Research On Automatic Scoring Technology Of Essay

With the maturity and widespread application of natural language processing technology,the research of natural language processing in the field of education is also showing a gradual upward trend.Composition writing assessment plays a more and more important role in the field of education in China,and the number of people taking various types of examinations is increasing every year.How to alleviate the unfairness caused by expert composition review and improve subjective consciousness has become an urgent problem to be solved.It is the trend of our times to use intelligent technology to solve the problem.Automatic Essay Scoring(AES)is a technology that uses computers to automatically grade the compositions of different languages and feedback them to the users to get corresponding guidance or suggestions.With the help of specific computer programs,the workload of grading teachers can be reduced to a certain extent,and the fairness and accuracy of grading can be improved.This is because in the process of grading manually,the subjective factors of the grading teachers may lead to deviations in the grading process,which may lead to challenges to the fairness of the examination.Therefore,it is of great practical significance to use computer to assist the reference of the composition grading.Based on the research of the related achievements of the automatic scoring technology of Chinese compositions at home and abroad,and combining the characteristics and scoring standards of Chinese compositions examination,this paper attempts to use the features of topic relevance and discourse coherence to effectively represent and predict the automatic scoring of Chinese compositions.The main research focuses include:(1)Aiming at the research of the test model of composition run-off,a test model of composition run-off based on LDA coupled space model is proposed.LDA topic extraction technology is used to extract the subject words of the composition title and the composition to(2)be evaluated respectively.Then the vector representation of the subject words is put into the coupling space to generate a pan-semantic matrix,and the relevance of the subject words is used to judge the degree of joint topic,so as to judge the degree of joint topic.Whether there is a problem of off-topic in the composition to be evaluated.(3)According to the feature extraction of discourse coherence,through careful study of a large number of compositions and scoring criteria,it is considered that discourse coherence is an important indicator to measure the quality of compositions.Therefore,a framework-based entity grid method is used to quantify discourse coherence,which is then used as one of the features to predict the overall score of compositions.(4)Attempt to use the elastic network regression model which is more suitable for the features with high correlation as the prediction model,and add the above extracted features into it respectively.Experiments are carried out and the experimental results are compared to explore the extraction features and the effectiveness of the prediction model.