Research On Automatic Scoring And Feedback Of Essays In International Chinese Language Education

As an important part of language teaching work,the composition test is a comprehensive inspection of learners’ language expression ability,material accumulation level and logical organization ability.Timely composition correction and effective feedback can improve students’writing level critically.In recent years,more and more foreigners are learning and using Chinese,and international Chinese language education has attracted great attention in China.The country’s 14th Five-Year Plan pointed out that strengthening international Chinese language education is the basic path to enhance the influence of Chinese culture.However,at present,the Chinese proficiency test for foreign candidates still completes the correction of the composition part manually,which has shortcomings such as low efficiency,high cost,poor objectivity,and lack of feedback.With the rapid development of artificial intelligence,more and more algorithms and models are used in the research of automatic composition scoring,and good results have been achieved in the field of automatic English composition scoring.However,the research on automatic scoring of Chinese compositions in China started late,and there is a lack of research on automatic scoring and feedback for international Chinese compositions.This paper aims to carry out research on automatic correction and feedback of international Chinese composition.The main research work is as follows:(1)By comparing and analyzing the current well-known interlanguage corpora in China,choose the "Chinese Interlanguage Corpus" and "HSK(Chinese Proficiency Test)Dynamic Composition Corpus" as the source of the original corpus,completed data capture and cleaning through relevant technical means,and constructed a training data set for the field of international Chinese composition.According to the relevant corpus that can be used at present,a two-classification model of spoken and written language and a multi-classification model of rhetorical devices were constructed based on the pre-trained language representation model BERT,which obtained better results by means of Dropout,EarlyStop,and visualization of the training process.The model effect proves the feasibility of applying the transfer learning method to the research of automatic feedback of international Chinese composition.(3)In this paper,considering the influence of grammatical errors,language richness and other shallow language features on the automatic scoring of essays,an automatic scoring model of BERT combined with shallow language features is constructed.The quadratic weighted Kappa(QWK)coefficient on the validation set of the model has an average value of 0.710,which ensures a high degree of consistency between the model’s predicted score and the actual score of the essay.(4)This paper uses the above model to design and implement an automatic scoring system for international Chinese composition with certain practical value.