development of online education in recent years,the traditional examination has a tendency to shift from offline examination to online examination.As an important part of online education,online assessment has also received increasing attentions.Automatic scoring technology for objective questions has been widely applied to various online education systems or automatic grading devices,which has improved the efficiency of teachers' scoring.However,there is still a lack of effective automatic scoring techniques for automatic scoring of subjective questions.With the rapid development of artificial intelligence science and natural language processing technology,subjective question automatic scoring technology.has also improved.At present,domestic and international researches on automatic scoring of subjective questions mainly calculate the similarity of two texts by analyzing the shallow semantics of the submitted answers and the reference answers,so as to judge the subjective question.Although this method has a certain effect,it needs to be further improved to be applied to the actual online evaluation environment.This paper uses two different methods to automatic scoring research on Chinese subjective questions,and designs and implements relevant online evaluation systems on this basis.The research content is as follows:(1)For the current situation that the accuracy rate of the automatic scoring algorithm for Chinese subjective questions is not high.An automatic Chinese subjective scoring algorithm based on Word Embedding is proposed.This algorithm analyzes and studies English sentence similarity algorithm,the process of serializing the sentences was extracted,and the characteristics of Chinese subjective questions were analyzed,and corresponding improvements were made.Combined with the automatic scoring formula,it can perform better on the automatic subjective task of Chinese subjective questions.Experiments show that this method is significantly improved compared to the traditional Chinese subjective automatic scoring method.Through the experimental research on the automatic scoring algorithm for Chinese subjective questions based on Word Embedding,it is found that the method based on Word Embedding has obvious deficiencies in the recognition of polysemy.(2)In order to make up for the shortcomings based on the Word Embedding method,An automatic scoring algorithm for Chinese subjective questions based on the BERT pre-training model is proposed.this algorithm uses the pre-training model for the automatic scoring task of Chinese subjective questions,by pre-training and fine-tuning the model,the serialized representation of words will be adjusted according to the context.Therefore,for polysemous words,the word corresponds to different semantics in different contexts,and the numerical sequence corresponding to the word also changes accordingly.Thereby solving the problems of Word Embedding.In the experimental environment of this paper,the automatic scoring algorithm for Chinese subjective questions based on the BERT model is better than the automatic Chinese subjective question scoring algorithm based on Word Embedding.(3)On the basis of the automatic scoring algorithm for Chinese subjective questions,this paper designs and implements a fully functional online evaluation system,and verified the practicability and effectiveness of the system.While the system is running,it has realized the function of collecting subjective question data in Chinese,and contributed to the supplement and improvement of the subjective question data set in Chinese.In the end,this paper summarizes the proposed algorithm,points out the shortcomings of the algorithm,and looks forward to the improvement of the algorithm in the future.