“…Literature [11] proposes the matching rate of n-order word elements as the scoring rule, and at the same time introduces the length penalty ratio to solve the problem of high scores for short sentences; literature [12] analyzes the writing scoring features from the perspectives of literal overlap, keywords, semantics, etc., and constructs the writing automatic scoring model. Writing automatic scoring model methods include feature engineering-based machine learning methods, deep learning-based methods, and hybrid model-based methods [13]. Literature [14] uses convolutional neural network to obtain sentence features and proposes an automatic scoring method for composition by fusing topic features; Literature [15] extracts artificial features based on grammaticalsyntactic rules for composition scoring and scores the composition by using a support vector machine; Literature [16] constructs an automatic scoring system for composition by extracting a variety of compositional features, such as words, sentences, and so on, and by using the random forest algorithm; Literature [ 17] extracts the feature set from three aspects, such as word frequency, word size, distribution location, etc., and uses random forest to score the composition; literature [18] uses a hybrid neural network model to identify and analyze the gracefully entering boards in the compositions; literature [19] firstly uses deep neural network to obtain the semantic representation of the compositions, and then, combined with manually extracted compositions features such as the number of misspelled words and the number of characters, it uses the XGboost classifier to predict and analyze the composition score.…”