Intelligent composition correction and promotion system and method

Abstract

The invention discloses an intelligent composition correction and promotion system and method, and relates to the technical field of natural language processing and education, and the method comprises the steps: receiving a composition of a user, converting the composition into a processable format, carrying out the preprocessing, carrying out the feature extraction of the preprocessed composition through a language model, collecting the writing historical record of the user, and carrying out the verification of the writing historical record. Using a clustering analysis method and a K-nearest neighbor regression model to determine scoring dimensions of the composition, calculating a feature value of each composition scoring dimension, integrating the feature value of each composition scoring dimension to integrally score the composition, creating a learning template, analyzing a user writing historical record, and combining the composition overall score to obtain a score of the composition; generating a personalized learning path and a practice task; according to the method, the composition scoring dimension is selected by using the DBSCAN clustering analysis method, the limitation of traditional single-dimension scoring is overcome, and the overall quality of the composition can be reflected more comprehensively.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing and educational technology, and in particular to an intelligent composition correction and improvement system and method. Background Art

[0002] With the rapid development of information technology, the field of education has gradually introduced a variety of intelligent tools to improve teaching efficiency and student learning outcomes. Especially in terms of essay marking and writing skills improvement, the traditional manual marking method is not only time-consuming and laborious, but also difficult to provide comprehensive and personalized feedback. With the advancement of machine learning, especially deep learning technology, the use of large-scale text data training models can more accurately understand the content of essays and provide more detailed scoring and feedback.

[0003] Although existing methods have improved the accuracy of scoring to a certain extent, they still face several challenges. First, existing methods often focus on scoring in a single dimension, while ignoring the overall structure and logical coherence of the essay. Second, most models fail to fully consider the personalized needs of users and cannot dynamically adjust the learning path and practice tasks based on the user's historical writing records, resulting in limited learning effects. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an intelligent composition correction and improvement system and method to solve the problem that existing methods often focus on scoring in a single dimension while ignoring the overall structure and logical coherence of the composition.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides an intelligent composition correction and improvement method, which comprises: Receive the user's composition, convert it into a processable format and preprocess it, use the language model to extract features from the preprocessed composition, and collect the user's writing history; Use cluster analysis and K-nearest neighbor regression model to determine the scoring dimensions of the essay, calculate the eigenvalue of each scoring dimension, and comprehensively score the essay based on the eigenvalue of each dimension. Create learning templates, analyze user writing history, and combine the overall score of the composition to generate personalized learning paths and practice tasks; Monitor users in real time while they are practicing tasks, and use the Q-learning algorithm to dynamically adjust the user's learning path.

[0007] As a preferred solution of the intelligent composition correction and improvement method of the present invention, the following steps are specifically included: receiving the composition of the user, converting the composition into a processable format and performing preprocessing. Users can input their compositions in three ways: text, voice, and image. Text input is set as the default input method, and users can directly type the composition text through the web page. For voice input, ASR is called to convert the voice signal into composition text. For uploading a handwritten composition image, OCR is used to extract the composition text. The preprocessing includes removing stop words, word segmentation and part-of-speech tagging.

[0008] As a preferred solution of the intelligent composition correction and improvement method of the present invention, the language model is used to extract features of the preprocessed composition, and the user's writing history record is collected, which specifically includes the following steps: Use BERT to extract semantic features of the preprocessed composition; The semantic features include vocabulary information, grammatical information, article structure and emotional tendency; Use principal component analysis to compress semantic features into low-dimensional features to form a low-dimensional feature vector; The user's writing history records include all previously submitted compositions and corresponding scoring results and feedback opinions for all compositions.

[0009] As a preferred solution of the intelligent composition correction and improvement method of the present invention, the scoring dimension of the composition is determined by using a cluster analysis method and a K nearest neighbor regression model, which specifically includes the following steps: Use DBSCAN to cluster the low-dimensional feature vectors to obtain N clusters and the cluster center of each cluster; The low-dimensional feature vector is input into the K nearest neighbor regression model to obtain the predicted value of the cluster to which the composition belongs; Use SHAP value analysis to calculate the importance of each feature in the low-dimensional feature vector of each cluster center, and obtain the feature that contributes most to each cluster center; According to the meaning and contribution of the features, a scoring dimension name is generated for each cluster center, and finally the scoring dimension and the corresponding description of the scoring dimension are obtained to form a scoring dimension set.

[0010] As a preferred solution of the intelligent composition correction and improvement method of the present invention, wherein: the calculation of the characteristic value of each composition scoring dimension and the overall scoring of the composition based on the characteristic value of each composition scoring dimension specifically include the following steps: The characteristic values ​​of each essay scoring dimension are obtained by direct reference, combination and use of external models, and then normalized; Set adjustment parameters that control the slope and position of the scoring curve; Use the Sigmoid function and calculate based on the normalized composition scoring dimension feature value and adjustment parameters to obtain the score of each composition scoring dimension; Based on the score of each essay scoring dimension, the geometric mean is used in combination with the exponential function to calculate and obtain the overall score of the essay.

[0011] As a preferred solution of the intelligent composition correction and improvement method of the present invention, the following steps are specifically included: creating a learning template, analyzing the user's writing history, and combining the overall composition score to generate a personalized learning path and practice tasks: By calculating the average of the scoring results in the user's writing history, and obtaining the lowest performance part of the user's composition based on the average; Set learning goals and learning paths based on the lowest performance part of the user's composition and the overall score of the composition; The learning path refers to a series of progressive practice tasks; Create learning templates containing practice tasks, add classification labels and difficulty levels to each practice task, and combine them into a template library; The learning template including the practice task includes setting the goal of the task, completion standards and recommended learning resources; By calculating the matching degree between the user's learning path and the templates in the template library, the template with the highest matching degree is selected as the practice task.

[0012] As a preferred solution of the intelligent composition correction and improvement method of the present invention, real-time monitoring is performed when the user is performing the practice task, and the learning path is dynamically adjusted, which specifically includes the following steps: Each time a user completes a practice task, their submission and related metadata will be automatically collected; Said metadata includes completion time, proportion of correct responses; Calculate the user's task performance score based on the collected content and metadata, and map the task performance score to an immediate reward; Define the state space and action space according to the learning objectives, and initialize the Q value table; The state space includes the current practice task type and difficulty level, and the action space includes increasing practice intensity and decreasing practice frequency; Based on the immediate reward, the Q-learning algorithm is used to update the Q-value table until the Q-value no longer changes in the iteration, indicating that the best action selection strategy has been found.

[0013] In a second aspect, the present invention provides an intelligent composition correction and improvement system, comprising: The composition collection module receives the user's composition, converts the composition into a processable format and performs preprocessing, uses the language model to extract features from the preprocessed composition, and collects the user's writing history; The composition scoring module uses cluster analysis and K-nearest neighbor regression model to determine the scoring dimensions of the composition, calculates the characteristic value of each composition scoring dimension, and comprehensively scores the composition based on the characteristic value of each dimension; The composition improvement module creates learning templates, analyzes the user's writing history, and combines the overall composition score to generate personalized learning paths and practice tasks; The learning adjustment module monitors the user in real time while performing the practice task and uses the Q-learning algorithm to dynamically adjust the user's learning path.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the intelligent composition correction and improvement method described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the intelligent composition correction and improvement method as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: by clustering low-dimensional feature vectors using the DBSCAN clustering analysis method, groups with similar characteristics in the composition can be identified, thereby determining multiple scoring dimensions. This method not only overcomes the limitations of traditional single-dimensional scoring, but also more comprehensively reflects the overall quality of the composition. In addition, by creating a learning template and combining the user's historical writing records and the overall score of the current composition, personalized learning paths and practice tasks can be generated. This customized design fully takes into account the unique needs and development stage of each student, ensuring that the practice tasks provided are both challenging and effective in improving writing skills. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0018] Figure 1 This is a flow chart of the intelligent composition correction and improvement method in Example 1.

[0019] Figure 2This is a schematic diagram of the overall scoring of the composition in Example 1. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides an intelligent composition correction and improvement method, comprising the following steps: S1. Receive the user's composition, convert it into a processable format and perform preprocessing.

[0024] The specific steps include: S1.1. In order to improve the convenience and flexibility of user interaction, three different composition input methods are supported, namely text input, voice input and image input.

[0025] Text input is the default input method, and users can directly type the content of the composition through the web interface. This method is simple and direct and suitable for most users.

[0026] Language input refers to the user inputting the composition through voice, using automatic speech recognition (ASR) technology to convert the user's voice signal into text in real time. This method is particularly suitable for users who are not convenient to type or prefer to express themselves verbally, such as children or students with special needs.

[0027] Image input refers to users uploading handwritten composition pictures and extracting text from the pictures through optical character recognition (OCR) technology. This input method is very suitable for users who are accustomed to handwriting, especially for situations where paper homework needs to be corrected.

[0028] Convert all received essays into a processable plain text format.

[0029] Further explanation, supporting multiple input methods (text, voice, image) greatly improves user-friendliness. Different types of users can choose the most suitable input method according to their preferences, thus lowering the usage threshold. For example, visually impaired users may prefer voice input, while users accustomed to handwriting can choose image input.

[0030] S1.2. For the composition converted to pure text format, stop words are removed. Stop words refer to words that frequently appear in natural language processing but contribute little to semantic understanding, such as "de", "shi", "zai", etc. Removing these words can reduce noise and improve the effect of feature extraction; then the continuous text is segmented into meaningful word units. For example, "I like reading" will be segmented into: I / like / reading. This step is particularly important for non-space-separated languages such as Chinese because they lack clear word boundaries; finally, part-of-speech tagging is performed. Part-of-speech tagging is the process of assigning a grammatical role to each word, such as noun, verb, adjective, etc. This process helps to further understand the sentence structure and semantic relationship.

[0031] Further explanation, removing stop words reduces the interference of meaningless words to the model, making the key information more prominent and improving the accuracy of feature extraction. Word segmentation provides clear word boundaries for subsequent semantic understanding and feature extraction, especially important when dealing with non-space-separated languages such as Chinese. Part-of-speech tagging can better understand the sentence structure and semantic relationship, thus more accurately evaluating the quality of the composition. For example, when judging whether a sentence is complete or has grammar errors, part-of-speech tagging provides important basis.

[0032] S2. Use a language model to extract features from the preprocessed composition and collect the user's writing history records at the same time.

[0033] Specifically, it includes the following steps. S2.1. After completing the preprocessing, use the language model BERT to extract the lexical information, grammatical information, article structure, and sentiment tendency features of the composition. BERT is a deep learning model based on the Transformer architecture, which can capture bidirectional context information in the text, thus providing a richer semantic representation.

[0034] Lexical information means that BERT generates context-sensitive embedding vectors for each word through its multi-layer Transformer encoder structure. It can not only recognize the meaning of the word itself but also understand its meaning in a specific context.

[0035] Grammatical information means that by analyzing the lexical relationships in the sentence, BERT can identify complex grammatical structures, such as subject-predicate-object relationships, clause nesting, etc., which is crucial for evaluating the grammatical correctness and complexity of the composition.

[0036] Article structure refers to the fact that BERT can help identify the logical connections between paragraphs and the overall structure of the article. For example, it can distinguish between the introduction, body, and supporting arguments, which helps to evaluate the organization and coherence of the essay.

[0037] Sentiment tendency means that BERT can be used for sentiment analysis to identify the emotional tendency expressed in an essay (such as positive, negative, or neutral), which is of great significance for evaluating the emotional expression ability of an essay and the author's attitude.

[0038] S2.1. In order to reduce the computational complexity and improve the efficiency of subsequent analysis, the principal component analysis (PCA) technique is used to reduce the dimensionality of the high-dimensional semantic features extracted by BERT. PCA is a commonly used linear dimensionality reduction method that can convert high-dimensional features into low-dimensional features and concatenate them into low-dimensional feature vectors while retaining the main information of the original data as much as possible. Its expression is: ; in, represents a low-dimensional feature vector, represents principal component analysis, The matrix representing the original high-dimensional features, Indicates the number of dimensions after dimensionality reduction, usually selected =50 to retain the main information.

[0039] It is further explained that the high-dimensional semantic features extracted by BERT are compressed into low-dimensional feature vectors through PCA technology, which significantly reduces the computational complexity of subsequent analysis. This not only improves the operating efficiency, but also makes large-scale data processing possible. Despite the dimensionality reduction process, PCA is able to retain the main information of the original data and ensure the accuracy of feature extraction. The low-dimensional feature vector provides a solid foundation for subsequent clustering analysis and scoring models.

[0040] S2.2. Collect user writing history records. Each time a user submits an essay, it will be scored and the scoring results will be stored in the database. These scores not only reflect the quality of the essay, but also provide a basis for the formulation of personalized learning paths.

[0041] S3. Use cluster analysis method and K nearest neighbor regression model to determine the scoring dimensions of the composition, calculate the eigenvalues ​​of each scoring dimension of the composition, and comprehensively score the composition based on the eigenvalues ​​of each scoring dimension.

[0042] The specific steps include: S3.1. Use DBSCAN to analyze low-dimensional feature vectors Perform clustering and output several clusters. Each cluster represents a potential scoring dimension, and obtain the center of each cluster, that is, the cluster center, which is expressed as: ; in, represents the output cluster, Represents the DBSCAN algorithm in cluster analysis, Indicates the radius of the field, Indicates the minimum number of samples.

[0043] Suppose DBSCAN generates three main clusters, corresponding to narrative coherence, emotional expression intensity, and opinion diversity.

[0044] Further explanation: DBSCAN clustering can identify different types of essay samples and divide them into multiple clusters. Each cluster represents a specific writing style or ability level, thus achieving multi-dimensional essay evaluation. Compared with a single-dimensional scoring method, this method can more comprehensively reflect the quality of the essay.

[0045] S3.2. Use the K nearest neighbor regression model as the explanatory model and transform the low-dimensional feature vector Input into the K nearest neighbor regression model to get the prediction result of which cluster the essay belongs to. The expression is: ; in, Represents the predicted value of the KNN model, that is, the predicted result of which cluster the essay belongs to. represents the K nearest neighbor regression model; For the low-dimensional feature vector of each cluster center, the SHAP value formula is used to calculate the importance of the features in each low-dimensional feature vector, and its expression is: ; in, Indicates The importance of features, Represents the set of all features, that is, all dimensions of the low-dimensional feature vector, Represents a feature subset, which is a set composed of all features A part of the feature combinations selected in , that is, the feature combinations currently being considered, Representing feature subsets The size of , that is, the number of features contained, Represents the total number of features.

[0046] According to the SHAP value analysis results, find the feature that contributes most to each cluster center. For example, if the important feature of a cluster center is the semantic similarity between adjacent sentences, then the cluster may correspond to narrative coherence. If the important feature of a cluster center is the distribution of sentiment scores, then the cluster may correspond to the intensity of sentiment expression.

[0047] To further explain, SHAP value is a technique for explaining model output. Through SHAP value analysis, it is possible to clarify which features have the greatest impact on the score of each cluster center.

[0048] Based on the most important features and the meaning of the features in the SHAP value analysis results, a scoring dimension name is generated for each cluster center.

[0049] For example, there are 3 cluster centers, and the feature that contributes most to cluster center 1 is (semantic similarity between adjacent sentences), is interpreted as narrative coherence, reflecting the fluency of the composition in terms of timeline or logic; the feature that contributes most to cluster center 2 is (emotion score distribution), which is interpreted as the intensity of emotional expression, reflecting the performance of the composition in emotional transmission. The most contributing feature of cluster center 3 is (Topic Distribution Entropy), interpreted as the diversity of viewpoints, reflects the richness of the topic distribution in the composition.

[0050] Finally, the obtained scoring dimensions and their corresponding explanations form a scoring dimension set ,For example: ; Further explanation: According to the characteristics and importance of each cluster center, a specific name and description are generated for each scoring dimension. This customized scoring dimension design not only makes the scoring more transparent and easy to understand, but also provides strong support for the formulation of personalized learning paths.

[0051] S3.3. If a certain scoring dimension directly corresponds to a dimension in a low-dimensional feature vector, the value of this dimension can be directly used as the feature value.

[0052] If a rating dimension needs to be represented by combining multiple features, the feature value can be calculated by weighted summation or nonlinear combination. For example, the diversity of opinions needs to be obtained by combining the topic distribution entropy and the number of topics.

[0053] If a certain scoring dimension requires more complex calculations, you can call an external model to calculate the feature value. For example, for the intensity of sentiment expression, you can use the TextBlob model to extract the sentiment score.

[0054] Normalize the eigenvalues ​​of each scoring dimension to between [0,1] to facilitate subsequent calculations.

[0055] S3.4. Set the slope of the control scoring curve and position adjustment parameters , the larger The value will make the score more sensitive to changes, and adjusting the parameter Determines the position of the scoring curve.

[0056] The Sigmoid function is used to calculate the score of each essay scoring dimension based on the normalized feature value of the essay scoring dimension and the adjustment parameter. The expression is: ; in, Indicates The scores of the essay scoring dimensions, Indicates the corresponding The characteristic value of the essay scoring dimension, represents the adjustment parameter that controls the slope of the scoring curve, represents the tuning parameter that controls the position of the scoring curve; It is further explained that the Sigmoid function limits the score value to between 0 and 1, which ensures the rationality of the score result, facilitates the subsequent comprehensive score calculation, makes the score result smoother, avoids the influence of extreme values, and enhances the stability of the score.

[0057] According to the score of each essay scoring dimension, the geometric mean is used in combination with the exponential function to calculate and obtain the overall score of the essay, which is expressed as follows: ; in, Indicates the overall score of the essay. Indicates the number of essay scoring dimensions, Indicates The weight of the dimension.

[0058] It is further explained that the use of the geometric mean combined with the exponential function to calculate the overall score of the essay ensures a balanced consideration of the scores of each dimension and avoids the excessive influence of too high or too low a score in one dimension on the overall score.

[0059] S4. Create learning templates, analyze user writing history, and combine the overall composition score to generate personalized learning paths and practice tasks.

[0060] The specific steps include: S4.1. Create learning templates containing practice tasks. Each template describes the task objectives, completion criteria, and recommended learning resources in detail, and adds classification labels to each practice task, which are then merged into a learning template library.

[0061] For example, add classification labels (such as vocabulary expansion, grammar correction, etc.) and difficulty levels (beginner, intermediate, advanced) to each practice task.

[0062] The goal of the task is to clarify the goal of each task. For example, through this task, you will learn how to use synonyms effectively in your writing.

[0063] Define the completion criteria for each task, for example, successfully replacing at least 5 common words with more advanced synonyms.

[0064] Provide relevant learning resources, such as online courses, reference books, or video tutorials, to help users better understand and complete tasks.

[0065] Organize all practice tasks according to category labels and difficulty levels to form a template library.

[0066] It is further explained that as the user's writing level improves, the difficulty of the tasks in the template library can be adjusted to adapt to the user's changing learning needs.

[0067] S4.2. Extract all the essays submitted by the user in the past and their corresponding rating results from the database. For each rating dimension, calculate the average of all the essay ratings.

[0068] For example, if the user's ratings for vocabulary richness in the rating dimension in the past five essays were 0.7, 0.6, 0.8, 0.5, and 0.6 respectively, then the average score for the dimension is 0.64.

[0069] Compare the average scores of each dimension and find the lowest performing area. For example, if the average score of vocabulary richness is 0.64, and the average score of logical coherence, another scoring dimension, is 0.9, you can determine that vocabulary richness is the area where the user needs the most improvement.

[0070] Based on the identified weak links and the overall score of the essay, specific learning goals and a progressive learning path are set for users.

[0071] First, set a learning goal. For example, if the user's average score in vocabulary richness is low, set a goal to improve the score of this dimension to above 0.8.

[0072] Secondly, design a series of progressive tasks based on the set goals. These tasks should go from simple to complex, gradually guiding users to master the required skills. For example, the initial task may be to identify and replace common words with synonyms, and subsequent tasks may involve more complex vocabulary collocation and sentence structure.

[0073] Further explanation: according to the user's performance and overall score, scientific and reasonable learning goals are set for them, and a series of progressive learning paths are designed. This kind of planning not only takes into account the short-term improvement needs, but also lays a solid foundation for long-term development.

[0074] Finally, by calculating the matching degree between the user's current learning path and the templates in the template library, the template with the highest matching degree is selected as the practice task.

[0075] Further explanation: By calculating the matching degree between the user's learning path and the templates in the template library, the most suitable practice tasks can be tailored for each user. This method ensures that the tasks provided are both challenging and effective in promoting the user's progress.

[0076] S5. Real-time monitoring is performed when the user is performing practice tasks, and the learning path is adjusted dynamically.

[0077] The specific steps include: Collect user-submitted practice tasks and related metadata The user's task performance score is calculated based on the collected practice tasks and metadata, and its expression is: ; in, Indicates the user's performance score in the practice task. Represents an adjustment factor, which is used to take into account the impact of additional factors such as creativity or complexity.

[0078] The calculated task performance score is directly used as an immediate reward .

[0079] In order to use the Q-learning algorithm for dynamic adjustment, it is necessary to define the state space and action space according to the learning objectives, and set the initial values ​​of the Q value table to 0, which means that at the beginning, there is no preference for any action in any state.

[0080] The state space contains the current practice task type and difficulty level. For example, the state can be represented as grammar correction task, medium difficulty.

[0081] The action space contains possible action options, such as increasing the intensity of practice (providing more complex tasks) or decreasing the frequency of practice (giving more time for review). These actions are designed to help users gradually improve their abilities or consolidate what they have learned.

[0082] According to the immediate reward, the Q-learning algorithm is used to update the Q value table until the Q value no longer changes in the iteration, indicating that the best action selection strategy has been found, and its expression is: ; in, Indicates in status Take action The updated Q value of Indicates in status Take action The current Q value of represents the learning rate, Indicates immediate reward, represents the discount factor, Indicates that in the next state Take action Q value; When the Q value no longer changes during the iteration, it means that the best action selection strategy has been found. At this point, the user's learning path can be dynamically adjusted according to this strategy to ensure that the provided practice tasks always meet the user's actual needs and development stage.

[0083] It is further explained that the use of Q-learning algorithm to continuously explore the best action selection strategy can find the most effective learning path in a complex learning environment. As the user's writing level improves, the strategy is dynamically adjusted to ensure that the practice tasks provided always meet the user's actual needs and development stage, supporting their long-term development.

[0084] This embodiment also provides an intelligent essay correction and improvement system, including: The composition collection module receives the user's composition, converts the composition into a processable format and performs preprocessing, uses the language model to extract features from the preprocessed composition, and collects the user's writing history; The composition scoring module uses cluster analysis and K-nearest neighbor regression model to determine the scoring dimensions of the composition, calculates the characteristic value of each composition scoring dimension, and comprehensively scores the composition based on the characteristic value of each dimension; The composition improvement module creates learning templates, analyzes the user's writing history, and combines the overall composition score to generate personalized learning paths and practice tasks; The learning adjustment module monitors the user in real time while performing the practice task and uses the Q-learning algorithm to dynamically adjust the user's learning path.

[0085] This embodiment also provides a computer device, which is suitable for the case of the intelligent composition correction and improvement method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the intelligent composition correction and improvement method proposed in the above embodiment.

[0086] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0087] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the intelligent composition correction and improvement method proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0088] In summary, the present invention uses the DBSCAN clustering analysis method to cluster low-dimensional feature vectors, which can identify groups with similar characteristics in the composition, thereby determining multiple scoring dimensions. This method not only overcomes the limitations of traditional single-dimensional scoring, but also reflects the overall quality of the composition more comprehensively. In addition, by creating a learning template and combining the user's historical writing records and the overall score of the current composition, personalized learning paths and practice tasks can be generated. This customized design fully takes into account the unique needs and development stage of each student, ensuring that the practice tasks provided are both challenging and effective in improving writing skills.

[0089] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent composition correction and improvement method, characterized by: include, Receive the user's composition, convert it into a processable format and preprocess it, use the language model to extract features from the preprocessed composition, and collect the user's writing history; Use cluster analysis and K-nearest neighbor regression model to determine the scoring dimensions of the composition, calculate the characteristic value of each scoring dimension, and comprehensively score the composition based on the characteristic value of each scoring dimension. Create learning templates, analyze user writing history, and combine the overall score of the composition to generate personalized learning paths and practice tasks; Monitor users in real time while they are practicing tasks, and use the Q-learning algorithm to dynamically adjust the user's learning path.

2. The intelligent composition correction and improvement method according to claim 1, characterized in that: Receive the user's composition, convert it into a processable format and perform preprocessing, which specifically includes the following steps: Users input their compositions through text, voice and images. Text input is set as the default input method, and users directly type the composition text through the web page; Through voice input, ASR is called to convert the voice signal into composition text; By uploading a handwritten composition image, OCR is used to extract the composition text; The preprocessing includes removing stop words, word segmentation and part-of-speech tagging.

3. The intelligent composition correction and improvement method as claimed in claim 2, characterized in that: Use the language model to extract features from the preprocessed composition and collect the user's writing history records. The specific steps include: Use BERT to extract semantic features of the preprocessed composition; The semantic features include vocabulary information, grammatical information, article structure and emotional tendency; Use principal component analysis to compress semantic features into low-dimensional features to form a low-dimensional feature vector; The user's writing history records include all previously submitted compositions and corresponding scoring results and feedback opinions for all compositions.

4. The intelligent composition correction and improvement method as claimed in claim 3, characterized in that: The cluster analysis method and K nearest neighbor regression model are used to determine the scoring dimensions of the composition, which specifically includes the following steps: Use DBSCAN to cluster the low-dimensional feature vectors to obtain N clusters and the cluster center of each cluster; The low-dimensional feature vector is input into the K nearest neighbor regression model to obtain the predicted value of the cluster to which the composition belongs; Use SHAP value analysis to calculate the importance of each feature in the low-dimensional feature vector of each cluster center, and obtain the feature that contributes most to each cluster center; According to the meaning and contribution of the features, a scoring dimension name is generated for each cluster center, and finally the scoring dimension and the corresponding description of the scoring dimension are obtained to form a scoring dimension set.

5. The intelligent composition correction and improvement method as claimed in claim 4, characterized in that: The step of calculating the characteristic value of each composition scoring dimension and comprehensively scoring the composition based on the characteristic value of each dimension specifically includes the following steps: The characteristic value of each essay scoring dimension is obtained by direct reference, combination and use of external models, and then normalized; Set adjustment parameters that control the slope and position of the scoring curve; Use the Sigmoid function and calculate based on the normalized essay scoring dimension feature values ​​and adjustment parameters to obtain the score of each essay scoring dimension; Based on the scores of each essay scoring dimension, the geometric mean is used in combination with the exponential function to obtain the overall score of the essay.

6. The intelligent composition correction and improvement method as claimed in claim 5, characterized in that: Create a learning template, analyze the user's writing history, and combine the overall score of the composition to generate a personalized learning path and practice tasks. The specific steps include the following: Calculate the average of the scoring results in the user's writing history, and obtain the lowest performance part of the user's composition based on the average; Set learning goals and learning paths based on the lowest performance part of the user's composition and the overall score of the composition; The learning path refers to a series of progressive practice tasks; Create learning templates containing practice tasks, add classification labels and difficulty levels to each practice task, and combine them into a template library; The learning template including the practice task includes setting the goal of the task, completion standards and recommended learning resources; By calculating the matching degree between the user's learning path and the templates in the template library, the template with the highest matching degree is selected as the practice task.

7. The intelligent composition correction and improvement method according to claim 6, characterized in that: Real-time monitoring is performed when users are performing practice tasks, and the learning path is adjusted dynamically. The specific steps include the following: Each time a user completes a practice task, their submission and related metadata will be automatically collected; Said metadata includes completion time, proportion of correct responses; Calculate the user's task performance score based on the collected content and metadata, and map the task performance score to an immediate reward; Define the state space and action space according to the learning objectives, and initialize the Q value table; The state space includes the current practice task type and difficulty level, and the action space includes increasing practice intensity and decreasing practice frequency; Based on the immediate reward, the Q-learning algorithm is used to update the Q-value table until the Q-value no longer changes in the iteration, indicating that the best action selection strategy has been found.

8. An intelligent composition correction and improvement system, based on the intelligent composition correction and improvement method according to any one of claims 1 to 7, characterized in that: include, The composition collection module receives the user's composition, converts the composition into a processable format and performs preprocessing, uses the language model to extract features from the preprocessed composition, and collects the user's writing history; The composition scoring module uses cluster analysis and K-nearest neighbor regression model to determine the scoring dimensions of the composition, calculates the characteristic value of each composition scoring dimension, and comprehensively scores the composition based on the characteristic value of each dimension; The composition improvement module creates learning templates, analyzes the user's writing history, and combines the overall composition score to generate personalized learning paths and practice tasks; The learning adjustment module monitors the user in real time while performing the practice task and uses the Q-learning algorithm to dynamically adjust the user's learning path.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent composition correction and improvement method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent composition correction and improvement method described in any one of claims 1 to 7 are implemented.

Images