Research On Model And Method Of Automated Essay Scoring

As a kind of essential question in language examination,writing usually needs enormous work in essay scoring.On the one hand,using manual scoring method waste a lot of human and material resources,on the other hand,manual score depends on the subjective judgment of teachers,which may have a certain error.With the development of natural language processing technology,it has made breakthrough progress in syntactic analysis,semantic analysis and emotional analysis.It is of great significance to use n atural language processing technology in automated essay scoring.The traditional automated essay scoring method mainly uses the statistics of whole article such as lexical,syntactic,semantic and other features to train the machine learning model and predict the score.However,the method usually depends on the artificial extraction of the regular features,stay in the overall quality of the chapter,and can't make good use of more detailed information,such as the context of the text.Based on these problems,this paper makes a further study on the characteristics of the sentence level upon the traditional scoring method,and utilizes the temporal context information of the chapter to extract the potential logic and coherence of the essay on the scoring m odel.In addition,according to the writing situation of the subject matter in the language examination,the effect of the subject relevance in the scoring model is studied from the point whether the essay is consistent with the meaning of the essay.The main content of this paper include the following three aspects:(1)This thesis studies the application of the sentence representation and deep learning model in automated essay scoring.In writing this particular task,composition is more suitably regarded as a logical statement sequence.In order to dig the hidden logical information between sentences,this paper mainly uses the unsupervised method to build the sentence vector,including Doc2 Vec,RAE,and so on.Then build model by using deep learning meth od.Based on the advantages of CNN model in extracting features and the characteristics of LSTM model for solving sequence question,this paper designs and implements a variety of model structure to find a suitable way to build the model.Finally,by the integrating with the traditional score model,improve the score of final model.(2)This thesis studies the significance of the relevance of topic in essay scoring.This paper mainly introduces the similarity between title and body,then introduces the concept of semantic dispersion based on topic,and analyzes its influence on the final score.(3)Design and implement an automated essay scoring system.This system can give feedback according to the quality of essay and give the feedback in terms of lexical,syntactic,logical and theme.And make an increase by complement the error correction module,which can provide the users with a better writing feedback experience.