CN113435179A

CN113435179A - Composition evaluation method, device, equipment and storage medium

Info

Publication number: CN113435179A
Application number: CN202110705457.2A
Authority: CN
Inventors: 巩捷甫; 章继东; 呼啸; 宋巍; 王士进; 胡国平; 秦兵; 刘挺
Original assignee: Hebei Xunfei Institute Of Artificial Intelligence; Zhongke Xunfei Internet Beijing Information Technology Co ltd; iFlytek Co Ltd
Current assignee: Hebei Xunfei Institute Of Artificial Intelligence; Zhongke Xunfei Internet Beijing Information Technology Co ltd; iFlytek Co Ltd
Priority date: 2021-06-24
Filing date: 2021-06-24
Publication date: 2021-09-24
Anticipated expiration: 2041-06-24
Also published as: CN113435179B

Abstract

The application provides a composition review method, a composition review device, composition review equipment and a storage medium, wherein the method comprises the following steps: detecting whether a target composition to be reviewed is an abnormal composition; if not, the target composition is modified from the word level, the sentence level and the chapter level respectively to obtain corresponding modification results of the target composition on the word level, the sentence level and the chapter level respectively; determining the grading of the target composition from the plurality of evaluation dimensions to obtain the grading of the target composition on the plurality of evaluation dimensions; and grading to generate the comment of the target composition according to the grading of the target composition on a plurality of scoring dimensions. The composition review method can automatically review the composition to be reviewed, and manual participation is not needed, so that the problems caused by manual participation are avoided.

Description

Composition evaluation method, device, equipment and storage medium

Technical Field

The application relates to the technical field of text data processing, in particular to a composition review method, device, equipment and storage medium.

Background

Writing is the essential practical skill in people's daily life and study, also is the essential ability that requires student's key mastery in the education of school.

Most of the current composition evaluation methods are manual evaluation methods, that is, the contents of the composition are checked by an evaluator, and the composition evaluation is subjectively given according to the understanding of the contents.

However, in some cases, the number of the compositions to be reviewed is often large, the compositions need to be reviewed one by the reviewer and then the reviewer can give comments, which is time-consuming and labor-consuming, and the review workload is large, so that the comments given by the reviewer are often very brief, and the instruction effect on the writer is limited.

Disclosure of Invention

In view of the above, the present application provides a composition review method, apparatus, device and storage medium, so as to solve the problems that the existing composition review method is time-consuming and labor-consuming, the given review is very simple, and the guidance effect on writers is limited, and the technical scheme is as follows:

a composition review method comprising:

detecting whether a target composition to be reviewed is an abnormal composition;

if not, modifying the target composition respectively from the word level, the sentence level and the chapter level to obtain corresponding modifying results of the target composition respectively on the word level, the sentence level and the chapter level;

determining a rating grade of the target composition from a plurality of rating dimensions to obtain a rating grade of the target composition on the plurality of rating dimensions;

and generating the comment of the target composition according to the grading of the target composition on the plurality of scoring dimensions.

Optionally, the multiple review dimensions are multiple review dimensions corresponding to the school segment to which the target composition belongs in multiple set dimensions;

determining a scoring profile for the target composition from a plurality of scoring dimensions to obtain a scoring profile for the target composition in the plurality of scoring dimensions, comprising:

and determining the grading of the target composition from a plurality of scoring dimensions corresponding to the school section to which the target composition belongs so as to obtain the grading of the target composition on the plurality of scoring dimensions corresponding to the school section to which the target composition belongs.

Optionally, the generating the comment of the target composition according to the grading of the target composition on the multiple scoring dimensions includes:

determining the comment corresponding to the grading grade of the target composition on each grading dimension based on the corresponding relation among the pre-constructed grading dimension, the grading grade and the comment;

and generating the comment of the target composition according to the comment corresponding to the grading grade of the target composition on each scoring dimension.

Optionally, the composition review method further includes:

determining the grading of the whole target composition;

the generating the comment of the target composition according to the grading of the target composition on the scoring dimensions comprises the following steps:

and generating the comment of the target composition according to the grading of the target composition on the plurality of scoring dimensions and the grading of the whole target composition.

Optionally, the determining the grading of the target composition as a whole includes:

predicting the overall score of the target composition as a first score of the target composition based on a pre-established score prediction model;

and/or predicting the score of the target composition and the model essay in the model essay set based on a pre-established score prediction model, predicting the integral score of the target composition as a second score of the target composition based on the score of the target composition and the model essay in the model essay set and the score of the model essay in the model essay set, wherein the model essay in the model essay set and the target essay belong to the same theme;

and determining the grading of the score of the whole target composition based on the first score and/or the second score of the target composition.

Optionally, the detecting whether the target composition to be reviewed is an abnormal composition includes:

if the target composition meets at least one of the following conditions, determining that the target composition is an abnormal composition:

the first condition is as follows: the similarity between the target composition and a text in a pre-constructed famous-brand material library is greater than a preset similarity threshold;

and a second condition: the occupation ratio of sentences of which the chaos degrees are larger than a preset chaos degree threshold value in the target composition is larger than a preset occupation ratio threshold value;

and (3) carrying out a third condition: sensitive information appears in the target composition.

Optionally, modifying the target composition from the word level includes:

performing one or more of the following treatments on the target composition: analyzing language accuracy, retrieving network vocabularies, identifying error punctuations, and generating a correction result corresponding to the target composition at a word level according to a processing result; wherein the language accuracy analysis comprises wrongly written characters, error detection of grammar and/or idiom type and/or ancient poetry;

amending the target composition from sentence level, comprising:

performing one or more of the following treatments on the target composition: graceful sentence recognition, high-level vocabulary statistics, modification method recognition, written sentence recognition, and generating a modification result corresponding to the target composition at a sentence level according to a processing result;

the target composition is amended from the chapter level, and the method comprises the following steps:

and performing chapter structure identification and/or theme identification on the target composition, and generating a batch modification result corresponding to the target composition at a chapter level according to an identification result.

Optionally, performing error punctuation identification on the target composition, including:

for each paragraph in the target article:

removing punctuation marks in the paragraph, and taking the paragraph with the punctuation marks removed as a target text;

predicting the label category corresponding to each word in the target text based on a pre-established punctuation prediction model; the label type corresponding to one word is used for indicating whether a punctuation mark exists behind the word or not, and the punctuation mark is of a certain type if the punctuation mark exists;

according to the label category corresponding to each word in the target text, punctuation marks predicted aiming at the target text are determined;

and determining the punctuation mark with error in the punctuation mark of the paragraph according to the punctuation mark in the paragraph and the punctuation mark predicted aiming at the target text.

Optionally, the process of performing the cross-word error correction on the target composition includes:

making each sentence in the text for the target:

detecting wrongly-written characters in the sentence, and acquiring a candidate character set corresponding to the wrongly-written characters;

covering wrong words in the sentence based on a pre-established mask language model, and predicting the probability that the words at the covering positions in the covered sentence are the candidate words in the candidate word set corresponding to the wrong words;

determining a correct word corresponding to the wrongly written word from the candidate word set corresponding to the wrongly written word based on the predicted probability;

and correcting the wrongly written characters in the sentence into correct characters corresponding to the wrongly written characters.

Optionally, the detecting the wrongly written characters in the sentence includes:

for each word in the sentence:

replacing the character in the sentence with each character in a confusion character set corresponding to the character, and forming a candidate sentence subset by each sentence after replacement;

calculating the confusion degree of the sentence and the confusion degree of the sentences in the candidate sentence set based on a pre-established statistical language model;

and determining whether the word is a wrongly-written word or not based on the confusion of the sentence and the confusion of the sentences in the candidate sentence set.

Optionally, the process of syntax error detection on the target composition includes:

making each sentence in the text for the target:

obtaining the syntactic dependency characteristics of the sentence, and the characteristics of each character, the participle characteristics of each character, the mutual information characteristics of each character and the part-of-speech characteristics of each character in the sentence;

determining a context vector of each word in the sentence according to the acquired features;

and determining whether the sentence has grammar errors or not and the specific grammar errors when the sentence has the grammar errors according to the context vector of each word in the sentence.

Optionally, the plurality of comment dimensions are a plurality of comment dimensions of one or more of the following four aspects: content, expression, structure, development, wherein:

the review dimensions for the content aspect include one or more of the following review dimensions: the thought is healthy, the emotion type is satisfied, the theme is satisfied, and the content is full;

the review dimensions for the expression aspects include one or more of the following review dimensions: basic expression, normative Chinese characters, fluent language and accordant style;

structural aspects of the review dimension include: the structure is strict;

developmental scoring dimensions include: and (5) culture and harvest.

Optionally, determining the scoring grade of the target composition on the scoring dimension of the thought health includes:

judging whether each sentence in the target writing text contains vulgar language;

according to the judgment result of each sentence in the target composition, determining the grading of the target composition on the evaluation dimension of the thought health;

determining a grading of the target composition on the scoring dimension of the emotion type, comprising:

identifying the emotion type expressed by the target composition;

according to the emotion type recognition result of the target composition, determining the grading of the target composition on the grading dimension of the emotion type;

determining a grading of the target composition on the scoring dimension of the subject-to-term, comprising:

acquiring a representation vector of each word in the title of the target composition and a representation vector of each word in the text of the target composition, and determining a target vector corresponding to each word in the title of the target composition based on the acquired vectors, wherein the target vector corresponding to one word can represent the correlation between the word and each word in the text of the target composition;

determining the matching degree of the title of the target composition and the text according to the target vector corresponding to each word in the title of the target composition, and determining the grading of the target composition on the grading dimension which is in line with the meaning of the title according to the matching degree of the title of the target composition and the text;

determining a rating grading of the target composition in the dimension of the content enrichment, comprising:

determining chapter representation vectors of the target composition based on the target composition, the basic information of the target composition and corresponding correction results of the target composition on sentence level and chapter level respectively;

and determining the grading of the target composition on the content enrichment scoring dimension by classifying the discourse representation vectors.

Optionally, determining the scoring grading of the target composition on the scoring dimension of the basic expression comprises:

performing new word and/or common idiom recognition on the target composition based on new word tables and/or common idiom libraries respectively corresponding to different school paragraphs to obtain word use conditions of the target composition on different school paragraphs;

carrying out sentence pattern recognition on the target composition based on a specified sentence pattern to obtain the sentence pattern use condition of the target composition;

determining grading of the target composition on a scoring dimension of basic expression based on word use conditions of the target composition on different school paragraphs and sentence use conditions of the target composition;

determining a scoring grading of the target composition in a scoring dimension of the literary specification, comprising:

detecting whether the title, and/or paragraph, and/or network expression, and/or punctuation mark, and/or discourse format of the target composition meet the specification;

according to the detection result of the target composition on the literary rule, determining the grading of the target composition on the scoring dimension of the literary rule;

determining a scoring grading of the target composition in a scoring dimension of the language fluency, comprising:

extracting feature vectors from the basic information of the target composition, statistical information of the results of the odd-character error correction and the grammar error detection and the matching and combination of the target composition;

determining a grading grade of the target composition on a scoring dimension of the language fluency by classifying the extracted feature vectors;

determining a scoring grading of the target composition on the dimension of the literary match with the review, comprising:

recognizing the genre of the target text based on a pre-established genre recognition model, wherein the genre recognition model is obtained by training a training composition marked with the genre;

and determining the grading of the target composition on the scoring dimension of the composition which accords with the genre according to the condition that whether the genre of the target composition is consistent with the specified genre or not.

Optionally, the recognizing the genre of the target composition based on the pre-established genre recognition model includes:

coding each word in the target composition based on the genre identification model to obtain a coding vector of each word in the target composition;

performing attention calculation on the coding vector of each word in the target composition based on the genre identification model to obtain an attention vector of each word in the target composition, and obtaining a representation vector of each sentence in the target composition based on the attention vector of each word in the target composition;

coding the representation vector of each sentence in the target composition based on the genre identification model to obtain a coding vector of each sentence in the target composition;

performing attention calculation on the coding vector of each sentence in the target composition based on the genre identification model to obtain an attention vector of each sentence in the target composition, and determining a chapter representation vector of the target composition based on the attention vector of each sentence in the target composition;

and classifying the discourse representation vectors of the target composition based on the genre identification model to obtain the genre of the target composition.

Optionally, the target composition includes a chapter structure modification result of the target composition in a chapter level modification result;

determining a scoring grading of the target composition in a scoring dimension of structural rigor, comprising:

and determining the grading of the target composition on the scoring dimension of strict structure according to the modification result of the target composition on the chapter structure.

Optionally, determining the scoring grade of the target composition in the scoring dimension of the literary sketch comprises:

determining a chapter expression vector of the target composition based on the target composition, the basic information of the target text and the modification result of the target composition at the sentence level;

and determining the grading of the target composition on the scoring dimension by classifying the discourse expression vectors.

A composition review device comprising: the system comprises a detection module, a correction module, a grading and grading determination module and a comment generation module;

the detection module is used for detecting whether the target composition to be reviewed is an abnormal composition;

the correcting module is used for correcting the target composition from a word level, a sentence level and a chapter level respectively when the target composition is not an abnormal composition so as to obtain corresponding correcting results of the target composition on the word level, the sentence level and the chapter level respectively;

the scoring and grading determination module is used for determining the scoring and grading of the target composition from a plurality of scoring dimensions so as to obtain the scoring and grading of the target composition on the plurality of scoring dimensions;

and the comment generating module is used for generating the comment of the target composition according to the grading of the target composition on the plurality of scoring dimensions.

A composition review device comprising: a memory and a processor;

the memory is used for storing programs;

the processor is used for executing the program to realize the steps of the composition review method.

A readable storage medium having stored thereon a computer program which, when executed by a processor, carries out the steps of the composition review method of any of the above.

According to the composition review method, the device, the equipment and the storage medium, whether the target composition to be reviewed is the abnormal composition is detected, if the target composition is not the abnormal composition, the target composition is revised from the word level, the sentence level and the chapter level respectively, and accordingly corresponding revision results of the target composition on the word level, the sentence level and the chapter level are obtained. The composition review method can automatically review the composition to be reviewed, and manual participation is not needed, so that the problems caused by manual participation are avoided, and the composition review method provided by the application not only can obtain the corresponding correction results of the target composition on the word level, the sentence level and the chapter level, but also can obtain the comments of the target composition on a plurality of review dimensions, namely the review results are rich, the rich review results can play a good guiding role for writers, and the user experience is good.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.

FIG. 1 is a schematic flow chart illustrating a composition review method provided in an embodiment of the present application;

fig. 2 is a schematic diagram of an example of shallow features to be extracted for composition of a treatise and other genres provided in an embodiment of the present application;

FIG. 3 is a diagram illustrating an example of the dimension of the word-level, sentence-level, and chapter-level endorsement and comment generation provided by an embodiment of the application;

fig. 4 is a topology structure diagram of a genre identification model provided in the embodiment of the present application;

fig. 5 is a schematic structural diagram of a GRU model provided in an embodiment of the present application;

FIG. 6 is a schematic flow chart illustrating the generation of comments of a target composition according to scoring and grading of the target composition in multiple scoring dimensions according to an embodiment of the present application;

FIG. 7 is a schematic structural diagram of a composition review device provided in an embodiment of the present application;

fig. 8 is a schematic structural diagram of composition review equipment provided in an embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In view of the problems of the existing manual review method, the inventor tries to provide an automatic review method, and therefore, the inventor conducts research, and the original thought is as follows: and (4) adopting a scoring model-based scoring method, namely training a scoring model by using the training composition marked with the score in advance, and then scoring the to-be-scored composition by using the scoring model obtained by training.

The inventor finds that the scoring model-based review method can only give a score for the composition to be reviewed, and the writer can only know about what level the composition is in by the score, but does not know about what problem the composition is, so the review result obtained by the scoring model-based review method has a limited guiding effect on the writer.

In view of the problems of the scoring model-based review method, the inventor further studies and finally provides a composition review method with a good effect through continuous research, the composition review method is suitable for any application scene needing to review the composition, the composition review method can realize automatic review of the composition, and a review result with strong guidance for writers can be obtained through the composition review method, and the basic concept of the composition review method is as follows: and modifying the target composition from the word level, the sentence level and the chapter level to obtain corresponding modification results of the target composition on the word level, the sentence level and the chapter level respectively, determining the grading grade of the target composition from a plurality of evaluation dimensions, and generating the comment of the target composition according to the grading grade of the target composition on the plurality of evaluation dimensions.

The composition review method provided by the application can be applied to electronic equipment with data processing capacity, and the electronic equipment can be a server on a network side and a terminal used by a user side, such as a PC, a notebook, a PAD, a smart phone and the like. The composition review method provided herein is next described by the following examples.

First embodiment

Referring to fig. 1, a schematic flow chart of a composition review method provided in an embodiment of the present application is shown, which may include:

step S101: whether the target composition to be reviewed is an abnormal composition is detected, if the target composition is not an abnormal composition, step S102a is executed, and if the target composition is an abnormal composition, step S102b is executed.

Specifically, the process of detecting whether the target composition to be reviewed is an abnormal composition may include: if the target composition meets at least one of the following conditions, determining that the target composition is an abnormal composition:

The first condition is that the target composition and the first text in the first condition are similar to each other by a preset similarity threshold, and most of the content of the target composition is the content of the text in the first condition.

The confusion degree of the sentence in the second condition can be determined based on a sentence-level language model established in advance, the confusion degree of the sentence represents the degree of 'sentence inadequacy', it needs to be explained that the occupation ratio of the sentence with the confusion degree larger than the preset confusion degree threshold value in the target composition is larger than the preset occupation ratio threshold value, and the occupation ratio of the sentence with the confusion degree larger than the preset confusion degree threshold value in the target composition is higher, under the condition, the target composition is judged to be an abnormal composition.

For the third condition, the method for detecting whether the sensitive information appears in the target composition may be: and searching whether the target composition has the sensitive words in the sensitive word list which is collected and sorted in advance, if the target composition has the sensitive words in the sensitive word list, determining that the target composition has sensitive information, and under the condition, judging that the target composition is an abnormal composition.

Step S102 a: and modifying the target composition from the word level, the sentence level and the chapter level respectively to obtain corresponding modification results of the target composition on the word level, the sentence level and the chapter level respectively.

S102 b: and generating an abnormal composition comment.

Specifically, if the target composition meets the first condition, the celebrity item type comment may be generated, optionally, the celebrity item type comment may include information (such as a name) of a text in the celebrity item material library, of which the similarity with the target composition is greater than a preset similarity threshold, if the target composition meets the second condition, the comment with the target composition being not satisfied/with a confusing sentence may be generated, optionally, a sentence with the confusing degree being greater than the preset confusing threshold (i.e., a sentence with "sentence not satisfied") may be marked out, if the target composition meets the third condition, the sensitive information type comment may be generated, optionally, the sensitive information type comment may include a sensitive word in the target composition, and optionally, the sensitive word in the target composition may be marked out.

Step S103: and determining the grading of the target composition from the plurality of scoring dimensions to obtain the grading of the target composition on the plurality of scoring dimensions.

In this step, "a plurality of review dimensions" are part or all of the set plurality of review dimensions.

Optionally, the comment dimensions in step S103 may be comment dimensions of one or more of the following four aspects: content, expression, structure, development, wherein:

the review dimensions for the content aspect may include one or more of the following review dimensions: the thought is healthy, the emotion type is satisfied, the theme is satisfied, and the content is full; the review dimensions of the expression aspects may include one or more of the following review dimensions: basic expression, normative Chinese characters, fluent language and accordant style; the review dimensions of the structural aspect may include: the structure is strict; developmental scoring dimensions may include: and (5) culture and harvest.

It should be noted that the multiple evaluation dimensions in step S103 may be the above 10 evaluation dimensions, or may be partial evaluation dimensions in the above 10 evaluation dimensions.

In the education field, in order to obtain a more targeted review result and better guide writers in consideration of different review dimensions concerned by different school paragraphs, the scoring grade of the target composition is determined from a plurality of review dimensions corresponding to the school paragraph to which the target composition belongs, so that the scoring grade of the target composition on the plurality of review dimensions corresponding to the school paragraph to which the target composition belongs is obtained. The following table shows the review dimensions of the various school segment concerns:

TABLE 1 review dimension of interest for each school segment

As shown in the table above, in the primary school grade 1-2, the thought health and the emotional type in the aspect of content are mainly considered, the basic expression application and the literary expression specification in the aspect of expression are realized, in the primary school grade 3-4, two evaluation dimensions of content enrichment and theme conformity are added in the aspect of content besides the requirement of meeting the basic requirement of the primary school grade 1-2, and in the primary school grade 5-6, two evaluation dimensions of language fluency and literacy are added. In the primary school section, the application of basic expression is no longer the focus of attention, and two evaluation dimensions of character conformity and strict structure are increased.

Step S104: and generating the comment of the target composition according to the grading of the target composition on a plurality of evaluation dimensions.

Optionally, the composition review method provided in this embodiment may further include: after the target composition is determined not to be the abnormal composition, before the comment of the target composition is generated according to the grading of the target composition on a plurality of evaluation dimensions, the grading of the whole target composition is determined. On the basis, the comment of the target composition can be generated according to the grading of the target composition on a plurality of scoring dimensions and the grading of the whole target composition.

The grading and grading implementation mode for determining the whole target composition is various:

in one possible implementation: the overall score of the target composition can be predicted based on a pre-established score prediction model, and the overall score grading of the target composition is determined according to the overall score of the target composition. The score prediction model is obtained by training a training composition marked with integral scores.

In another possible implementation: the score of the target composition and the score of the model essay in the model essay set can be predicted based on a pre-established score prediction model (the score can be positive or negative), the overall score of the target composition is predicted based on the score of the target composition and the model essay in the model essay set and the score of the model essay in the model essay set, and the overall score grading of the target composition is determined according to the overall score of the target composition. The model essay in the model essay set in the implementation mode and the target essay belong to the same theme, and the score prediction model is obtained by training the training essay and the model essay which belongs to the same theme as the training essay and is marked with the integral score.

Based on the difference between the target composition and the model essay in the model essay set and the score of the model essay in the model essay set, the process of predicting the overall score of the target composition comprises the following steps: aiming at each model essay in the model essay set, determining the score of the target essay according to the score of the model essay and the score difference between the target essay and the model essay, and taking the score of the target essay on the model essay; and (4) calculating the average value of the scores of the target composition on each model text in the model text set, and calculating the average value to be used as the integral score of the target composition.

In yet another possible implementation: predicting the integral score of the target composition as a first score of the target composition based on a pre-established score prediction model; predicting the score of a target composition and a model document in a model document set based on a pre-established score prediction model, and predicting the integral score of the target composition as a second score of the target composition based on the score of the target composition and the model document in the model document set and the score of the model document in the model document set; and determining the grading of the target composition on the basis of the first grading of the target composition and the second grading of the target composition.

It should be noted that, regardless of the above-mentioned scoring prediction model or the above-mentioned scoring prediction model, the idea of prediction is to obtain two aspects of features of the target composition, wherein on one hand, the two aspects are shallow features including word use features, and/or sentence-level evaluation features, and/or chapter structure analysis features, and on the other hand, the two aspects of features are deep semantic features, and vector representation of the target composition is obtained according to the two aspects of features of the target composition, and then, overall scoring or scoring prediction is performed according to the vector representation of the target composition. In addition, it should be noted that, considering that the discussion paper has uniqueness compared with other genres, the uniqueness of the discussion paper is mainly reflected in that it has certain requirements on chapter structures, and in view of this, when scoring or difference prediction is performed on the discussion paper, chapter structure characteristics need to be obtained, while for other genres, such as description, there is no more rigid requirement on chapter structures, but there is usually a certain requirement on expression modes, and therefore, there are characteristics related to expression modes, and of course, there are some characteristics in common. Fig. 2 shows a schematic diagram of an example of shallow features to be extracted for composition of discussion papers and other genres.

In view of the difference in characteristics used for score prediction or difference prediction of the composition of different genres, the genre of the target composition may be determined first before score prediction or difference prediction of the target composition. In addition, considering that different academic sections have different investigation contents on the composition, scoring prediction models and/or difference prediction models of different academic sections can be established, and when scoring prediction or difference prediction is carried out on the target composition, the scoring prediction models and/or difference prediction models of the academic sections to which the target composition belongs are adopted for prediction.

The composition review method provided by the embodiment of the application comprises the steps of firstly detecting whether a target composition to be reviewed is an abnormal composition, and if the target composition is not the abnormal composition, modifying the target composition respectively from the word level, the sentence level and the chapter level so as to obtain corresponding modification results of the target composition respectively on the word level, the sentence level and the chapter level. The composition review method provided by the embodiment of the application can automatically review the composition to be reviewed, and manual participation is not needed, so that the problems caused by manual participation are avoided.

Second embodiment

The present embodiment mainly deals with the "step S102 a: and modifying the target composition respectively from the word level, the sentence level and the chapter level to obtain corresponding modification results of the target composition respectively on the word level, the sentence level and the chapter level.

First, a process of correcting a target composition from a word level will be described.

The process of modifying the target composition from the word level comprises the following steps: the target composition is subjected to one or more (preferably more) of the following treatments: the method comprises the steps of language accuracy analysis, network expression retrieval, error punctuation identification and generation of a correction result corresponding to a target composition at a word level according to a processing result.

The language accuracy analysis comprises the steps of correcting errors of different characters, and/or detecting errors of grammar, and/or retrieving idioms, and/or detecting errors of ancient poetry. Preferably, the language accuracy analysis includes both typographic error correction, grammatical error detection, idiomatic error detection, and ancient poetry error detection.

The process of carrying out the cross word error correction on the target composition comprises the following steps: for each sentence in the target composition, executing:

step a1, detecting wrongly written characters in the sentence, and obtaining a candidate character set corresponding to the wrongly written characters.

Specifically, the process of detecting wrongly written characters in the sentence includes: aiming at each character in the sentence, the character in the sentence is respectively replaced by each character in a confusion character set corresponding to the character, each sentence after replacement forms a candidate sentence subset, and the confusion degree of the sentence and the confusion degree of the sentences in the candidate sentence set are calculated based on a pre-established statistical language model; based on the confusion of the sentence and the confusion of the sentences in the candidate sentence set, it is determined whether the word is a wrongly-written word. The confusing word set corresponding to a word may include the shape word, the pronunciation word and the homophone of the word.

Optionally, the statistical language model in this embodiment may be a deep neural network-based language model, such as an ELMO-based language model, a BERT-based language model, and the like, and these models may learn the syntactic semantic information of a language at a deeper level through a large number of parameters.

Since the manner of determining whether each word in a sentence is a wrongly written word is the same, this embodiment takes the word "shop" in the sentence "i see shop" as an example, and describes the process of determining whether the word "shop" is a wrongly written word:

assuming that a set of confusion words corresponding to "shop" is { e, p, m }, first replace "shop" in "i'm view of shop" with "e", "p", and "m", respectively, to obtain a subset of candidate sentences { i am watching tv, i am watching me watching tv }, after obtaining the subset of candidate sentences, calculate a perplexity of "i am watching shop" based on a statistical language model established in advance, and a perplexity of "i am watching tv", and "i am watching tv", calculated, the perplexity of "i am watching shop" is 55, the perplexity of "i am watching tv" is 23, the perplexity of "i am watching tv" is 61, the perplexity of "i watching me" is 42, since the perplexity of "i watching tv" and "i watching me" is higher than "shop watching tv", it can be determined that "shop" is a wrong word, the 'electricity' and the 'occupation' can be combined into a candidate word set corresponding to the wrongly written 'shop', and then the correct word corresponding to the 'shop' is determined from the 'electricity' and the 'occupation'.

Step a2, covering the wrong words in the sentence based on the pre-established mask language model, and predicting the probability that the words at the covered positions in the sentence are each candidate word in the candidate word set corresponding to the wrong words.

For the example "i see shop" above, the "shop" is masked based on a pre-established mask language model, predicting the probability that the word at the masked location is "electricity" and the probability of being "occupied".

Optionally, the mask language model in this embodiment may be a Bert-based mask language model.

Step a3, based on the predicted probability, determining the correct word corresponding to the wrongly written word from the candidate word set corresponding to the wrongly written word.

For the example "i see shop" described above, assuming that the probability that the word at the mask position is predicted to be "electricity" via step a2 is 0.99, and assuming that the probability that the word at the mask position is "occupied" is 0.13, it is determined that the correct word corresponding to the wrongly-written word "shop" is "electricity".

Step a4, correcting the wrongly written characters in the sentence into correct characters corresponding to the wrongly written characters.

For the example "i see shop," after determining that the correct word corresponding to the wrongly written word "shop" is "e," the "i see shop" is modified to "i see tv.

The process of syntax error detection of the target composition comprises the following steps: for each sentence in the target composition, executing:

and b1, obtaining the syntactic dependency characteristics of the sentence, and the characteristics of each word in the sentence, the participle characteristics of each word, the mutual information characteristics of each word and the part-of-speech characteristics of each word.

The word segmentation characteristic of a word is used for representing the word and which word is segmented into a word, the mutual information characteristic of a word is used for representing the co-occurrence condition of the word and the content before and after the word, and the part-of-speech characteristic of a word is used for representing the part-of-speech of the word in which the word is located.

Step b2, determining the context vector of each word in the sentence according to the acquired features.

Optionally, a Bert model and a long-and-short memory network LSTM (preferably bi-directional LSTM) may be used to obtain a context vector for each word in a sentence.

The Bert model is a pre-trained neural network model, the Bert model is pre-trained on a large-scale corpus by using a Transformer, so that the model learns general sentence representation, the parameters of the Bert model need to be finely adjusted in the training process of a grammar error detection task because the pre-trained Bert model learns the general sentence representation, so that the model learns the context representation related to the grammar error detection task, the LSTM is a recurrent neural network widely applied to a sequence task, a sequence can be well modeled to capture long-distance dependency information, and the bidirectional LSTM network represents bidirectional context information through splicing vectors represented by forward LSTM vectors and reverse LSTM vectors, so that the representation capability of the LSTM model is further enhanced.

Step b3, determining whether there is grammar error in the sentence according to the context vector of each word in the sentence, and determining the concrete grammar error when there is grammar error.

Alternatively, the context vector of each word in the sentence can be input into the CRF model for sequence labeling to perform syntax error detection. The CRF model can calculate sentence-level normalized probability so as to select a globally optimal labeling sequence, the output labeling labels also have a certain dependency relationship for the sequence labeling task, and the CRF algorithm can model the dependency relationship among the labeling labels by learning the transition probability among the labeling labels, so that the model is ensured to select a more reasonable labeling result during decoding.

For idiom type error detection, fuzzy matching algorithm can be adopted to carry out fuzzy matching on the target composition and idioms in a pre-constructed idiom library, and the idioms with writing errors in the target composition are determined through fuzzy matching results. Similar to idiom error detection, a fuzzy matching algorithm can be adopted to carry out fuzzy matching on the target composition and ancient poetry texts in a pre-constructed ancient poetry library, and the ancient poetry texts with writing errors in the target composition are determined through fuzzy matching results.

For network expression retrieval, network non-standard expressions appearing in the target composition can be retrieved based on a pre-constructed network non-standard expression resource table.

The punctuation errors generally include two types, one is punctuation errors of a format type, for example, punctuation continuous use, paired punctuation mismatch, whole punctuation missing, pause number and conjunctions common, conjunctions pre-misuse colon number, and the other is punctuation errors of a semantic type.

For punctuation errors in format classes, a rule-based error detection mode can be collected, specifically, a corresponding regular expression is written for each punctuation error in advance, so that a rule base for punctuation error detection is constructed, when punctuation symbols of a target composition are subjected to error detection, the punctuation symbols in the paragraphs are matched with rules in the rule base by taking the paragraphs as a unit, and if the punctuation symbols are matched with the rules, the corresponding punctuation symbols are determined to be the wrong punctuation symbols.

For punctuation errors of semantic classes, mainly solved are use errors of periods and commas related to semantics in texts, and the use errors of the periods and commas related to semantics cannot be solved through rules. For punctuation errors of semantic classes, the present embodiment provides the following recognition methods:

for each paragraph in the target composition, performing:

and c1, removing punctuation marks in the paragraph, and taking the paragraph with the punctuation marks removed as a target text.

And c2, predicting the label category corresponding to each word in the target text based on the pre-established punctuation prediction model.

The label type corresponding to a word is used for indicating whether a punctuation mark exists after the word, and the punctuation mark is of a certain type if the punctuation mark exists.

Specifically, the punctuation prediction model may include a coding module (such as MacBERT), a mapping module, and a classification module, wherein:

the coding module is used for each character c in the target text_iEncoding is performed to output a representative vector h for each word_i：

h_i＝MacBERT(c_i) (1)

Each word c in the target text_iIs represented by vector h_iInput mapping module for mapping the expression vector h of each word_iMapping to the space of the label category, outputting the mapping result, inputting the mapping result into a classification module, and calculating the probability distribution p of each word in the target text on the space of the label category by the classification module based on the softmax function and the input mapping result_i：

p_i＝Softmax(Linear(h_i)) (2)

Obtaining a probability distribution p of each word in the target text on a label category space_iThen, the probability distribution p of each word in the target text on the label category space can be determined_iAnd determining the label category corresponding to each word in the target text, namely the label category predicted for each word in the target text.

And c3, determining punctuation marks predicted for the target text according to the label category corresponding to each word in the target text.

And c4, determining the punctuation mark with error in the punctuation mark of the paragraph according to the punctuation mark in the paragraph and the punctuation mark predicted for the target text.

By comparing the symbols in the paragraph with punctuation predicted for the target text, punctuation symbols in the paragraph that are used incorrectly can be determined.

The correction result of the target composition at the word level can indicate the problems of the target composition at the word level, such as wrongly written characters, wrongly grammatical grammar, use of network non-standard expressions, wrongly written idioms, wrongly written ancient poems, wrongly used punctuations and the like.

Next, a process of correcting the target composition from the sentence level is introduced.

The process of modifying the target composition from sentence level includes: and performing graceful sentence recognition, and/or correction method recognition, and/or written sentence recognition on the target composition, and generating a correction result corresponding to the target composition at a sentence level according to the recognition result. Preferably, graceful sentence recognition, correction method recognition and written sentence recognition can be simultaneously performed on the target composition, and a correction result corresponding to the target composition at the sentence level can be generated according to the three recognition results.

Wherein, the process of graceful sentence recognition to the target composition comprises the following steps: and inputting each sentence in the target written text into a pre-established graceful sentence judgment model to obtain a judgment result of each sentence in the target written text. It should be noted that the grace sentence discrimination model is a binary classification model that classifies the input sentences as grace sentences/non-grace sentences, and specifically, the binary classification model that inputs each sentence in the target sentence as grace sentences/non-grace sentences extracts feature vectors for the input sentences first, and then determines whether the input sentence is a grace sentence or a non-grace sentence based on the extracted feature vectors.

There are many retrieval methods, which are commonly named as alignment, reference, metaphor and anthropomorphic, and the following describes the procedures of alignment retrieval identification, reference retrieval identification, metaphor retrieval identification and anthropomorphic retrieval identification of target composition.

The process of carrying out the ranking, retrieval and identification on the target composition comprises the following steps: firstly, constructing an inverted storage structure for a target composition according to paragraphs, respectively extracting candidate alignment sentences, filtering the candidate alignment sentences, distinguishing the candidate alignment sentences and the like on the basis, and then identifying complete alignment sentences by using mechanisms such as candidate alignment sentence segmentation, candidate alignment sentence completion and the like.

The process of identifying the reference and the retrieval of the target composition comprises the following steps: and searching each sentence in the target working text in a pre-constructed reference resource library, if the sentence is searched in the reference resource library, determining that the sentence is the reference sentence, and if the sentence is not searched in the reference resource library, determining that the sentence is not the reference sentence. The quoting resource library comprises classic sentences in common ancient poetry, dialect, modern celebrity languages, adage, poem and common songs and the like.

The process of metaphorical recognition of the target composition comprises the following steps: for each sentence in the target work text, a metaphor (such as 'like', 'Buddha', 'good-looking' and the like) is firstly screened from the sentence, then a first vector capable of representing the sentence component is obtained based on the metaphor recognition model and the screened metaphor, and a second vector representing the sentence (especially the part in front of the metaphor and the part behind the metaphor) is obtained, and whether the sentence is the metaphor is determined based on the metaphor recognition model and the obtained vector.

The process of anthropomorphic paraphrasing and identifying the target composition comprises the following steps: and aiming at each sentence in the target working text, inputting the sentence into a pre-established anthropomorphic sentence recognition model, and obtaining a recognition result which is output by the anthropomorphic sentence recognition model and used for indicating whether the sentence is an anthropomorphic sentence. After a sentence to be recognized is input into the anthropomorphic sentence recognition model, the anthropomorphic sentence recognition model firstly extracts features and then classifies the extracted features, so that a classification result indicating whether the input sentence is an anthropomorphic sentence or not is obtained.

The process of identifying the descriptive sentence of the target composition comprises the following steps: for each sentence in the target written sentence, the type of the writing of the sentence and whether the sentence is literary or not are identified based on a pre-established writing sentence identification model, that is, whether each sentence in the target written sentence is a certain writing sentence with literary or not can be identified based on the writing sentence identification model.

The descriptive sentence recognition model is obtained by training a training text with two labels, wherein one of the two labels is used for indicating the descriptive type of the training text, and the other label is used for indicating whether the training text is literary or not. Alternatively, the type of depiction of a sentence may be one of the following types of depictions: language, action, appearance, expression, mind, and scene. Optionally, the written sentence recognition model may include a semantic representation vector determination module (such as bi-directional LSTM) for determining a semantic representation vector of the input sentence, a text recognition module for determining whether the input sentence has a text based on the semantic representation vector of the input sentence, and a written type recognition module for determining a written type of the input sentence based on the semantic representation vector of the input sentence.

The result of the correction of the target composition at the sentence level can indicate whether or not the target composition uses a correction method and/or an elegant sentence and/or a written sentence, and when the correction method is used, it can indicate which correction method is used and which part of the target composition uses the correction method, and when the elegant sentence is used, it can indicate which part of the target composition is the elegant sentence, and when the written sentence is used, it can indicate which type of literary/non-literary written sentence is used, and it can indicate which part of the target composition is the literary/non-literary written sentence of the type.

And finally, introducing the process of correcting the target text from the chapter level.

The process of documenting the target from the chapter level comprises the following steps: and performing chapter structure analysis and/or theme analysis on the target composition, and generating batch modification results corresponding to the target composition at chapter levels according to analysis results. Preferably, the chapter structure analysis and the theme analysis can be simultaneously performed on the target composition, and the batch modification result corresponding to the target composition at the chapter level is generated according to the two analysis results.

For the chapter structure identification, the chapter structure of the target composition can be analyzed based on the set analysis rules, such as whether the target composition is open and has a question, whether the target composition is in head-to-tail correspondence, whether the structure is clear, whether the view is clear, and the like. For topic analysis, topic identification can be performed on a target composition based on a pre-established topic identification model, and on the basis, off-topic detection can be further performed, namely whether the topic identified for the target composition deviates from a specified topic is detected.

The topic identification model is a classification model in essence, and determines the topic to which the target composition belongs from the topics contained in the pre-constructed topic system.

The middle part of fig. 3 shows a schematic diagram of the modification of the target composition from the word level, sentence level, chapter level, and the modification process described above can obtain the corresponding modification results of the target composition at the word level, sentence level, and chapter level, and the corresponding modification results of the target composition at the word level, sentence level, and chapter level can enable the writer to know the information of the wrongly written characters, the language sickness, the irregular expression, the punctuation error, the graceful sentence being used, the dictionary manipulation used, the description expression mode used, the chapter structure, and the composition theme.

Third embodiment

The present embodiment mainly deals with the "step S103: and determining the grading of the target composition from the plurality of evaluation dimensions to obtain the grading of the target composition on the plurality of evaluation dimensions, and introducing the grading.

The assessment dimension of each aspect is provided according to the assessment specification of the college entrance examination composition (the college entrance examination composition requires to assess the quality of the student composition from the aspects of content, expression and development grade 3), preferably, as shown in fig. 3, the assessment dimension of the content aspect can comprise thought health, emotion type, subject to question and full content, the assessment dimension of the expression aspect can comprise basic expression, line specification, language fluency and text coincidence, the assessment dimension of the structure aspect can comprise strict structure, and the assessment dimension of the development aspect can comprise text acquisition. It should be noted that, in practical application, the scoring and grading of the target composition may be determined from all the above-mentioned scoring dimensions, or a part of the scoring dimensions may be selected from the above-mentioned scoring dimensions, the scoring and grading of the target composition may be determined from the selected scoring dimensions, and the scoring and grading of the target composition may be determined from which scoring dimensions may be selected according to specific requirements.

The process of determining the scoring of the target composition in each of the above-mentioned scoring dimensions is described next.

(1) Determining scoring profiles for target composition in four scoring dimensions in content

(1-1) grading the scoring of the target composition on the evaluation dimension of the thought health

The process of determining the grading of the scoring of the target composition on the scoring dimension of the thought health comprises the following steps: judging whether each sentence in the target writing text contains vulgar language; and determining the grading of the target composition on the evaluation dimension of the healthy thought according to the judgment result of each sentence in the target composition.

The process of determining whether each sentence in the target writing text contains a vulgar language may include: and inputting each sentence in the target writing text into a pre-established sentence classification model to obtain a classification result which is output by the sentence classification model and used for indicating whether each sentence in the target writing text contains vulgar language.

Illustratively, the evaluation dimension of the thought health comprises two grading grades, namely a first grade and a second grade, if the target composition comprises the vulgar language, the grading grade of the target composition on the evaluation dimension of the thought health is determined to be the second grade, and if the target composition does not comprise the vulgar language, the grading grade of the target composition on the evaluation dimension of the thought health is determined to be the first grade. It should be noted that the evaluation dimension of thought health including two grading levels is only an example, and the number of grading levels may be determined according to specific requirements in practical applications.

(1-2) grading the scoring of the target composition on the scoring dimension of emotion type

The process of determining the grading of the scoring of the target composition on the scoring dimension of the emotion type comprises the following steps: identifying the emotion type expressed by the target composition; and determining the grading of the target composition on the evaluation dimension of the emotion type according to the emotion type identification result of the target composition.

In one possible implementation manner, the emotion type expressed by the target composition can be determined based on an emotion classification method of an emotion dictionary, in another possible implementation manner, the emotion type expressed by the target composition can be determined based on an emotion classification model obtained through pre-training, and optionally, the emotion type expressed by the target composition can be one of positive emotion and negative emotion. Optionally, when determining the emotion type expressed by the target composition, the intensity of the emotion type expressed by the target composition may also be determined, and then the scoring level of the target composition in the sentiment type scoring dimension is determined according to the emotion type expressed by the target composition and the intensity of the emotion type expressed by the target composition.

(1-3) grading the rating of the target composition in the dimension of the score according with the theme

Determining the grading of the target composition on the scoring dimension of the corresponding theme, comprising the following steps: acquiring a representation vector of each word in the title of the target composition and a representation vector of each word in the text of the target composition, and determining a target vector corresponding to each word in the title of the target composition based on the acquired vectors, wherein the target vector corresponding to one word can represent the correlation between the word and each word in the text of the target composition; determining the matching degree of the title of the target composition and the text according to the target vector corresponding to each word in the title of the target composition, and determining the grading of the target composition on the grading dimension which is in line with the title according to the matching degree of the title of the target composition and the text.

It should be noted that the matching degree between the topic of the target composition and the text can represent the condition that the target composition conforms to the topic meaning, and the higher the matching degree between the topic of the target composition and the text is, the more the target composition conforms to the topic meaning.

(1-4) grading the rating of the target composition in the dimension of content enrichment

In view of the above, the present embodiment determines the grading level of the target composition in the dimension of content enrichment by combining the modification result with the above-mentioned classification result, because the modification result of the target composition includes some useful information, such as which sentences are graceful sentences, which sentences include revisions, which sentences include descriptions, and the like.

The process of determining the grading of the target composition on the scoring dimension of the content enrichment comprises the following steps: determining a chapter representation vector of the target composition based on the target composition, basic information of the target composition, a corresponding correction result of the target composition on a sentence level and a corresponding correction result of the target composition on a chapter level (especially a separation detection result of the target composition); and classifying the discourse expression vectors through a classifier to obtain the grading of the target composition on the content enrichment grading dimension. Wherein, the basic information of the target composition may include, but is not limited to, one or more of the following information: the part of speech of the word used in the target writing, the sentence pattern used in the target writing, the length of the target writing, the number of words in the target writing, the number of words appearing in the target writing, the number of paragraphs in the target writing, the number of sentences in the target writing, and the like.

Optionally, after obtaining the score grading of the target composition in the evaluation dimension of the full content based on the classifier, it may be determined whether the score grading obtained based on the classifier is reasonable based on the modification result (such as the number of the written sentences in the target composition) corresponding to the target composition in the sentence level, and if not, the score grading is corrected.

(2) Determining a scoring profile for a target composition over multiple scoring dimensions in expression

(2-1) grading the scoring of the target composition on the scoring dimension of basic expression

The basic expression aspect is mainly to judge whether the basic capability points (specifically words and sentences) of the composition are reasonable or not. The process of determining the grading of the scoring of the target composition in the scoring dimension of the basic expression can comprise the following steps: performing new word and/or common idiom recognition on the target composition based on the new word tables and/or common idiom libraries respectively corresponding to different school paragraphs to obtain the word use conditions of the target composition on different school paragraphs; carrying out sentence pattern recognition (such as exclamation sentences and question sentences) on the target composition based on the specified sentence patterns so as to obtain the sentence pattern use condition of the target composition; and determining the grading of the target composition on the scoring dimension of the basic expression according to the word use condition of the target composition on different learning sections and the sentence use condition of the target composition.

Optionally, when the scoring and grading of the target composition in the scoring dimension of the basic expression is determined according to the use condition of the words of the target composition on different school paragraphs and the use condition of the sentence of the target composition, the scoring and grading of the target composition in the use condition of the words may be determined based on the use condition of the words of the target composition on different school paragraphs, the scoring and grading of the target composition in the use condition of the sentence may be determined according to the use condition of the sentence of the target composition, and then the scoring and grading of the target composition in the scoring dimension of the basic expression may be determined according to the scoring and grading of the target composition in the use condition of the words and the scoring and grading of the target composition in the use condition of the sentence.

Optionally, the word usage may be set to three grading levels, which are a first level, a second level, and a third level, respectively, where the third level represents that the word usage is lower than the current learning segment, the second level represents that the word usage is synchronous with the current learning segment, and the first level represents that the word usage exceeds the current learning segment. The embodiment counts the usage number of the new words and/or idioms of each school paragraph of the target composition, and determines the grading of the target composition on the usage condition of the words based on the counted number and the set number threshold. It should be noted that, in order to obtain a reasonable grading, the number threshold should be set appropriately.

(2-2) grading the scoring of the target composition on the scoring dimension of the line specification

The process of determining the grading of the target composition on the scoring dimension of the line specification comprises the following steps: detecting whether the title, and/or paragraph, and/or network wording, and/or punctuation mark, and/or the character format of the target composition meet the specification to obtain the detection result of the target composition on the literary specification, and determining the grading of the target composition on the scoring dimension of the literary specification according to the detection result of the target composition on the literary specification.

For the detection of the title and the paragraph, the title and the segmentation result of the target composition can be obtained based on a text processing mode, and whether the title and the paragraph of the target composition meet the specification or not is determined according to the obtained title and the segmentation result; for the detection of the network expression, the network expression in the target composition text can be detected based on a pre-constructed network expression table, and whether the network expression of the target composition text meets the specification or not is determined according to the detection result; for the punctuation detection, whether the punctuation of the target composition is wrong or not can be detected based on rules or models (if the punctuation error detection is performed when the target composition is modified from the word level, the punctuation error detection result is directly used here); for detecting the genre format, the target genre (i.e. the genre of the target text) can be obtained, and whether the format book of the target text meets the format requirement of the target genre is detected.

(2-3) grading the scoring of the target composition in the scoring dimension of language fluency

Fluency of language is a dimension evaluated from the perspective of grammatical norms, fluent lines, and the like in composition. The process of determining the grading of the scoring of the target composition on the scoring dimension of language fluency comprises the following steps: extracting characteristic vectors from basic information of the target composition, statistical information of the results of the odd word error correction and the grammar error detection and combination information; and classifying the extracted feature vectors through a classifier to obtain the grading of the target composition on the scoring dimension of fluent language.

Wherein, the basic information of the target composition may include, but is not limited to, one or more of the following information: the part of speech of the word used in the target writing, the sentence pattern used in the target writing, the length of the target writing, the number of words in the target writing, the number of words appearing in the target writing, the number of paragraphs in the target writing, the number of sentences in the target writing, and the like. The statistical information of the wrongly written characters and the syntax error detection result of the target composition is obtained by counting wrongly written characters and syntax errors of the target composition, and may include: the number of wrongly written words, the number of grammar errors, the contents of wrongly written words, the positions of grammar errors, and the like, and the combination information of the target composition includes the collocation appearing in the target composition, punctuation mark combination among sentences, and the like.

It should be noted that the association between fluent language and grammar error is relatively close, and the grading of fluent language is greatly reduced under the condition of the occurrence of wrongly written characters or more language diseases, therefore, the statistical information of the results of error correction and grammar error detection of written characters is used as an important basis for determining the grading of the target composition on the grading dimension of fluent language.

Optionally, after the score of the target composition in the evaluation dimension of fluent language is obtained based on the classifier, whether the score obtained based on the classifier is reasonable or not can be determined based on statistical information of the error correction and syntax detection results of the target composition, and if the score is not reasonable, the score of the target composition in the evaluation dimension of fluent language is corrected.

(2-4) grading the scoring of the target composition on the scoring dimension of the cultural relics conforming to the cultural relics

The genre conformity is mainly to determine whether the genre of the target composition is consistent with the specified genre (for example, if the writer is required to write a discussion paper, the specified genre is the discussion paper).

The process of determining the grading of the scoring of the target composition on the cultural relic conformity dimension comprises the following steps: and identifying the genre of the target composition based on a pre-established genre identification model, and determining the grading of the target composition on the aspect that the genre accords with the grading dimension according to the fact that whether the genre of the target composition is consistent with the specified genre. The genre identification model is obtained by training a training composition marked with the genre.

The process of identifying the genre of the target composition based on the pre-established genre identification model comprises the following steps: and determining the probability that the genre of the target composition is the set each genre based on the genre identification model, and determining the genre of the target composition according to the probability that the genre of the target composition is the set each genre. Wherein, the set individual genres can include one or more of the following genres, but are not limited to the following genres: diary, letter, lecture, postreading, poetry, narrative, writer, scene, discussion, description, and the like.

More specifically, the process of determining the probability that the genre of the target composition is the set genre based on the genre identification model includes:

and d1, encoding each word in the target composition based on the genre identification model to obtain an encoding vector of each word in the target composition.

Referring to fig. 4, a topology of a genre recognition model is shown, and as shown in fig. 4, the genre recognition model includes a word encoding module, a word-level attention module, a sentence encoding module, a sentence-level attention module, and a classification module.

When the genre of the target composition is identified, the target composition is firstly subjected to sentence segmentation, then each sentence obtained by the sentence segmentation is subjected to word segmentation, each word obtained by the word segmentation is input into a word coding module of a genre identification model for coding, and the word coding module outputs a coding vector of each word in the target composition. The purpose of encoding a word is to map the word to a high-dimensional semantic vector space.

And d2, performing attention calculation on the coding vector of each word in the target composition based on the genre identification model to obtain an attention vector of each word in the target composition, and obtaining a representation vector of each sentence in the target composition based on the attention vector of each word in the target composition.

The method comprises the steps of inputting a coding vector of each word in a target composition into a word-level attention module, determining an attention weight of each word in the target composition by the word-level attention module, obtaining the attention vector of each word in the target composition based on the coding vector and the attention weight of each word in the target composition, and further obtaining a representation vector of each sentence in the target composition based on the attention vector and the attention weight.

W in FIG. 4_itT-th word, alpha, representing the ith sentence in the target composition_itThe expression w_itAttention weight of (1), in accordance with w_itAnd a is_itCan determine w_itAccording to the attention vector of each word in the ith sentence, a representation vector s of the ith sentence in the target sentence can be determined_i. U in FIG. 2_wRepresenting the query vector when the term-level attention weight is computed.

It should be noted that the attention calculation of the encoding vector of each word in the target composition is performed to highlight the important word information, so that the subsequent operation can focus on the important word information.

And d3, coding the representation vector of each sentence in the target composition based on the genre identification model to obtain a coding vector of each sentence in the target composition.

The sentence coding module in this embodiment may be a bidirectional GRU model, fig. 5 shows a structure of the GRU model, and the operation of the GRU model is as follows:

z_t＝σ(W_zx_t+U_zh_t-1+b_z) (4)

r_t＝σ(W_rx_t+U_rh_t-1+b_r) (6)

wherein x is_tInput representing the current time step, h_t-1Hidden vector representing last time step, z_tAnd r_tRespectively representing an update gate and a reset gate, W_z，U_z，b_zAnd W_r，U_r，b_rThe parameters of the update gate and the reset gate respectively,

and h_tAnd respectively storing the memory vector of the current time step and the hidden vector finally output by the current time step.

It should be noted that, the GRU model has two gates, which are an update gate and a reset gate, and this structural arrangement overcomes the problem that RNN cannot solve the remote dependence well, and has stronger characterization capability for longer sentences. To this end, the present embodiment uses a bidirectional GRU model to characterize the sentence (i.e. encode the sentence representation vector obtained in the previous step):

wherein e is_iA vector representation representing the ith word in the sentence.

And combining the results of the bidirectional GRU models to obtain a sentence coding result:

and d4, performing attention calculation on the coding vector of each sentence in the target composition based on the genre identification model to obtain an attention vector of each sentence in the target composition, and determining a chapter representation vector of the target composition based on the attention vector of each sentence in the target composition.

The encoding vector of each sentence in the target composition is input into the sentence-level attention module, the sentence-level attention module can determine the attention weight of each sentence in the target composition, and then the attention vector of each sentence in the target composition can be obtained based on the encoding vector and the attention weight of each sentence in the target composition, and further the chapter expression vector of the target composition can be obtained on the basis.

It should be noted that the purpose of performing attention calculation on the encoding vector of each sentence in the target composition is to further highlight important word information, so that the important word information can be focused when performing classification later.

And d5, classifying the chapter representation vectors of the target composition based on the genre identification model to obtain the probability that the genre of the target composition is the set genre.

Specifically, the chapter expression vectors of the target composition are input into a classification module of the genre identification model for classification. Optionally, keywords (such as titles) may be extracted from the target composition, the expression vectors of the keywords and the discourse expression vectors of the target composition are input into the classification module of the genre identification model for classification, and the classification effect of the model can be improved by adding the keyword information.

(3) Determining scoring profiles for structural scoring dimensions of a target composition

The structural evaluation dimension of the target composition mainly comprises the structural rigor evaluation dimension. Determining the grading of the scoring of the target composition on the scoring dimension of strict structure, comprising the following steps: and determining the grading of the target composition on the scoring dimension of strict structure according to the modification result of the target composition on the chapter structure.

(4) Determining scoring profiles for development scoring dimensions for a target composition

The evaluation dimension of the target composition in development mainly comprises the evaluation dimension of literary mining. If the composition has bright spots in any aspect of language expression, vocabulary, and expressions, it can be regarded as having characters.

The process of determining the grading of the target composition on the scoring dimension of the literary sketch comprises the following steps: determining a chapter expression vector of the target composition based on the target composition, the basic information of the target text and the modification result of the target composition at the sentence level; grading of the target composition on the scoring dimension is determined by classifying the discourse expression vectors. It should be noted that, when determining the chapter representation vector of the target composition based on the target composition, the basic information of the target text, and the modification result of the target composition at the sentence level, the chapter representation vector of the target composition may be determined based on the information related to the text mining (e.g., the number of good words, the word frequency of the words, the number of idioms used, the richness of the language, the organization arrangement of the paragraphs, etc.) in the three information, namely, the target composition, the basic information of the target text, and the modification result of the target composition at the sentence level.

Illustratively, the evaluation dimension of the literature can include three grades, namely a first grade, a second grade and a third grade, wherein the first grade represents the literature, the whole vocabulary and the sentences of the explanatory article are excellent, a plurality of graceful expressions are provided, the bright spots are very prominent, the second grade represents the better literature, the local vocabulary and the sentences of the explanatory article are better, the bright spots are general, the third grade represents the insufficient literature, the explanatory article lacks rich vocabulary, sentences and bright spots are insufficient.

Fourth embodiment

The present embodiment mainly deals with the "step S104: and generating a realization process of a comment of the target composition for introduction according to the grading of the target composition on a plurality of evaluation dimensions.

Referring to fig. 6, a schematic flow chart illustrating a process of generating a comment of a target composition according to scoring of the target composition in a plurality of review dimensions may include:

step S601: and determining the comment corresponding to the grading grade of the target composition on each grading dimension based on the corresponding relation among the pre-constructed grading dimension, the grading grade and the comment.

The following table shows an example of correspondence of the scoring dimensions, scoring grades and comments:

TABLE 2 correspondence between scoring dimensions, scoring profiles, and comments

Illustratively, the "multiple scoring dimensions" in step S104 include "meeting the subject" and "enriching the content", the scoring of the target composition in the scoring dimension of "meeting the subject" is classified into one, the scoring of the target composition in the scoring dimension of "enriching the content" is also classified into one, based on the correspondence between the scoring dimension, the scoring classification and the comment shown in the above table, the comment corresponding to the scoring of the target composition in the scoring dimension of "meeting the subject" is obtained as "the content of the composition has clear title meaning", the comment corresponding to the scoring of the target composition in the scoring dimension of "enriching the content" is obtained as "the content is abundant, and the personality of the person is clearly highlighted".

It should be noted that, in the correspondence relationship between the review dimension, the score and the comment, one score may correspond to one comment, or may correspond to multiple comments, for example, if the content in the table is enriched, two comments correspond to one document in the review dimension, in this case, if multiple comments exist in the correspondence relationship in the score of the target composition in a certain dimension, one comment may be randomly selected from the multiple comments, for example, if the score of the target composition in the "content enriched" review dimension is one document, and in the correspondence relationship in the table, one document in the "content enriched" review dimension corresponds to two comments, one document in the "content enriched" review dimension may be randomly selected from the two documents, and used as the comment corresponding to the score of the target composition in the "content enriched" review dimension.

Step S602: and generating the comment of the target composition according to the comment corresponding to the grading grade of the target composition on each evaluation dimension.

In a possible implementation mode, the comment corresponding to the grading grade of the target composition on each evaluation dimension can be directly used as the comment of the target composition; in another possible implementation manner, based on the comment template, the comments corresponding to the grading grades of the target composition in each evaluation dimension are spliced and combined by using the connecting words and/or the turning words, so that a section of complete comprehensive comment is obtained and is used as the comment of the target composition. The following is an example of a comprehensive comment:

"this composition writes so well that you really expect you to get the next composition to do so!

In the aspect of content, the thought of the text is positive, not only the title meaning of composition content is clear, but also the content is full, and the personality of the character is distinct and outstanding. Expression, more advanced words are used, exclamation sentence pattern expression emphasis is used, sentence expression is rich, and partial punctuation marks are not used properly.

Some promotion suggestions are given to you, it is critical to master the applicable objects of words, and in addition, question sentence patterns are remembered when expressing questions, and most importantly, the punctuations are required to meet the use norms.

Child, "know everything, read ancient book", we learn to draw nutrition from the book. "

In the above embodiment, it is mentioned that, in addition to determining the scoring grades of the target composition in the multiple scoring dimensions, the scoring grades of the whole target composition may also be determined, in this case, besides the comments corresponding to the scoring grades of the target composition in each scoring dimension, the comments corresponding to the scoring grades of the whole target composition may also be determined, when the comments corresponding to the scoring grades of the whole target composition are determined, the comments corresponding to the scoring grades of the whole target composition may be determined based on the pre-established corresponding relationship between the whole scoring grades and the comments, and finally, the comprehensive comment of the target composition may be generated based on the comments corresponding to the scoring grades of the target composition in each scoring dimension and the comments corresponding to the scoring grades of the whole target composition.

By the method, the comments with rich contents and strong instructive property can be obtained.

Fifth embodiment

The embodiment of the application also provides a composition review device, which is described below, and the composition review device described below and the composition review method described above can be correspondingly referred to each other.

Referring to fig. 7, a schematic structural diagram of a composition review device provided in an embodiment of the present application is shown, which may include: a detection module 701, a correction module 702, a score grading determination module 703 and a comment generation module 704.

The detection module 701 is used for detecting whether a target composition to be reviewed is an abnormal composition;

a modification module 702, configured to, when the target composition is not an abnormal composition, modify the target composition from a word level, a sentence level, and a chapter level, respectively, to obtain modification results corresponding to the target composition at the word level, the sentence level, and the chapter level, respectively;

a scoring and grading determination module 703, configured to determine scoring and grading of the target composition from multiple scoring dimensions to obtain scoring and grading of the target composition in the multiple scoring dimensions;

and a comment generating module 704, configured to generate a comment of the target composition according to the grading of the target composition on the multiple scoring dimensions.

In a possible implementation manner, the multiple review dimensions are multiple review dimensions corresponding to the school section to which the target composition belongs in multiple set dimensions;

the scoring and grading determination module 703 is specifically configured to determine, from the multiple scoring dimensions corresponding to the school segment to which the target composition belongs, the scoring and grading of the target composition, so as to obtain the scoring and grading of the target composition on the multiple scoring dimensions corresponding to the school segment to which the target composition belongs.

In one possible implementation, comment generating module 704 includes: a comment determining submodule and a comment generating submodule.

The comment determining submodule is used for determining the comments corresponding to the grading grades of the target composition on each scoring dimension based on the corresponding relation among the pre-constructed scoring dimensions, the grading grades and the comments;

and the comment generation submodule is used for generating the comment of the target composition according to the comment corresponding to the grading of the target composition on each evaluation dimension.

In one possible implementation, the composition review device may further include: and the overall grading and grading determination module.

And the integral grading determination module is used for determining the grading of the whole target composition.

The comment generating module 704 is specifically configured to generate the comment of the target composition according to the grading of the target composition on the multiple scoring dimensions and the grading of the whole target composition.

In one possible implementation, the overall score grading determination module may include: a first score determination sub-module, and/or a second score determination sub-module, and an overall score grading determination sub-module.

The first score determining submodule is used for predicting the integral score of the target composition based on a pre-established score prediction model to be used as the first score of the target composition;

the second score determining submodule is used for predicting the score difference between the target composition and the model essay in the model essay set based on a pre-established score difference prediction model, predicting the integral score of the target composition based on the score difference between the target composition and the model essay in the model essay set and the score of the model essay in the model essay set, and taking the integral score as the second score of the target composition, wherein the model essay in the model essay set and the target composition belong to the same theme;

and the overall grading determination submodule is used for determining the grading of the overall target composition based on the first grading of the target composition and/or the second grading of the target composition.

In a possible implementation manner, the detecting module 701 is specifically configured to determine that the target composition is an abnormal composition if the target composition meets at least one of the following conditions:

In one possible implementation, the wholesale module 702 includes: a word-level correction module, a sentence-level correction module, and a chapter-level correction module.

The word level correcting module is used for performing one or more of the following treatments on the target composition: analyzing language accuracy, retrieving network vocabularies, identifying error punctuations, and generating a correction result corresponding to the target composition at a word level according to a processing result; wherein the language accuracy analysis comprises wrongly written characters, error detection of grammar and/or idiom type and/or ancient poetry;

a sentence-level modification module, configured to perform one or more of the following processes on the target composition: graceful sentence recognition, high-level vocabulary statistics, modification method recognition, written sentence recognition, and generating a modification result corresponding to the target composition at a sentence level according to a processing result;

and the chapter-level correction module is used for performing chapter structure identification and/or theme identification on the target composition and generating a correction result corresponding to the target composition at the chapter level according to the identification result.

In a possible implementation manner, the word-level modification module, when performing error punctuation recognition on the target composition, is specifically configured to, for each paragraph in the target composition:

In a possible implementation manner, the word-level modifying module is specifically configured to, when performing the difference correction on the target composition, perform, for each sentence in the target composition:

In a possible implementation manner, the word-level modifying module, when detecting the wrongly-typed word in the sentence, is specifically configured to, for each word in the sentence:

In a possible implementation manner, the word-level modification module, when performing syntax error detection on the target composition, is specifically configured to, for each sentence in the target composition:

In one possible implementation, the plurality of comment dimensions are a plurality of comment dimensions for one or more of the following four aspects: content, expression, structure, development, wherein:

structural aspects of the review dimension include: the structure is strict;

developmental scoring dimensions include: and (5) culture and harvest.

In a possible implementation manner, the scoring grading determination module 703, when determining the scoring grading of the target composition in the scoring dimension of the thought health, is specifically configured to:

judging whether each sentence in the target writing text contains vulgar language, and determining grading of the target writing text on the evaluation dimension of the healthy thought according to the judgment result of each sentence in the target writing text;

the scoring and grading determination module 703 is specifically configured to, when determining the scoring and grading of the target composition in the scoring dimension of the emotion type:

identifying the emotion type expressed by the target composition, and identifying a result according to the emotion type of the target composition; determining the grading of the target composition on the scoring dimension of the emotion type;

the scoring and grading determination module 703, when determining the scoring and grading of the target composition in the scoring dimension of the subject, is specifically configured to:

acquiring a representation vector of each word in the title of the target composition and a representation vector of each word in the text of the target composition, and determining a target vector corresponding to each word in the title of the target composition based on the acquired vectors, wherein the target vector corresponding to one word can represent the correlation between the word and each word in the text of the target composition; determining the matching degree of the title of the target composition and the text according to the target vector corresponding to each word in the title of the target composition, and determining the grading of the target composition on the grading dimension which is in line with the meaning of the title according to the matching degree of the title of the target composition and the text; determining a rating grading of the target composition in the dimension of the content enrichment, comprising: determining chapter representation vectors of the target composition based on the target composition, the basic information of the target composition and corresponding correction results of the target composition on sentence level and chapter level respectively; and determining the grading of the target composition on the content enrichment scoring dimension by classifying the discourse representation vectors.

In a possible implementation manner, the scoring grading determination module 703 is specifically configured to, when determining the scoring grading of the target composition in the scoring dimension of the basic expression,:

performing new word and/or common idiom recognition on the target composition based on new word tables and/or common idiom libraries respectively corresponding to different school paragraphs to obtain word use conditions of the target composition on different school paragraphs; carrying out sentence pattern recognition on the target composition based on a specified sentence pattern to obtain the sentence pattern use condition of the target composition; determining grading of the target composition on a scoring dimension of basic expression based on word use conditions of the target composition on different school paragraphs and sentence use conditions of the target composition;

the scoring and grading determination module 703, when determining the scoring and grading of the target composition in the scoring dimension of the line specification, is specifically configured to:

detecting whether the title, and/or paragraph, and/or network expression, and/or punctuation mark, and/or discourse format of the target composition meet the specification; according to the detection result of the target composition on the literary rule, determining the grading of the target composition on the scoring dimension of the literary rule;

the scoring and grading determination module 703, when determining the scoring and grading of the target composition in the scoring dimension of the fluent language, is specifically configured to:

extracting feature vectors from the basic statistical information of the target composition, the statistical information of the results of the odd-character error correction and the grammar error detection and the matching and combination of the target composition; determining a grading grade of the target composition on a scoring dimension of the language fluency by classifying the extracted feature vectors;

the scoring and grading determination module 703 is specifically configured to, when determining that the genre of the target composition meets the scoring and grading in the scoring dimension:

recognizing the genre of the target text based on a pre-established genre recognition model, wherein the genre recognition model is obtained by training a training composition marked with the genre; and determining the grading of the target composition on the scoring dimension of the composition which accords with the genre according to the condition that whether the genre of the target composition is consistent with the specified genre or not.

The scoring and grading determination module 703 is specifically configured to, when identifying the genre of the target text based on a pre-established genre identification model:

coding each word in the target composition based on the genre identification model to obtain a coding vector of each word in the target composition; performing attention calculation on the coding vector of each word in the target composition based on the genre identification model to obtain an attention vector of each word in the target composition, and obtaining a representation vector of each sentence in the target composition based on the attention vector of each word in the target composition; coding the representation vector of each sentence in the target composition based on the genre identification model to obtain a coding vector of each sentence in the target composition; performing attention calculation on the coding vector of each sentence in the target composition based on the genre identification model to obtain an attention vector of each sentence in the target composition, and determining a chapter representation vector of the target composition based on the attention vector of each sentence in the target composition; and classifying the discourse representation vectors of the target composition based on the genre identification model to obtain the genre of the target composition.

In one possible implementation manner, the target composition comprises the target composition on a chapter structure in a chapter-level modification result;

the scoring and grading determination module 703, when determining the scoring and grading of the target composition in the scoring dimension of structural rigor, is specifically configured to:

In a possible implementation manner, the scoring and grading determination module 703, when determining the scoring and grading of the target composition in the scoring dimension, is specifically configured to:

determining a chapter expression vector of the target composition based on the target composition, the basic information of the target text and the modification result of the target composition at the sentence level; and determining the grading of the target composition on the scoring dimension by classifying the discourse expression vectors.

The composition review device that this application embodiment provided can treat automatically that the composition reviewed reviews the composition, because do not need artifical participation, consequently, the problem that artifical participation brought has been avoided, and, not only can obtain the mark-and-error result that the target composition corresponds in the word level respectively, the sentence level, the chapter level through the composition review device that this application embodiment provided, still can obtain the comment of target composition on a plurality of evaluation dimensions, it is comparatively abundant to review the result promptly, abundant review result can play fine guide effect to the writer, user experience is better.

Sixth embodiment

The embodiment of the present application further provides a composition review device, please refer to fig. 8, which shows a schematic structural diagram of the composition review device, and the composition review device may include: at least one processor 801, at least one communication interface 802, at least one memory 803, and at least one communication bus 804;

in the embodiment of the present application, the number of the processor 801, the communication interface 802, the memory 803, and the communication bus 804 is at least one, and the processor 801, the communication interface 802, and the memory 803 complete communication with each other through the communication bus 804;

the processor 801 may be a central processing unit CPU, or an application Specific Integrated circuit asic, or one or more Integrated circuits configured to implement embodiments of the present invention, or the like;

the memory 803 may include a high-speed RAM memory, and may further include a non-volatile memory (non-volatile memory) or the like, such as at least one disk memory;

wherein the memory stores a program and the processor can call the program stored in the memory, the program for:

Alternatively, the detailed function and the extended function of the program may be as described above.

Seventh embodiment

Embodiments of the present application further provide a computer-readable storage medium, which may store a program adapted to be executed by a processor, where the program is configured to:

Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A composition review method, comprising:

2. The composition review method according to claim 1, wherein the plurality of review dimensions are a plurality of review dimensions corresponding to the school section to which the target composition belongs among the set plurality of dimensions;

3. The composition review method of claim 1, wherein generating the target composition reviews according to the grading of the target composition scoring in the plurality of review dimensions comprises:

4. The composition review method according to claim 1, further comprising:

determining the grading of the whole target composition;

5. The composition review method of claim 4, wherein said determining the scoring profile of the target composition as a whole comprises:

6. The composition review method according to claim 1, wherein said detecting whether the target composition to be reviewed is an abnormal composition comprises:

7. The composition review method as claimed in claim 1, wherein the amending of the target composition from the word level comprises:

amending the target composition from sentence level, comprising:

8. The composition review method as claimed in claim 7, wherein the error punctuation recognition of the target composition comprises:

for each paragraph in the target article:

9. The composition review method as claimed in claim 7, wherein the process of correcting the target composition by the difference word comprises:

making each sentence in the text for the target:

10. The composition review method of claim 9, wherein said detecting wrongly written characters in the sentence comprises:

for each word in the sentence:

11. The composition review method of claim 7, wherein the process of syntax error detection of the target composition comprises:

making each sentence in the text for the target:

12. The composition review method of claim 1, wherein the plurality of review dimensions are a plurality of review dimensions for one or more of the following four aspects: content, expression, structure, development, wherein:

structural aspects of the review dimension include: the structure is strict;

developmental scoring dimensions include: and (5) culture and harvest.

13. The composition review method of claim 12 wherein determining a rating scale for the target composition in the review dimension of thought health comprises:

identifying the emotion type expressed by the target composition;

14. The composition review method of claim 12 wherein determining a rating scale for said target composition in the review dimension of said basal expression comprises:

15. The composition review method as claimed in claim 14, wherein said identifying the genre of the target composition based on the pre-established genre identification model comprises:

16. The composition reviewing method as claimed in claim 12, wherein said target composition includes a chapter-level approval result of said target composition on a chapter structure;

17. The composition review method of claim 12 wherein determining the scoring profile of the target composition in the reading dimension comprises:

18. A composition review device, comprising: the system comprises a detection module, a correction module, a grading and grading determination module and a comment generation module;

19. A composition review device, comprising: a memory and a processor;

the memory is used for storing programs;

the processor, for executing the program, implementing the steps of the composition review method as claimed in any one of claims 1 to 17.

20. A readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the composition review method as claimed in any one of claims 1 to 17.