Personalized service configuration method and system of education intelligent cloud platform

Abstract

The application discloses an individualized service configuration method and system of an educational intelligent cloud platform, and relates to the technical field of service configuration. The individualized service configuration method of the educational intelligent cloud platform obtains correction image data and writing interaction data of each writing task of each student user of the educational intelligent cloud platform; analyzes the correction image data of each writing task based on a pre-trained correction semantic analysis model to obtain writing transition characteristic values of the corresponding student user; analyzes feedback research characteristic values of the corresponding student user based on the writing interaction data of each writing task of each student user; and performs corresponding service configuration processing on each student user based on the writing transition characteristic values and the feedback research characteristic values, so that the platform can realize differentiated allocation and accurate intervention of service configuration, and the service configuration can be based on a more real learning track, thereby improving overall learning effect and individualized experience.

Description

Technical Field

[0001] This invention relates to the field of service configuration technology, specifically to a personalized service configuration method and system for an educational intelligent cloud platform. Background Technology

[0002] With the development of educational informatization and artificial intelligence technologies, intelligent educational platforms based on cloud computing and big data are gradually becoming important support tools for teaching and learning. These platforms can connect to multiple terminal devices via the internet, providing students with functions such as online writing training, automatic grading, feedback generation, and learning resource delivery. In writing training scenarios, students can upload their essays to the platform, which then uses optical character recognition and AI recognition to analyze and correct errors. Furthermore, intelligent educational cloud platforms are gradually developing towards intelligence, refinement, and personalization, ensuring grading efficiency while dynamically adapting to the different learning needs of students, thus driving educational services from uniformity to differentiation.

[0003] The limitations of existing technologies include at least the following problems: existing technologies lack the ability to track the dynamic evolution of students' writing abilities in continuous tasks, making it difficult to fully reflect students' true writing levels. For example, when a student scores low in a task due to a temporary poor performance, existing technologies often directly push low-level learning services based on this, but fail to identify the overall progress trend shown in multiple consecutive writing tasks. This can easily lead to a mismatch between service configuration and the student's actual development direction, making it difficult to adjust personalized learning service configuration measures in a timely manner when the student's writing ability improves or declines. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a personalized service configuration method and system for an educational intelligent cloud platform, which solves the problem that existing technologies struggle to track writing evolution, leading to a mismatch between service configuration and student development.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a personalized service configuration method for an educational intelligent cloud platform, comprising the following steps: acquiring the grading image data and writing interaction data of each student user's writing task for each specified educational intelligent cloud platform; analyzing the writing change feature values ​​of the corresponding student user based on the grading image data of each specified educational intelligent cloud platform's writing task for each specified educational intelligent cloud platform, and combining it with a pre-trained grading semantic parsing model; analyzing the feedback and learning feature values ​​of the corresponding student user based on the writing interaction data of each specified educational intelligent cloud platform's writing task for each specified educational intelligent cloud platform; and performing corresponding service configuration processing for each student user on the specified educational intelligent cloud platform based on the writing change feature values ​​and feedback and learning feature values.

[0006] Furthermore, the corrected image data specifically refers to the pixel value and two-dimensional coordinates of each pixel in the corrected image, and the corrected semantic parsing model includes a corrected recognition subnetwork, a corrected evaluation subnetwork, and a corrected annotation subnetwork.

[0007] Furthermore, the specific steps for analyzing the writing evolution feature values ​​of each student user on the educational intelligent cloud platform are as follows: Input the grading image data of each writing task of each student user on the educational intelligent cloud platform into the pre-trained grading semantic parsing model, and analyze the grading evaluation feature set of the corresponding student user, including the grading evolution feature set and the annotation evolution feature set; Based on the grading evaluation feature set of each student user on the educational intelligent cloud platform, analyze the writing evolution feature values ​​of the corresponding student user.

[0008] Furthermore, the specific steps for analyzing the grading and evaluation feature set of each student user on the educational intelligent cloud platform are as follows: In the grading recognition sub-network of the grading semantic parsing model, based on the grading image data of each writing task of each student user on the educational intelligent cloud platform, the grading evaluation text information and grading annotation text information of the corresponding writing task are extracted; In the grading evaluation sub-network of the grading semantic parsing model, based on the grading evaluation text information of each writing task of each student user on the educational intelligent cloud platform, the grading evolution feature set of the corresponding student user is extracted; In the grading annotation sub-network of the grading semantic parsing model, based on the grading annotation text information of each writing task of each student user on the educational intelligent cloud platform, the annotation evolution feature set of the corresponding student user is extracted.

[0009] Furthermore, the grading and evaluation sub-network includes an evaluation input layer, an evaluation feature extraction layer, and a task sequence association output layer. The specific steps for extracting the grading evolution feature set of each student user on the designated educational intelligent cloud platform are as follows: In the evaluation input layer of the grading and evaluation sub-network, the grading and evaluation text information of each writing task of each student user on the designated educational intelligent cloud platform is received and preprocessed; in the evaluation feature extraction layer of the grading and evaluation sub-network, based on the preprocessed grading and evaluation text information of each writing task of each student user on the designated educational intelligent cloud platform, the evaluation feature vector of the corresponding writing task is extracted; in the task sequence association output layer of the grading and evaluation sub-network, based on the evaluation feature vector of each writing task of each student user on the designated educational intelligent cloud platform, the grading evolution feature set of the corresponding student user is extracted.

[0010] Furthermore, the writing interaction data includes submission duration, feedback immersion duration, feedback response time difference, iterative submission ratio, and pause peak ratio. The specific steps for analyzing the feedback learning characteristic values ​​of each student user on the set education intelligent cloud platform are as follows: Based on the writing interaction data of each student user's writing task on the set education intelligent cloud platform, analyze the writing response characteristic set of the corresponding writing task, including writing efficiency characteristic value and feedback reading characteristic value; perform time-series analysis on the writing response characteristic set of each student user's writing task on the set education intelligent cloud platform to obtain the corresponding student user's feedback learning characteristic value.

[0011] Furthermore, the specific steps for analyzing the writing response feature set of each student user's writing task on the educational intelligent cloud platform are as follows: Based on the submission duration and iteration submission ratio of each student user's writing task on the educational intelligent cloud platform, analyze the writing efficiency feature value of the corresponding writing task; Based on the feedback immersion duration, feedback response time difference, and pause peak ratio of each student user's writing task on the educational intelligent cloud platform, analyze the feedback reading feature value of the corresponding writing task.

[0012] Furthermore, the specific steps of the time series analysis are as follows: A comprehensive analysis is conducted on the writing efficiency characteristic value and feedback study characteristic value of each writing task for each student user on the educational intelligent cloud platform to obtain the initial feedback study characteristic value of the corresponding writing task; based on the initial feedback study characteristic value of each writing task for each student user on the educational intelligent cloud platform, the study fluctuation characteristic value, study trajectory characteristic value, and study equilibrium characteristic value of the corresponding student user are analyzed and comprehensively analyzed to obtain the feedback study characteristic value of the corresponding student user.

[0013] Furthermore, the specific steps for configuring services for each student user on the educational intelligent cloud platform based on writing change characteristic values ​​and feedback training characteristic values ​​are as follows: Normalize the writing change characteristic values ​​and feedback training characteristic values ​​for each student user on the educational intelligent cloud platform; compare the normalized writing change characteristic values ​​and feedback training characteristic values ​​with several preset service configuration adjustment ranges; and take corresponding service configuration measures for each student user on the educational intelligent cloud platform based on the results of the analysis.

[0014] The personalized service configuration system of the educational intelligent cloud platform includes: a data acquisition module, used to acquire the grading image data and writing interaction data of each student user's writing task for each specified educational intelligent cloud platform; a writing change analysis module, used to analyze the writing change feature values ​​of the corresponding student user based on the grading image data of each student user's writing task for each specified educational intelligent cloud platform, combined with a pre-trained grading semantic parsing model; a feedback and learning analysis module, used to analyze the feedback and learning feature values ​​of the corresponding student user based on the writing interaction data of each student user's writing task for each specified educational intelligent cloud platform; and a service configuration feedback module, used to perform corresponding service configuration processing for each student user of the specified educational intelligent cloud platform based on the writing change feature values ​​and feedback and learning feature values.

[0015] The present invention has the following beneficial effects:

[0016] (1) The personalized service configuration method of the educational intelligent cloud platform can dynamically generate writing change feature values ​​and feedback training feature values ​​by continuously collecting the correction image data and writing interaction data of each student user's writing task. Based on this, the platform can dynamically push service configuration measures that match the development direction of the corresponding student user, thereby enabling the platform to achieve differentiated allocation and precise intervention of service configuration. This ensures that high-level students receive high-level challenges and that students with weak foundations receive targeted reinforcement, thereby improving the overall learning effect and personalized experience.

[0017] (2) The personalized service configuration method of the educational intelligent cloud platform introduces a semantic parsing model to conduct in-depth analysis of the parsing image data of each writing task, thereby accurately extracting the parsing evaluation text information and the parsing annotation text information. Under the collaborative parsing of the parsing evaluation sub-network and the parsing annotation sub-network, a parsing evolution feature set and an annotation evolution feature set are formed, thereby capturing the performance change trend of students in different tasks and generating writing change feature values. This enables the service configuration to be based on a more realistic learning trajectory and realizes the reliability of personalized service configuration.

[0018] (3) The personalized service configuration method of the educational intelligent cloud platform introduces writing interaction data and performs in-depth analysis, so that the student's task participation in each task can be accurately expressed and the feedback training feature value is generated by time-series processing. This enables the platform to dynamically identify the degree of student's absorption of feedback, and thus the platform can push personalized services to students to ensure that the service configuration can match the student's actual input level, thereby improving the overall learning effectiveness.

[0019] (4) The personalized service configuration system of the educational intelligent cloud platform accurately realizes the personalized service configuration of student users through the collaborative analysis between modules. The data acquisition module provides synchronous input of grading image data and writing interaction data. The writing change analysis module generates writing change feature values ​​that reflect the dynamic changes of students' writing ability based on the grading semantic analysis model. The feedback and training analysis module analyzes the students' learning engagement in the grading feedback process from the writing interaction data and generates feedback and training feature values. The service configuration feedback module outputs personalized service configuration measures based on the two types of feature values, thereby achieving a high degree of matching of service configuration, which enables the platform to provide more targeted learning support for students in different developmental states.

[0020] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0021] Figure 1 This is a flowchart of the personalized service configuration method for the educational intelligent cloud platform of the present invention.

[0022] Figure 2 This is a flowchart illustrating the specific steps involved in analyzing and setting the writing change characteristic values ​​for each student user on the educational intelligent cloud platform in the personalized service configuration method of the educational intelligent cloud platform of the present invention.

[0023] Figure 3 This is a schematic diagram illustrating the setting of comprehensive evaluation and annotation data for student user sequences in the personalized service configuration method of the educational intelligent cloud platform of the present invention.

[0024] Figure 4 This is a block diagram of the personalized service configuration system of the educational intelligent cloud platform of the present invention. Detailed Implementation

[0025] Please see Figure 1This invention provides a technical solution: a personalized service configuration method for an educational intelligent cloud platform, comprising the following steps: within a set sliding period (e.g., 3 or 5 writing tasks as a sliding period, and when the number of writing tasks does not meet the sliding period requirement, a preset conventional service configuration strategy is invoked to execute the default personalized service configuration for student users, such as pushing model essays, assigning basic tasks, etc.), the correction image data and writing interaction data of each writing task of each student user on the set educational intelligent cloud platform are obtained; based on the correction image data of each writing task of each student user on the set educational intelligent cloud platform, and combined with a pre-trained correction semantic parsing model, the writing change feature values ​​of the corresponding student users are analyzed; based on the writing interaction data of each writing task of each student user on the set educational intelligent cloud platform, the feedback and learning feature values ​​of the corresponding student users are analyzed; based on the writing change feature values ​​and feedback and learning feature values, corresponding service configuration processing is performed on each student user on the set educational intelligent cloud platform.

[0026] The specific steps for configuring services for each student user on the educational intelligent cloud platform based on writing change characteristic values ​​and feedback training characteristic values ​​are as follows: Normalize the writing change characteristic values ​​and feedback training characteristic values ​​for each student user on the educational intelligent cloud platform (i.e., calculate the minimum and maximum values ​​of these characteristic values ​​within the current sampling period, and perform a linear mapping by subtracting the minimum value from the original value and then dividing by the difference between the maximum and minimum values, unifying the values ​​to a standard range of 0 to 1); Then, compare the normalized writing change characteristic values ​​and feedback training characteristic values ​​of each student user on the educational intelligent cloud platform with several preset service configuration adjustment intervals. Each service configuration adjustment interval includes one writing change interval and one feedback training interval, and each service configuration adjustment interval corresponds to one service configuration measure.

[0027] Based on the judgment and analysis results, corresponding service configuration measures are adopted for each student user of the set-up education intelligent cloud platform. Specifically, based on the normalized writing change characteristic value and feedback training characteristic value of each student user of the set-up education intelligent cloud platform being within the preset service configuration adjustment range, the service configuration control processing is performed on each student user of the set-up education intelligent cloud platform, including but not limited to the following examples:

[0028] Interval Group 1 (Low Transition, High Training):

[0029] Writing ability fluctuation range: 0.0–0.3 (decline in writing ability);

[0030] Feedback on training range: 0.7–1.0 (high engagement in learning);

[0031] Measures: Recommend the student with sample essays and writing skills sections, reduce AI-assisted writing, encourage the student to upload and revise repeatedly, and have the teacher assign targeted essay tasks;

[0032] Interval Group 2 (Intermediate Change, Intermediate Training):

[0033] Writing ability improvement range: 0.4–0.6 (slight improvement in writing ability);

[0034] Feedback on training: 0.4–0.6 (slight increase in engagement);

[0035] Measures: Appropriately enable AI prompts and AI-assisted writing functions for the student, guide the student to participate in interactive comments on the essay platform, increase interest in writing, and at the same time retain the teacher's group assignments and corrections;

[0036] Interval Group 3 (High Transition, Low Training):

[0037] Writing performance improvement range: 0.7–1.0 (improvement in writing ability);

[0038] Feedback on training: 0.0–0.3 (decline in learning engagement);

[0039] Measures: Reduce the amount of automated AI-assisted writing for the student, focus on pushing interactive scenarios such as writing activities and group discussions, require the student to submit multiple times in the group and view the corrections, and strengthen their attention to the correction reports;

[0040] Interval Group 4 (High Transition, High Training):

[0041] Writing performance range: 0.7–1.0;

[0042] Feedback period: 0.7–1.0;

[0043] Measures: Utilize both writing skills and feedback effectively, then provide the student user with higher-level challenges, such as AI-generated imitation and multiple rounds of iterative submissions, while reducing basic prompts and encouraging independent exploration;

[0044] Interval Group 5 (Low Change, Low Training):

[0045] Writing variation range: 0.0–0.3;

[0046] Feedback / training range: 0.0–0.3;

[0047] Measures: Emphasize the use of AI prompts and AI-assisted writing to lower the writing threshold for this student, combine this with mandatory essay assignments in teacher groups to increase the frequency of student writing, and guide them to view the required steps of the correction report.

[0048] The image data to be corrected specifically includes the pixel value and two-dimensional coordinates of each pixel in the image. The semantic parsing model for correction includes a correction recognition subnetwork, a correction evaluation subnetwork, and a correction annotation subnetwork.

[0049] Specifically, such as Figure 2 As shown, the specific steps for analyzing the writing evolution feature values ​​of each student user on the educational intelligent cloud platform are as follows: Input the grading image data of each writing task of each student user on the educational intelligent cloud platform into the pre-trained grading semantic parsing model, and analyze the corresponding student user's grading evaluation feature set, including the grading evolution feature set (including coverage evolution feature value, feedback coordination feature value, and subdivision smoothing feature value) and the annotation evolution feature set (including annotation change ratio feature value, annotation distribution differential feature value, annotation load bias feature value, and annotation balance difference feature value); Based on the grading evaluation feature set of each student user on the educational intelligent cloud platform, analyze the corresponding student user's writing evolution feature values, specifically as follows:

[0050] The coverage evolution feature value, feedback coordination feature value, and subdivision smoothing feature value of each student user on the educational intelligent cloud platform are weighted and processed. The results are then mapped between 0 and 1 based on the Sigmoid function to obtain the comprehensive evaluation feature value of the corresponding student user (used to characterize the comprehensiveness of the student user's essay improvement needs in continuous writing tasks; the larger the value, the more improvement needs are reflected in the overall text). It should be noted that in this weighting process, the subdivision smoothing feature value is expressed as 1 / (1+subdivision smoothing feature value).

[0051] It should be noted that in this implementation example, the weighting coefficients for each parameter in the weighted processing can be obtained using sample entropy weighting. Taking the weighted processing process for obtaining the comprehensive evaluation feature value as an example, for instance, the coverage evolution feature value, feedback coordination feature value, and subdivision smoothing feature value are set for each student user of the educational intelligent cloud platform. Then, a directional inverse transformation is performed on the subdivision smoothing feature value of each student user, i.e., 1 / (1+subdivision smoothing feature value). After that, the corresponding information entropy value is extracted, and then the corresponding information entropy value is subjected to reciprocal suppression mapping. The function f(x) = 1 / (1+x) is transformed, such as 1 / (1+information entropy value of covering evolution feature value), and summed to obtain the information entropy sum value. The corresponding transformed information entropy values ​​are then compared with the information entropy sum value to obtain the weight coefficients corresponding to each parameter. It should be added that if all parameters are positively correlated with the weighted result during the weighting process, then there is no need to perform directional reverse transformation before performing entropy weight calculation. Directional reverse transformation is only performed when there is a parameter that is negatively correlated with the weighted result.

[0052] The annotation variation ratio, annotation distribution differential, annotation load bias, and annotation balance difference characteristics of each student user on the educational intelligent cloud platform are weighted and processed. The results are then mapped between 0 and 1 using the Sigmoid function to obtain the comprehensive annotation characteristic value of the corresponding student user (used to characterize the comprehensive degree of annotation errors of the student user in continuous writing tasks; the smaller the value, the less severe the overall error situation). This value is then combined with the comprehensive evaluation characteristic value of the corresponding student user to obtain the writing change characteristic value of the corresponding student user (used to characterize the degree of change of the student user's writing ability in continuous writing tasks; the larger the value, the stronger the writing ability, and vice versa).

[0053] The specific formula for calculating the writing change characteristic value of a specific student user on the educational intelligent cloud platform is as follows: ;in, To define the writing change characteristic values ​​for a specific student user on the educational intelligent cloud platform, To set comprehensive evaluation feature values ​​for a specific student user on the educational intelligent cloud platform, To define the comprehensive labeled feature values ​​for a specific student user on the educational intelligent cloud platform, These are the coordination coefficients stored in the database. The difference adjustment coefficients are stored in the database, and in this embodiment, the coordination adjustment coefficients are stored in the database. Comprehensive adjustment coefficient The values ​​were 0.784 and 0.284, respectively.

[0054] The following is a specific implementation example of calculating the writing change characteristic value of a student user on the educational intelligent cloud platform. The available data includes the comprehensive evaluation characteristic value and comprehensive annotation characteristic value of five randomly selected student users on the educational intelligent cloud platform, as detailed in Table 1 and... Figure 3 As shown:

[0055] Table 1. Example of Comprehensive Evaluation and Labeling Data for Student User Sequences on the Educational Intelligent Cloud Platform

[0056] Comprehensive evaluation feature value Comprehensive annotation feature values Student User 1 0.268 0.358 Student User 2 0.548 0.482 Student User 3 0.781 0.693 Student User 4 0.652 0.526 Student User 5 0.426 0.396

[0057] Coordination coefficients stored in the database The value is: 0.784;

[0058] Comprehensive adjustment coefficients stored in the database The value is: 0.284;

[0059] Substituting the data from Table 1 and the row number coefficient into the specific formula for calculating the writing change characteristic value of a student user on the educational intelligent cloud platform, we obtain:

[0060] The characteristic value of the writing change of the first student user on the educational intelligent cloud platform is set as 0.784×exp(-√(0.286×0.358))+ln(1+0.284×((1 / (1+0.286))+(1 / (1+0.358))))≈0.935;

[0061] Set the writing change characteristic value of the second student user on the educational intelligent cloud platform as 0.784×exp(-√(0.548×0.482))+ln(1+0.284×((1 / (1+0.548))+(1 / (1+0.482))))≈0.787;

[0062] The writing change characteristic value of the third student user on the educational intelligent cloud platform is set as 0.784×exp(-√(0.781×0.693))+ln(1+0.284×((1 / (1+0.781))+(1 / (1+0.693))))≈0.660;

[0063] Set the writing change characteristic value of the fourth student user on the education intelligent cloud platform as 0.784×exp(-√(0.652×0.526))+ln(1+0.284×((1 / (1+0.652))+(1 / (1+0.526))))≈743;

[0064] Set the writing change characteristic value of the fifth student user on the education intelligent cloud platform as 0.784×exp(-√(0.426×0.396))+ln(1+0.284×((1 / (1+0.426))+(1 / (1+0.396))))≈0.858.

[0065] The pre-training steps of the semantic parsing model are as follows:

[0066] The labeled dataset consists of grading image data, grading evaluation text data, and grading annotation text data from the educational intelligent cloud platform. The data was manually annotated by a team of educational experts based on historical essay task samples. The annotation content includes: the bounding box coordinates of the scoring area, grading evaluation area, and grading annotation area in each essay image; the ground truth text labels for each area (such as score characters, evaluation terms, and annotation category); and the feature ground truth labels corresponding to the grading evaluation and annotation descriptions (such as question coverage category, prompt word distribution, annotation error category, etc.). Each sample in the labeled dataset has complete annotations and has undergone preprocessing, including image size normalization, character encoding standardization, and invalid symbol removal. The preprocessed dataset is divided into training set, validation set, and test set, for example: 80% for training, 10% for validation, and 10% for testing.

[0067] The semantic parsing model for corrections is trained. Taking the correction recognition sub-network as an example, the correction image is input into a convolutional neural network region detection algorithm to extract multi-scale feature maps of the image. Candidate regions are generated through a region generation network. The classification branch outputs region category labels (score, evaluation, annotation), and the localization branch outputs bounding box coordinates. Based on this, an OCR recognition network is called to convert the pixels in the candidate regions into character encoding sequences, generating score strings, evaluation text strings, and annotation description strings. These are then compared with the real text labels in the labeled dataset. The cross-entropy loss function is used to minimize the character recognition error, achieving joint optimization of region segmentation and text recognition.

[0068] During training, optimization algorithms (such as the Adam optimizer or RMSProp) are used to update model parameters. Hyperparameters (such as learning rate, number of LSTM hidden layer units, batch size, etc.) are tuned using a validation set to ensure the convergence and stability of the training process. Finally, the generalization ability of the model is evaluated using a test set, with metrics including character recognition accuracy, evaluation feature extraction accuracy, and labeled feature extraction accuracy. After training, the optimal model parameters are saved for subsequent practical applications on the educational intelligent cloud platform, enabling online essay correction semantic parsing and service configuration.

[0069] In this implementation plan, a pre-trained semantic parsing model is used to parse the grading image data of each student user's writing task into a grading evolution feature set and an annotation evolution feature set. The grading evolution feature set comprehensively reflects the intensity of improvement needs revealed by the grading text, while the annotation evolution feature set reflects the error distribution and burden. Based on the grading evolution feature set and the annotation evolution feature set, corresponding feature values ​​are generated. Then, combined with the co-regulation coefficient, differential regulation coefficient, and smoothing regulation coefficient, a comprehensive analysis is performed. This allows the writing change feature values ​​to truly reflect the overall trend of students' changes in continuous writing tasks. At the same time, it can also grasp the overall trend of students' writing development when configuring services, thus giving the platform more basis for service configuration and enabling it to push more suitable services according to the overall direction of students' changes, thereby improving the targeting of services.

[0070] Specifically, the specific steps for analyzing the grading and evaluation feature set of each student user on the educational intelligent cloud platform are as follows: In the grading recognition sub-network of the grading semantic parsing model, based on the grading image data of each student user's writing task on the educational intelligent cloud platform, the grading evaluation text information and grading annotation text information of the corresponding writing task are extracted. Specifically, the grading recognition sub-network includes a grading input partitioning layer and a recognition output layer.

[0071] In the grading input segmentation layer of the grading recognition subnetwork, the grading image data of each student user's writing task on the designated educational intelligent cloud platform is received and processed. Specifically, based on the convolutional neural network region detection algorithm, the grading image is segmented to distinguish the score region, the grading evaluation text region, and the grading annotation region. That is, in the convolutional neural network region detection algorithm, multi-layer convolution and pooling operations are performed on the grading image to extract multi-scale feature maps representing image features. Then, based on the multi-scale feature maps, the regions in the image that may contain text are scanned through a region generation mechanism (such as using a region generation network) to obtain several candidate regions, each corresponding to a potential text information region. The candidate regions are then input into the classification and localization branch of the convolutional neural network, and the candidate regions are classified using a fully connected layer combined with a Softmax classifier, outputting their category labels (score region, grading evaluation text region, grading annotation region), and simultaneously outputting the bounding box coordinate information of the regions. Finally, redundant bounding boxes are removed based on the non-maximum suppression method to obtain the target region segmentation result of the grading image, namely the score region, the grading evaluation text region, and the grading annotation region.

[0072] In the recognition output layer of the correction recognition subnetwork, based on the correction image data of each student user's writing task on the designated educational intelligent cloud platform after partitioning, the corresponding correction evaluation text information and correction annotation text information are output. That is, based on the OCR network stored in the database, the pixels in the score area, the correction evaluation text area, and the correction annotation area are sequentially processed for character recognition to obtain the corresponding character encoding sequence. The score area is recognized as a sequence of numeric characters, the correction evaluation text area is recognized as a sequence of Chinese characters or alphabetic characters, and the correction annotation area is recognized as a sequence of misspelled words and annotation prompts. The recognized character codes are then sorted according to their spatial position within the image area. The relationship is sequentially concatenated. Characters within the same line are arranged in left-to-right coordinate order, and characters between different lines are arranged in top-to-bottom line order to form corresponding text strings. The scoring area forms a score string, the grading and evaluation text area forms an evaluation text string, and the grading and annotation area forms an annotation and explanation string. The above strings are then normalized, including removing extra spaces, standardizing the encoding format, and standardizing punctuation. The normalized strings are used as the output results of the grading and evaluation text information and the grading and annotation text information, respectively. The score string and the evaluation text string are merged to form the grading and evaluation text information, and the annotation and explanation string forms the grading and annotation text information.

[0073] In the grading evaluation subnetwork of the grading semantic parsing model, based on the grading evaluation text information of each student user's writing task on the designated educational intelligent cloud platform, the corresponding student user's grading evolution feature set is extracted; in the grading annotation subnetwork of the grading semantic parsing model, based on the grading annotation text information of each student user's writing task on the designated educational intelligent cloud platform, the corresponding student user's annotation evolution feature set is extracted, specifically as follows:

[0074] The annotation subnetwork includes an annotation input parsing layer, an annotation sequence perception layer, and an annotation output layer. In the annotation input parsing layer of the annotation subnetwork, the annotation text information of each student user's writing task on the set educational intelligent cloud platform is parsed to obtain the annotation feature vector of the corresponding writing task. That is, the annotation description strings in the annotation text information are identified one by one, and matched and classified based on a preset annotation category database (such as typos, punctuation errors, grammatical errors, etc.). At the same time, the number of annotations in each category is extracted, and the number of annotations in each category is summed to obtain the total annotation feature. The number of annotations in each category is compared with the total number of annotations, and the sample entropy is processed based on the result (Shannon entropy calculation method can be used) to extract the annotation distribution feature of each writing task (used to characterize the distribution degree of annotations among different categories. The larger the value, the more balanced the distribution of annotations among multiple categories, and the more dispersed the problems in the essay).

[0075] In the database, weight coefficients are preset for different annotation categories. For example, the weight of grammatical errors is higher than that of punctuation errors, and the weight of punctuation errors is higher than that of typos. The number of annotations in each category is multiplied by its corresponding weight coefficient, and the product results of all categories are summed to extract the annotation load features of each writing task (used to characterize the comprehensive error load of different categories of errors in this writing task).

[0076] The number of each type of annotation is read and the ratio of the number of annotations to the total number of annotations, i.e. the proportion of each type of annotation. The result is squared and summed, and then the difference is processed, i.e., 1 - the squared result is summed, in order to extract the annotation balance feature of each writing task (used to characterize the degree of distribution balance of different types of annotations in the overall error of the writing task). The annotation total feature, annotation distribution feature, annotation load feature, and annotation balance feature of each writing task are concatenated into an annotation feature vector.

[0077] In the annotation sequence perception layer of the annotation sub-network, based on the annotation feature vector of each student user's writing task on the educational intelligent cloud platform, the corresponding student user's annotation evolution feature set is extracted. That is, this layer uses LSTM, and LSTM updates its hidden state based on the annotation feature vector of each writing task, thereby capturing the evolution trend of annotation features in different writing tasks, such as:

[0078] For the total number of annotations in the annotation feature vector of each writing task, the total number of annotations of adjacent writing tasks are successively processed by ratio, and the results are weighted to obtain the annotation change ratio feature, which is used to characterize the degree of change in the number of annotations in consecutive writing tasks. The larger the value, the more it reflects the increasing trend of the total number of annotations between different writing tasks, and vice versa.

[0079] For the annotation distribution features in the annotation feature vector of each writing task, the annotation distribution features of the last writing task and the first writing task are subtracted, and the result is compared with the annotation distribution features of the first writing task to obtain the annotation distribution trend features. The annotation distribution features of adjacent writing tasks are subtracted in turn, and the root mean square is processed based on the results. The processed results are weighted with the annotation distribution trend features to extract the annotation distribution differential features, which are used to characterize the overall change of the annotation distribution of the student user over time in continuous writing tasks.

[0080] For the annotation load features in the annotation feature vector of each writing task, the annotation load features of adjacent writing tasks are successively processed by difference to obtain several sets of annotation load feature differences between adjacent writing tasks. Positive annotation load feature differences (i.e., the sum of all annotation load feature differences higher than or equal to 0) and negative annotation load feature differences (i.e., the sum of all annotation load feature differences less than 0, at which point the absolute value is removed) are extracted and ratio processed to extract annotation load bias features, which are used to characterize the relative bias of annotation load changes between the upward and downward directions in continuous writing tasks. The larger the value and the higher the value is than 1, the more the error burden increases as the task evolves. The smaller the value and the lower the value is than 1, the less the error burden decreases as the task evolves.

[0081] For the annotation balance feature in the annotation feature vector of each writing task, the maximum value, minimum value and mean value of the annotation balance feature are extracted respectively. The difference between the maximum value and the minimum value of the annotation balance feature is compared with the mean value of the annotation balance feature to extract the annotation balance difference feature. This feature is used to characterize the relative difference in the degree of annotation balance between different writing tasks in the continuous writing tasks of the student user. It can reflect the fluctuation of the annotation balance feature in the overall task sequence.

[0082] The annotation variation ratio feature, annotation distribution differential feature, annotation load bias feature, and annotation balance difference feature are activated by the Sigmoid function to obtain annotation variation ratio feature value, annotation distribution differential feature value, annotation load bias feature value, and annotation balance difference feature value between 0 and 1, which are used as the annotation evolution feature set.

[0083] The grading and evaluation subnetwork includes an evaluation input layer, an evaluation feature extraction layer, and a task sequence association output layer. The specific steps for extracting the grading evolution feature set of each student user on the designated educational intelligent cloud platform are as follows: In the evaluation input layer of the grading and evaluation subnetwork, the grading and evaluation text information of each writing task of each student user on the designated educational intelligent cloud platform is received and preprocessed. Specifically, the received grading and evaluation text information is processed by sentence segmentation. The text can be divided into several clauses based on separators such as periods, semicolons, and exclamation marks. Within each clause, word segmentation is performed to divide the continuous text into words or character sequences. For score strings, numerical processing is performed directly to obtain the total score of the essay. The sentence segmentation and word segmentation can be implemented using existing natural language processing methods, such as rule-based word segmentation, statistical word segmentation, or deep learning-based word segmentation methods.

[0084] In the evaluation feature extraction layer of the grading and evaluation subnetwork, based on the preprocessed grading and evaluation text information of each student user's writing task on the designated educational intelligent cloud platform, the evaluation feature vector of the corresponding writing task is extracted, specifically as follows:

[0085] For each pre-processed writing task, several words in the evaluation text are matched one by one with a pre-set set of keywords stored in the database (e.g., theme keywords include idea and main idea, structural keywords include beginning, transition, and ending, language keywords include word choice, expression, and sentence structure, and content keywords include details, arguments, and materials). If a match is detected with a keyword in the keyword set, the keyword is determined to be triggered, and the triggered keyword category is recorded. Finally, the number of triggered keyword categories is counted, and the number of each keyword category is compared with the total number of keyword categories. The results of the ratio processing are weighted to extract the evaluation coverage feature, which is used to characterize the comprehensiveness of the evaluation of writing performance. The larger the feature, the more comprehensive the problems in the essay.

[0086] Several sets of prompt words representing improvement needs are preset and stored in the database, such as need, should, insufficient, pending, and to be improved. The word sequence of the preprocessed grading evaluation text information is compared with the prompt word set one by one. If a matching prompt word is detected, it is counted to obtain the total number of improvement prompt words. The total score of the essay is compared with the full score of the essay for this writing task stored in the database, i.e., full score of essay - total score of essay. The total number of improvement prompt words is then compared with the result of the difference to extract the improvement feedback intensity feature, which is used to characterize the intensity level of the feedback. When the improvement feedback intensity feature value is large, it indicates that there are still many improvement prompts within the limited score improvement space.

[0087] The number of improvement prompts in each clause is counted and compared with the total number of words in each clause to obtain the prompt percentage of each clause. Simultaneously, the number of improvement prompts in each clause is compared with the total number of improvement prompts, and the results are used as the weights of the prompt percentages for the corresponding sentences. These weights are then used to extract the evaluation detail feature, which characterizes the level of detail in the feedback. The larger this feature is, the more numerous and scattered the essay problems are. The evaluation coverage feature, improvement feedback intensity feature, and evaluation detail feature are concatenated into an evaluation feature vector.

[0088] In the task sequence association output layer of the grading and evaluation sub-network, based on the evaluation feature vector of each student user's writing task on the educational intelligent cloud platform, the corresponding student user's grading evolution feature set is extracted. Specifically, this layer uses LSTM, which sequentially receives evaluation feature vector inputs from multiple writing tasks. The LSTM network processes the input sequence step by step through the mechanism of recurrent neural networks (RNN), updating the hidden state at each time step and combining the current input with the historical hidden state, thereby realizing the modeling of the dependency relationship between task sequences. Furthermore, the LSTM layer can selectively remember and forget historical information through gating mechanisms (including input gate, forget gate, and output gate), thereby capturing the long-term dependency relationship and short-term change pattern between different writing tasks, and extracting the grading evolution feature set reflecting the student's performance change trend in continuous writing tasks, such as:

[0089] For each student user's evaluation feature vector for each writing task, the evaluation coverage features of adjacent writing tasks are processed for trend analysis. For example, the absolute value of the difference between the evaluation coverage features of the first and second writing tasks is divided by the evaluation coverage feature of the second writing task. This is then weighted to obtain the coverage fluctuation feature. Simultaneously, the mean of the evaluation coverage feature is extracted and standardized with the coverage fluctuation feature (using the Min–Max interval standardization method). Based on the results, a weighted analysis is performed to extract the coverage evolution feature, which characterizes the range of issues involved in the student user's grading and evaluation in consecutive writing tasks. The larger the value, the wider the range of issues involved and the more obvious the fluctuation.

[0090] For each student user's evaluation feature vector for each writing task, the maximum value and mean value of the improvement feedback intensity feature are extracted and then processed together as 1 - (mean value of improvement feedback intensity feature / maximum value of improvement feedback intensity feature) to extract the feedback coordination feature. This feature is used to characterize the overall coordination of the feedback intensity of the student user in continuous writing tasks. The larger the value, the less the need for improvement in most writing tasks.

[0091] For each student user's evaluation feature vector for each writing task, a moving average is applied to extract the subdivision smoothing feature, which is used to characterize the smoothness of the distribution of feedback at the sentence level in continuous writing tasks. The coverage evolution feature, feedback coordination feature, and subdivision smoothing feature are then activated by the Sigmoid function to obtain coverage evolution feature values, feedback coordination feature values, and subdivision smoothing feature values ​​between 0 and 1, which are used as the feedback evolution feature set.

[0092] In this implementation plan, a semantic parsing model is used to identify and separate the various types of information originally mixed in the score area, evaluation area, and annotation area in the grading image. Combined with region detection and OCR recognition from a convolutional neural network, different contents are extracted as score strings, evaluation text, and annotations, thus avoiding omissions or ambiguities that may occur during manual reading. Secondly, the extracted content is transformed into corresponding features. Subsequently, with the help of the sequence perception capability of LSTM, the evolution process of these features in multiple tasks is linked together to capture the trajectory of changes in students' writing performance, thereby identifying the overall improvement trend and problem distribution of students over a period of time. This provides a more reliable basis for subsequent personalized service configuration. Finally, the platform can achieve more targeted dynamic adjustments in service configuration, which not only helps teachers save energy but also allows students to receive a guidance experience that is highly consistent with their actual learning status.

[0093] Specifically, the writing interaction data includes submission duration, feedback immersion duration, feedback response time difference, iterative submission ratio, and pause peak ratio. The specific steps for analyzing the feedback learning characteristics of each student user on the set-up education intelligent cloud platform are as follows: Based on the writing interaction data of each student user on the set-up education intelligent cloud platform for each writing task, analyze the writing response characteristic set of the corresponding writing task, including writing efficiency characteristic value and feedback reading characteristic value; perform time series analysis on the writing response characteristic set of each student user on the set-up education intelligent cloud platform for each writing task to obtain the corresponding student user's feedback learning characteristics.

[0094] The submission time value is the time difference between when a student enters the essay writing interface and when the essay is submitted. It can be obtained by performing a difference calculation on the task database of the education intelligent cloud platform, which records the system timestamp of the student's first entry into the writing task interface and the system timestamp of the final submission of the essay.

[0095] The feedback immersion time value is the cumulative time a student spends on the grading report interface. It can be obtained by recording the start and exit timestamps of the student opening the grading report interface in the interaction log database of the education intelligent cloud platform, and the difference between them is taken as the dwell time. If a student enters the grading report interface multiple times in one task, the total dwell time is summed to obtain the feedback immersion time value for that writing task.

[0096] The feedback response time difference is the time interval from the time the grading report is generated to the time when the student first opens the grading report interface. It can be obtained by recording the system timestamp of the grading result generation in the education intelligent cloud platform and the timestamp of the student's first click to enter the grading report interface of the essay in the interaction log, and performing a difference calculation on the two to obtain the feedback response time difference of this writing task.

[0097] The iterative submission ratio is the ratio between the number of times a student submits to the same essay task and the maximum number of submissions allowed for that task. It can be obtained by counting the number of submissions a student makes for the same essay task in the task database of the education intelligent cloud platform, reading the maximum number of submissions set for that essay task, and performing ratio processing on the two to obtain the iterative submission ratio for that writing task.

[0098] The pause peak ratio is the ratio between the maximum single dwell time and the average dwell time of all students during the process of browsing the grading report. It can be obtained by recording the timestamp interval of each dwell time of the student on the grading report interface in the interaction log of the education intelligent cloud platform, calculating the dwell time of each dwell time, extracting the maximum dwell time and the average dwell time of all dwell times, and performing ratio processing on the two to obtain the pause peak ratio of the writing task.

[0099] The specific steps for analyzing the writing response feature set of each student user's writing task on the educational intelligent cloud platform are as follows: Based on the submission time and iterative submission ratio of each student user's writing task on the educational intelligent cloud platform, analyze the writing efficiency feature value of the corresponding writing task. That is, standardize the submission time and iterative submission ratio of each student user's writing task (Min-Max interval standardization method can be used for processing), and perform weighted processing based on the standardization results to obtain the writing efficiency feature value of the corresponding writing task (used to characterize the student's execution efficiency in the writing and submission process; the larger the value, the more effectively the student can complete the task within a limited time and demonstrate a stronger ability to revise and improve). In this weighted processing, the submission time value is expressed as 1 / (1+standardized submission time value).

[0100] Based on the feedback immersion time, feedback response time difference, and peak pause ratio of each student user's writing task on the educational intelligent cloud platform, the feedback reading characteristics of the corresponding writing task are analyzed. That is, the feedback immersion time, feedback response time difference, and peak pause ratio of each student user's writing task are standardized (Min–Max interval standardization method can be used for processing). The standardized results are then weighted to obtain the feedback reading characteristics of the corresponding writing task (used to characterize the student's absorption depth when using the correction information). In this weighted processing, the feedback response time difference is expressed as 1 / (1 + standardized feedback response time difference).

[0101] The specific steps of the time series analysis are as follows: First, comprehensively analyze (i.e., weighted process) the writing efficiency characteristic values ​​and feedback study characteristic values ​​of each student user's writing task on the set educational intelligent cloud platform for each writing task to obtain the initial feedback study characteristic value of the corresponding writing task (used to characterize the student's learning engagement in a single writing task; the larger the value, the greater the learning engagement). Based on the initial feedback study characteristic value of each student user's writing task on the set educational intelligent cloud platform, analyze the corresponding student user's study fluctuation characteristic value (i.e., analyze the feedback study characteristic value of each writing task). The following features are used: variance processing and the result is used as the feature value; training trajectory feature value (i.e., the ratio of the difference between the feedback training feature value of the last and the first writing task within the sliding cycle to the feedback training feature value of the first writing task; the larger and positive the value, the higher the student participation; the larger and negative the value, the lower the student participation); and training equilibrium feature value (i.e., the mean of the feedback training feature value of each writing task is processed and the result is used as the feature value). A comprehensive analysis is then performed to obtain the corresponding student user feedback training feature value.

[0102] The specific formula for calculating the feedback and learning characteristic value of a specific student user on the educational intelligent cloud platform is as follows: ;in, To set feedback and learning characteristic values ​​for a specific student user on the educational intelligent cloud platform, To define the study fluctuation characteristic value for a specific student user on the educational intelligent cloud platform, The training fluctuation adjustment coefficient is stored in the database. To set the learning trajectory characteristic values ​​for a specific student user on the educational intelligent cloud platform, The training trajectory adjustment coefficients are stored in the database. To set the study balance characteristic value for a specific student user on the educational intelligent cloud platform, The training balance adjustment coefficient is stored in the database. Furthermore, in this implementation example, the training fluctuation adjustment coefficient is stored in the database. Training trajectory adjustment coefficient The training balance adjustment coefficients are taken as 0.385, 0.243, and 0.372 respectively.

[0103] In this implementation plan, the writing interaction data of each student user for each writing task is gradually transformed into feature values, and then linked together through time series analysis to obtain a feedback training feature value that reflects continuous learning input. This avoids biased conclusions caused by abnormal status of a single task and more stably captures the student's true input over a period of time. At the same time, the three types of adjustment coefficients set in the formula—fluctuation, trajectory, and equilibrium—allow the feedback training feature value to reflect both short-term fluctuations and overall directionality and stability. Ultimately, this enables the platform to more clearly identify whether a student is maintaining stable effort or showing signs of declining input, thereby providing students with accurate personalized service configurations.

[0104] Please see Figure 4 This invention provides a technical solution: a personalized service configuration system for an educational intelligent cloud platform, comprising: a data acquisition module for acquiring grading image data and writing interaction data for each writing task of each student user on the educational intelligent cloud platform; a writing change analysis module for analyzing the writing change feature values ​​of the corresponding student user based on the grading image data of each writing task of each student user on the educational intelligent cloud platform and in conjunction with a pre-trained grading semantic parsing model; a feedback and learning analysis module for analyzing the feedback and learning feature values ​​of the corresponding student user based on the writing interaction data of each writing task of each student user on the educational intelligent cloud platform; and a service configuration feedback module for performing corresponding service configuration processing on each student user on the educational intelligent cloud platform based on the writing change feature values ​​and the feedback and learning feature values.

[0105] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0106] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for configuring personalized services on an intelligent education cloud platform, characterized in that, Includes the following steps: Acquire the grading image data and writing interaction data of each student user's writing task for each specified educational intelligent cloud platform; Based on the grading image data of each student user's writing task on the educational intelligent cloud platform, and combined with a pre-trained grading semantic parsing model, the writing change feature values ​​of the corresponding student users are analyzed, specifically as follows: The image data of each student user's writing task on the educational intelligent cloud platform is input into a pre-trained semantic parsing model for grading. The model analyzes the corresponding student user's grading evaluation feature set, including the grading evolution feature set and the annotation evolution feature set, specifically: In the correction recognition subnetwork of the correction semantic parsing model, the correction image data of each student user's writing task on the set education intelligent cloud platform are used to extract the correction evaluation text information and correction annotation text information of the corresponding writing task. In the correction and evaluation subnetwork of the correction semantic parsing model, the correction and evaluation text information of each student user's writing task on the set education intelligent cloud platform is used to extract the corresponding student user's correction evolution feature set. In the correction annotation subnetwork of the correction semantic parsing model, the corresponding student user's annotation evolution feature set is extracted based on the correction annotation text information of each student user's writing task on the set education intelligent cloud platform. Based on the grading and evaluation feature set of each student user on the educational intelligent cloud platform, we analyze the writing change feature values ​​of the corresponding student users. Based on the writing interaction data of each student user's writing task on the educational intelligent cloud platform, including submission duration, feedback immersion duration, feedback response time difference, iterative submission ratio, and pause peak ratio, the feedback learning characteristic values ​​of the corresponding student users are analyzed, specifically as follows: Based on the writing interaction data of each student user in each writing task on the set education intelligent cloud platform, analyze the writing response feature set of the corresponding writing task, including writing efficiency feature value and feedback reading feature value. A time-series analysis was performed on the writing response feature set of each student user on the educational intelligent cloud platform for each writing task to obtain the corresponding student user's feedback and learning feature value, which is as follows: Based on the submission time and iteration submission ratio of each student user's writing task on the educational intelligent cloud platform, the writing efficiency characteristics of the corresponding writing task are analyzed. Based on the feedback immersion time, feedback response time difference, and pause peak ratio of each student user's writing task on the educational intelligent cloud platform, the feedback reading characteristics of the corresponding writing task are analyzed. Based on writing change characteristic values ​​and feedback training characteristic values, corresponding service configurations are performed for each student user on the educational intelligent cloud platform. Specifically: Normalize the writing change characteristic value and feedback and training characteristic value of each student user on the educational intelligent cloud platform. The normalized writing change characteristic value and feedback training characteristic value of each student user on the educational intelligent cloud platform are compared with several preset service configuration adjustment ranges for judgment and analysis. Based on the judgment and analysis results, corresponding service configuration measures are adopted for each student user of the educational intelligent cloud platform.

2. The personalized service configuration method for the educational intelligent cloud platform according to claim 1, characterized in that, The corrected image data specifically refers to the pixel value and two-dimensional coordinates of each pixel in the corrected image, and the corrected semantic parsing model includes a corrected recognition subnetwork, a corrected evaluation subnetwork, and a corrected annotation subnetwork.

3. The personalized service configuration method for the educational intelligent cloud platform according to claim 1, characterized in that, The grading and evaluation sub-network includes an evaluation input layer, an evaluation feature extraction layer, and a task sequence association output layer. The specific steps for extracting the grading evolution feature set of each student user on the educational intelligent cloud platform are as follows: In the evaluation input layer of the grading and evaluation subnetwork, the grading and evaluation text information of each student user's writing task on the set educational intelligent cloud platform is received and preprocessed. In the evaluation feature extraction layer of the grading and evaluation subnetwork, the evaluation feature vector of each writing task is extracted based on the grading and evaluation text information of each student user of the preprocessed setting education intelligent cloud platform for each writing task. In the task sequence association output layer of the grading and evaluation subnetwork, the grading evolution feature set of each student user is extracted based on the evaluation feature vector of each writing task of each student user on the set educational intelligent cloud platform.

4. The personalized service configuration method for the educational intelligent cloud platform according to claim 1, characterized in that, The specific steps of time series analysis are as follows: By comprehensively analyzing the writing efficiency characteristic value and feedback study characteristic value of each student user's writing task on the educational intelligent cloud platform, the initial feedback study characteristic value of the corresponding writing task is obtained. Based on the initial feedback learning characteristic values ​​of each student user's writing task on the educational intelligent cloud platform, the learning fluctuation characteristic values, learning trajectory characteristic values, and learning equilibrium characteristic values ​​of the corresponding student users are analyzed, and a comprehensive analysis is conducted to obtain the feedback learning characteristic values ​​of the corresponding student users.

5. A personalized service configuration system for an educational intelligent cloud platform, used to implement the personalized service configuration method for an educational intelligent cloud platform as described in any one of claims 1-4, characterized in that, include: The data acquisition module is used to acquire the grading image data and writing interaction data of each student user's writing task on the set educational intelligent cloud platform; The writing change analysis module is used to analyze the writing change feature values ​​of each student user based on the grading image data of each writing task of each student user on the set education intelligent cloud platform, and combined with the pre-trained grading semantic parsing model. The feedback and training analysis module is used to analyze the feedback and training characteristics of each student user based on the writing interaction data of each writing task of each student user on the set education intelligent cloud platform. The service configuration feedback module is used to perform corresponding service configuration processing for each student user on the educational intelligent cloud platform based on writing change feature values ​​and feedback training feature values.

Images