A Study Of Automatic Essay Scoring In HSK

At present,automatic essay scoring technology has been practically applied in some standardized examinations in foreign countries,which greatly saves manpower and financial resources and improving the efficiency.In China,the writing part of the HSK exam is marked manually in a more traditional way,therefore,the marking process is inevitably subjective.Automatic scoring technology has a lot of advantages,such as objectivity and neutrality that manual scoring does not have,while at the same time being efficient,time-saving and labor-saving.The ultimate research objective of this paper is to implement a usable automatic scoring system for HSK essays.Automatic scoring of essays is first and foremost a multiclassification task,and there are different algorithms for multiclassification tasks,such as Support Vector Machine(SVM),Linear Regression,Naive Bayesian algorithm,etc.,.With the continuous development of deep learning technology,the application of artificial neural network technology in the field of natural language processing has been phenomenal,which provides excellent technical support for the study of automatic scoring of HSK essays.Most of the previous HSK essay scoring studies have been based on linear regression to calculate the correlation between different indicators and essay scores,but there are no studies on the superiority or inferiority of linear regression over techniques such as support vector machines,Park Bayesian and deep learning in automatic scoring of HSK essays.Based on such research gaps and questions,the paper first collected the data set from the HSK Dynamic Composition Corpus developed by Beijing Language and Culture University.Second,based on this data-set,four models were trained in this paper using linear regression,Support Vector Machine(SVM)and LSTM technique respectively.According to the results,the scoring models built with the SVM and LSTM algorit hms performed well on a certain part of the data-set,while they performed poorly on th e rest parts.Meanwhile,the scoring model based on the multiple linear regression meth od performs consistently well on all parts of the data-set.Therefore,this paper conclude s that the multiple linear regression model is most applicable to the HSK essay automati c scoring task.Finally,this paper further deploys a scoring model based on multiple linear regression to the Web,turning the theoretical technical research into practical and actionable applications.