Automatic composition scoring and feedback method and system based on large model

Abstract

The invention discloses an automatic composition scoring and feedback method and system based on a large model. According to the method, a student composition image is converted into a text through super-resolution reconstruction and an OCR technology, and a composition text data set is constructed after calibration of an error correction model; then adopting a multi-expert distributed scoring mechanism to carry out multi-dimensional manual labeling on the composition text, and generating a standard scoring label; fine-tuning the large language model by using the tagged data, and constructing a composition scoring model with a multi-dimensional automatic scoring capability; and finally, generating a multi-dimensional scoring result for student compositions by using the composition scoring model, dynamically generating cue words together with text contents, and driving a large language model to output targeted improvement suggestions. According to the invention, targeted composition improvement feedback can be generated efficiently and objectively, and technical support is provided for personalized learning in an online learning platform.

Description

Technical Field

[0001] The present invention belongs to the field of generative artificial intelligence, and specifically relates to a method and system for automated composition scoring and feedback based on a large model. Background Art

[0002] Essay scoring and feedback has always been one of the core issues in the field of education. Traditional essay scoring methods usually rely on manual scoring, which is time-consuming and highly subjective, making it difficult to achieve fair and consistent evaluation. In recent years, with the development of deep learning technology, automated essay scoring systems based on large models have gradually become possible. These systems can automatically evaluate the quality of essays through natural language processing (NLP) models and provide personalized feedback to students.

[0003] However, existing automated essay scoring systems mainly focus on a single scoring dimension (such as grammar or content) and lack the ability to conduct comprehensive evaluations in multiple dimensions. At the same time, when generating personalized feedback, existing technologies often fail to provide in-depth and targeted suggestions based on students' specific writing problems, resulting in insufficient feedback effectiveness. In addition, existing systems usually rely on limited data sets and annotations, making it difficult to adapt to the scoring needs of various essay types and styles.

[0004] With the development of large pre-trained language models, pre-training based on large-scale corpora and then fine-tuning can significantly improve the accuracy and reliability of automatic scoring. These models can capture the deep rules in language, thereby improving the accuracy of scoring. At the same time, generative models perform well in natural language generation tasks, making personalized feedback generation possible.

[0005] However, despite the significant progress made by large models in the field of natural language processing, how to combine these models for efficient essay scoring and feedback generation, especially how to provide students with specific improvement suggestions, is still a problem that needs in-depth research. Therefore, how to combine the powerful capabilities of large models to provide more comprehensive, accurate and personalized essay scoring and feedback has become an important challenge in current technology. Summary of the invention

[0006] The purpose of the present invention is to solve the above-mentioned problems in the prior art and to provide a method and system for automated essay scoring and feedback based on a large model.

[0007] The specific technical solutions adopted by the present invention are as follows:

[0008] In a first aspect, the present invention provides a method for automated essay scoring and feedback based on a large model, comprising:

[0009] S1. Obtain a first data set consisting of student composition images, perform super-resolution reconstruction on the student composition images in the first data set, and then convert each student composition from the reconstructed image into text by optical character recognition (OCR) technology, and then input the converted text into an error correction model to correct recognition errors, thereby establishing a second data set consisting of student composition texts;

[0010] S2, dividing the second data set into multiple sub-data sets, randomly sending each sub-data set to multiple different expert scoring clients for experts to perform multi-dimensional scoring on each student composition text therein, performing quality control and result fusion on the multi-dimensional scoring data returned by all expert scoring clients, and obtaining a multi-dimensional scoring label for each student composition text in the second data set;

[0011] S3, using the second data set with the multi-dimensional scoring labels as a fine-tuning data set, fine-tuning the first language model to obtain a composition scoring model capable of performing multi-dimensional scoring on student composition texts;

[0012] S4. For the target composition image uploaded by the student client, it is converted into a target composition text and then input into the composition scoring model to obtain a multi-dimensional scoring result. The target composition text and the multi-dimensional scoring result are then used to replace the placeholders in the prompt word template to form a prompt word text for driving the model to generate multi-dimensional composition improvement suggestions. Finally, the prompt word text is input into the composition feedback model constructed based on the second largest language model, and the generated multi-dimensional composition improvement suggestions are fed back to the student client.

[0013] As a preferred embodiment of the first aspect, the error correction model adopts a BiLSTM-CRF model.

[0014] As a preferred embodiment of the above-mentioned first aspect, the student composition images in the first data set need to be input into a pre-trained dual-channel CNN model to evaluate image quality, and samples with image quality scores less than a threshold are eliminated.

[0015] As a preferred embodiment of the first aspect mentioned above, when performing quality control and result fusion on the multi-dimensional scoring data returned by all expert scoring clients, it is necessary to eliminate outliers for all multi-dimensional scoring data of the same student composition text, and then average the remaining multi-dimensional scoring data dimension by dimension to obtain the multi-dimensional average score as the multi-dimensional scoring label corresponding to this student composition text.

[0016] As a preferred embodiment of the above-mentioned first aspect, when fine-tuning the first large language model, the loss function is the mean square error between the multi-dimensional scoring result output by the model and the multi-dimensional scoring label obtained by expert scoring.

[0017] As a preferred embodiment of the first aspect above, the scoring dimensions adopted in the multi-dimensional scoring include grammar, content, fluency, logic and creativity.

[0018] In a second aspect, the present invention provides a system for automated essay scoring and feedback based on a large model, which comprises:

[0019] A training data construction module is used to obtain a first data set consisting of student composition images, perform super-resolution reconstruction on the student composition images in the first data set, and then convert each student composition from the reconstructed image to text using optical character recognition (OCR) technology, and then input the converted text into an error correction model to correct recognition errors, thereby establishing a second data set consisting of student composition texts;

[0020] A scoring label generation module is used to divide the second data set into multiple sub-data sets, each of which is randomly sent to multiple different expert scoring clients for experts to perform multi-dimensional scoring on each student composition text therein, and to perform quality control and result fusion on the multi-dimensional scoring data returned by all expert scoring clients to obtain a multi-dimensional scoring label for each student composition text in the second data set;

[0021] A model fine-tuning module, used to fine-tune the first language model using the second data set with the multi-dimensional scoring labels as a fine-tuning data set, so as to obtain a composition scoring model capable of performing multi-dimensional scoring on student composition texts;

[0022] The scoring feedback module is used to convert the target composition image uploaded by the student client into a target composition text and then input it into the composition scoring model to obtain a multi-dimensional scoring result, and then replace the placeholder in the prompt word template with the target composition text and the multi-dimensional scoring result to form a prompt word text for driving the model to generate multi-dimensional composition improvement suggestions. Finally, the prompt word text is input into the composition feedback model built based on the second largest language model, and the generated multi-dimensional composition improvement suggestions are fed back to the student client.

[0023] In a third aspect, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can implement the method of automated essay scoring and feedback based on a large model as described in any of the schemes of the first aspect above.

[0024] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for automated essay scoring and feedback based on a large model as described in any of the schemes of the first aspect above can be implemented.

[0025] In a fifth aspect, the present invention provides a computer electronic device comprising a memory and a processor;

[0026] The memory is used to store computer programs;

[0027] The processor is used to implement the method of automated essay scoring and feedback based on a large model as described in any of the solutions of the first aspect above when executing the computer program.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] The present invention can use a composition scoring model fine-tuned based on a large language model to automatically score composition texts in multiple dimensions, and then use the large language model to combine the scoring results with the composition content, automatically generate improvement suggestions for the composition, and help students improve specific composition dimensions. Compared with traditional manual scoring methods, the present invention has the advantages of high efficiency, objectivity, and scalability, and can provide more targeted feedback, helping students get timely improvement suggestions in writing and improve their writing skills. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic diagram of the steps of the method for automated essay scoring and feedback based on a large model;

[0031] Figure 2 This is a schematic diagram of the system modules for automated essay scoring and feedback based on a large model;

[0032] Figure 3 A schematic diagram of the components of computer electronic equipment.

[0033] Figure 4 Schematic diagram of the interaction between the online learning platform and the client. DETAILED DESCRIPTION

[0034] In order to make the above-mentioned purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation mode of the present invention is described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. The technical features in each embodiment of the present invention can be combined accordingly without conflicting with each other.

[0035] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for the purpose of distinguishing descriptions, and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features.

[0036] In a preferred embodiment of the present invention, a method for automated essay scoring and feedback based on a large model is provided, which includes:

[0037] S1. Obtain a first data set consisting of student composition images, perform super-resolution reconstruction on the student composition images in the first data set, and then convert each student composition from the reconstructed image into text through optical character recognition (OCR) technology, and then input the converted text into an error correction model to correct OCR recognition errors, thereby establishing a second data set consisting of student composition texts.

[0038] It should be noted that the first dataset can be constructed based on handwritten composition images uploaded by students themselves or by batch scanning of handwritten composition images on paper. In addition, super-resolution reconstruction of student composition images can be achieved based on deep learning models (such as ESRGAN, SRCNN), or by using traditional interpolation algorithms (such as bicubic interpolation).

[0039] In addition, the process of converting each student composition from the reconstructed image after super-resolution reconstruction to text by using (OCR technology belongs to the prior art, but recognition errors are prone to occur when performing OCR recognition on handwritten text. Therefore, the present invention further introduces an error correction model to correct recognition errors in OCR recognized text. In an embodiment of the present invention, a BiLSTM-CRF model can be used as an error correction model to correct OCR error text.

[0040] In addition, it should be noted that, since some handwritten composition images themselves have image quality problems or the handwriting of young authors is difficult to recognize, it is preferred in the present invention to screen the student composition images in the first data set. In an embodiment of the present invention, the student composition images in the first data set can be input into the pre-trained dual-channel CNN model one by one to evaluate the image quality, and samples with image quality scores less than the threshold are eliminated. The dual-channel CNN model can evaluate image quality from different dimensions. For example, the dual-channel CNN model is used to evaluate handwriting coherence (stroke connectivity>85%) and layout regularity (text area offset<5%). When the score of any of the two indicators is low, the composition image sample needs to be eliminated to avoid quality problems in the composition text obtained by recognition. Such composition texts with quality problems will introduce unqualified large model fine-tuning samples when sent to experts for scoring. Only qualified images are converted into text data using OCR technology, which can ensure that the composition scoring model obtained by subsequent fine-tuning can accurately generate real scores for the composition content.

[0041] S2. Divide the second data set into multiple sub-datasets, and randomly send each sub-dataset to multiple different expert scoring clients for experts to perform multi-dimensional scoring on each student composition text therein. Perform quality control and result fusion on the multi-dimensional scoring data returned by all expert scoring clients to obtain the multi-dimensional scoring label for each student composition text in the second data set.

[0042] In an embodiment of the present invention, the purpose of dividing the second data set into multiple sub-data sets is to improve the objectivity of expert scoring and improve the quality of scoring labels by cross-validating the scores of different experts. In theory, each sub-data set should be sent to as many expert scoring clients as possible to obtain as much multi-dimensional scoring data as possible, but in practical applications, it is necessary to consider time and labor costs and select an appropriate number of transmissions. For each student composition text, the scoring expert needs to mark the corresponding multiple-dimensional scoring labels 𝑆(𝐶) on the expert scoring client, and the specific scoring dimensions can be selected and optimized according to the actual composition scoring criteria. In an embodiment of the present invention, the scoring dimensions used in the multi-dimensional scoring can include 5 dimensions, namely grammar (Grammar), content (Content), fluency (Fluency), logic (Logic) and creativity (Creativity). Moreover, the expert scoring client needs to show the definitions of these 5 dimensions to the experts, or the specific points that should be considered, and at the same time, provide selection buttons for the scoring values ​​of each dimension on the GUI interface to facilitate accurate scoring by experts. For example, grammar is defined as: evaluating the correct application of language rules, including sentence structure, punctuation usage, spelling accuracy, etc.; content is defined as: measuring the degree of fit between the composition and the requirements of the topic and the literary quality of the composition content; fluency is defined as: evaluating the naturalness and coherence of language expression; logic is defined as: testing the rationality of the composition structure and the rigor of the argumentation; creativity is defined as: evaluating the novelty of the composition conception and theme, and the uniqueness of the expression.

[0043] In addition, when quality control and result fusion are performed on the multi-dimensional scoring data returned by all expert scoring clients in the present invention, the quality control method and result fusion method can be selected according to actual needs. For example, in quality control, the original collected data can be cleaned, outliers, missing values ​​and logical errors can be processed, and consistency checks can be performed, such as whether the scores of different experts on the same object conflict, and rationality checks, such as whether the scores are within a reasonable range. If the prompts on the expert scoring client are clear and the scoring input requirements are standardized, generally speaking, the quality of the scoring data obtained is better, as long as some outliers are removed. In result fusion, statistical methods, machine learning and hybrid models can be used. Statistical methods such as mean, median, weighted average, etc. are suitable for simple scenarios; cluster analysis, regression analysis and classification algorithms in machine learning can handle complex patterns. This can be achieved by using statistical methods in the scenario of the present invention.

[0044] In an embodiment of the present invention, outliers can be eliminated for all multi-dimensional scoring data of the same student composition text, and then the remaining multi-dimensional scoring data can be averaged dimension by dimension. The average scores of the five dimensions constitute the multi-dimensional average score, which can be used as the multi-dimensional scoring label corresponding to this student composition text.

[0045] Finally, multi-dimensional scoring can be performed for each student composition text in the second data set. After quality control and result fusion of the multi-dimensional scoring data returned by all expert scoring clients, the multi-dimensional scoring labels of each student composition text in the second data set are obtained. In order to facilitate subsequent calls, the processed student composition text data and multi-dimensional scoring labels can be combined into structured data and stored in a unified JSON format to ensure efficient reading and management of data during fine-tuning.

[0046] S3. Using the second data set with the multi-dimensional scoring labels as a fine-tuning data set, fine-tune the first language model to obtain a composition scoring model that can perform multi-dimensional scoring on student composition texts.

[0047] It should be noted that the fine-tuning technology of the large language model (LLM) belongs to the existing technology. The first large language model used to fine-tune the composition scoring model can be any large language model that can achieve this task, such as GPT-4, Claude3, DeepSeek-R1, Baidu Wenxin (ERNIE), Tongyi Qianwen (Qwen), etc. can all be used as basic models for fine-tuning. For Chinese compositions, Chinese optimized large models such as Baidu Wenxin, Tongyi Qianwen, DeepSeek-R1, etc. are more in line with language habits, and the depth of understanding of Chinese context is more prominent. The analysis of common implicit expressions in compositions is better than other large models trained in English context.

[0048] In an embodiment of the present invention, the specific implementation steps of the above step S2 include:

[0049] S21: Construct an essay scoring model based on a large language model. The model receives student essay text 𝐶 as input, extracts features from the student essay text data internally, and outputs the predicted scores for each scoring dimension after passing through a multi-layer deep learning network. The score of the essay in each dimension d can be recorded as , where d∈{grammar, content, fluency, logic, creativity}.

[0050] S23: The initially constructed essay scoring model needs to be trained to optimize the prediction accuracy of each scoring dimension. For each scoring dimension d, samples are collected in batches from the second data set and input into the essay scoring model. The mean square error (MSE) is calculated using the multi-dimensional scoring results output by the model and the multi-dimensional scoring labels obtained by expert scoring, and the mean square error is used as the loss function. Perform reverse parameter update on the essay scoring model:

[0051]

[0052] Among them, 𝑁 is the number of training samples, is the predicted score of the 𝑖th sample on the rating dimension d, is the true score of the 𝑖th sample on the scoring dimension d (determined by the aforementioned multi-dimensional scoring label).

[0053] The iterative fine-tuning process makes the predicted score close to the real score annotated by experts. The fine-tuned essay scoring model can receive the student essay text 𝐶 as input and then output a multi-dimensional score close to the expert scoring standard.

[0054] S4. For the target composition image uploaded by the student client, it is converted into a target composition text and then input into the composition scoring model to obtain a multi-dimensional scoring result. The target composition text and the multi-dimensional scoring result are then used to replace the placeholders in the prompt word template to form a prompt word text for driving the model to generate multi-dimensional composition improvement suggestions. Finally, the prompt word text is input into the composition feedback model constructed based on the second largest language model, and the generated multi-dimensional composition improvement suggestions are fed back to the student client.

[0055] It should be noted that the method of converting the target composition image uploaded by the student client into the target composition text is the same as in step S1. It is necessary to first perform super-resolution reconstruction on the target composition image, and then convert the reconstructed image of the target composition image into text through optical character recognition (OCR) technology. The converted text is then input into the error correction model to correct the OCR recognition errors, thereby obtaining the target composition text.

[0056] It should also be noted that the model specifically used in the second language model used to construct the composition feedback model in the present invention and the first language model can be any large language model that can achieve the task, such as GPT-4, Claude 3, DeepSeek-R1, Baidu Wenxin (ERNIE), Tongyi Qianwen (Qwen). These large language models themselves already have good task generalization capabilities, so they can be directly used for composition suggestion generation tasks. Of course, if the effect is not good, they can be further fine-tuned. It should also be noted that the above-mentioned composition scoring model and composition feedback model can use the same large language model, or two different large language models can be constructed separately, and there is no restriction on this.

[0057] It should also be noted that the specific practice of setting the prompt word template mentioned above is a common practice in the prompt word (Prompt) project. The prompt word template can be designed according to the task requirements, and a placeholder can be preset in the prompt word template for each personalized input content, so that the program can automatically replace the placeholder and generate the prompt word text input by the large language model (LLM). An exemplary prompt word template is as follows: "Please perform a task to generate improvement suggestions for student essays. The current essay text is [placeholder 1], and the current multi-dimensional score of the essay text by the expert is [placeholder 2]. Please combine the above multi-dimensional scores for the student essay text and generate essay improvement suggestions on each scoring dimension in a targeted manner."

[0058] [Placeholder 1] and [Placeholder 2] in the above prompt word template correspond to the target composition text and the multi-dimensional scoring result, respectively. The replacement of the placeholders in the prompt word template can be automated by a script program to facilitate the automation of the present invention.

[0059] Therefore, the present invention can use the composition scoring model fine-tuned based on the large language model to automatically score the composition text in multiple dimensions, and then use the large language model to combine the scoring results with the composition content through a sequence-to-sequence (Seq2Seq) method to automatically generate improvement suggestions for the composition, helping students to improve on specific composition dimensions.

[0060] It should also be noted that the method for automated composition scoring and feedback based on a large model described in S1 to S4 of the above embodiments can essentially be executed by a computer program or module. These computer programs or modules can be run on the server of the online learning platform, and the server can interact with the student client and the expert scoring client for data, thereby completing the above-mentioned method for automated composition scoring and feedback based on a large model of the present invention. Figure 4The diagram is a diagram of the interaction between the online learning platform and the student client and the expert scoring client. The student client and the expert scoring client can be in any form as long as they can provide data input, display, and communication transmission functions. For example, they can be PCs, laptops, mobile phones, tablets, and other mobile devices. The student client can upload the composition image by configuring a scanner or taking a photo with the camera on the mobile device.

[0061] Therefore, based on the same inventive concept, Figure 2 As shown, another preferred embodiment of the present invention also provides a system for automatic essay scoring and feedback based on a large model, which includes:

[0062] A training data construction module is used to obtain a first data set consisting of student composition images, perform super-resolution reconstruction on the student composition images in the first data set, and then convert each student composition from the reconstructed image to text using optical character recognition (OCR) technology, and then input the converted text into an error correction model to correct OCR recognition errors, thereby establishing a second data set consisting of student composition texts;

[0063] A scoring label generation module is used to divide the second data set into multiple sub-data sets, each of which is randomly sent to multiple different expert scoring clients for experts to perform multi-dimensional scoring on each student composition text therein, and to perform quality control and result fusion on the multi-dimensional scoring data returned by all expert scoring clients to obtain a multi-dimensional scoring label for each student composition text in the second data set;

[0064] A model fine-tuning module, used to fine-tune the first language model using the second data set with the multi-dimensional scoring labels as a fine-tuning data set, so as to obtain a composition scoring model capable of performing multi-dimensional scoring on student composition texts;

[0065] The scoring feedback module is used to convert the target composition image uploaded by the student client into a target composition text and then input it into the composition scoring model to obtain a multi-dimensional scoring result, and then replace the placeholder in the prompt word template with the target composition text and the multi-dimensional scoring result to form a prompt word text for driving the model to generate multi-dimensional composition improvement suggestions. Finally, the prompt word text is input into the composition feedback model built based on the second largest language model, and the generated multi-dimensional composition improvement suggestions are fed back to the student client.

[0066] Similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer program product corresponding to the method for automated composition scoring and feedback based on a big model provided in the above embodiment, including a computer program / instructions. When the computer program / instructions are executed by a processor, the method for automated composition scoring and feedback based on a big model in the above embodiment can be implemented.

[0067] Similarly, based on the same inventive concept, Figure 3 As shown, another preferred embodiment of the present invention also provides a computer electronic device corresponding to the method for automatic essay scoring and feedback based on a large model provided in the above embodiment, which includes a memory and a processor;

[0068] The memory is used to store computer programs;

[0069] The processor is used to implement the method of automatic composition scoring and feedback based on a large model in the above embodiment when executing the computer program.

[0070] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0071] Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer-readable storage medium corresponding to the method for automated composition scoring and feedback based on a large model provided in the above embodiment, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the method for automated composition scoring and feedback based on a large model in the above embodiment can be implemented.

[0072] It is understandable that the above storage medium may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage. The storage medium may also be a U disk, a mobile hard disk, a magnetic disk or an optical disk, etc., which can store program codes.

[0073] It is understandable that the above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0074] It should also be noted that those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the various embodiments provided in this application, the division of steps or modules in the system and method is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or steps can be combined or integrated together, and a module or step can also be split.

[0075] The above-described embodiments are only some preferred implementations of the present invention, but are not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A method for automated essay scoring and feedback based on a large model, characterized in that: include: S1, obtaining a first data set consisting of student composition images, performing super-resolution reconstruction on the student composition images in the first data set, and then converting each student composition from the reconstructed image into text by optical character recognition technology, and then inputting the converted text into an error correction model to correct recognition errors, thereby establishing a second data set consisting of student composition texts; S2, dividing the second data set into multiple sub-data sets, randomly sending each sub-data set to multiple different expert scoring clients for experts to perform multi-dimensional scoring on each student composition text therein, performing quality control and result fusion on the multi-dimensional scoring data returned by all expert scoring clients, and obtaining a multi-dimensional scoring label for each student composition text in the second data set; S3, using the second data set with the multi-dimensional scoring labels as a fine-tuning data set, fine-tuning the first language model to obtain a composition scoring model capable of performing multi-dimensional scoring on student composition texts; S4. For the target composition image uploaded by the student client, it is converted into a target composition text and then input into the composition scoring model to obtain a multi-dimensional scoring result. The target composition text and the multi-dimensional scoring result are then used to replace the placeholders in the prompt word template to form a prompt word text for driving the model to generate multi-dimensional composition improvement suggestions. Finally, the prompt word text is input into the composition feedback model constructed based on the second largest language model, and the generated multi-dimensional composition improvement suggestions are fed back to the student client.

2. The method for automated essay scoring and feedback based on a large model as claimed in claim 1, characterized in that: The error correction model adopts the BiLSTM-CRF model.

3. The method for automated essay scoring and feedback based on a large model as claimed in claim 1, characterized in that: The student composition images in the first data set need to be input into a pre-trained dual-channel CNN model to evaluate image quality, and samples with image quality scores less than a threshold are eliminated.

4. The method for automated essay scoring and feedback based on a large model as claimed in claim 1, characterized in that: When performing quality control and result fusion on the multi-dimensional scoring data returned by all expert scoring clients, it is necessary to eliminate outliers for all multi-dimensional scoring data of the same student composition text, and then average the remaining multi-dimensional scoring data dimension by dimension to obtain the multi-dimensional average score as the multi-dimensional scoring label corresponding to this student composition text.

5. The method for automated essay scoring and feedback based on a large model as claimed in claim 1, characterized in that: When fine-tuning the first largest language model, the loss function is the mean square error between the multi-dimensional scoring result output by the model and the multi-dimensional scoring label obtained by expert scoring.

6. The method for automated essay scoring and feedback based on a large model as claimed in claim 1, characterized in that: The scoring dimensions used in the multi-dimensional scoring include grammar, content, fluency, logic and creativity.

7. A system for automated essay scoring and feedback based on a large model, characterized in that: include: A training data construction module is used to obtain a first data set consisting of student composition images, perform super-resolution reconstruction on the student composition images in the first data set, and then convert each student composition from the reconstructed image to text using optical character recognition technology, and then input the converted text into an error correction model to correct recognition errors, thereby establishing a second data set consisting of student composition texts; A scoring label generation module is used to divide the second data set into multiple sub-data sets, each of which is randomly sent to multiple different expert scoring clients for experts to perform multi-dimensional scoring on each student composition text therein, and to perform quality control and result fusion on the multi-dimensional scoring data returned by all expert scoring clients to obtain a multi-dimensional scoring label for each student composition text in the second data set; A model fine-tuning module, used to fine-tune the first language model using the second data set with the multi-dimensional scoring labels as a fine-tuning data set, so as to obtain a composition scoring model capable of performing multi-dimensional scoring on student composition texts; The scoring feedback module is used to convert the target composition image uploaded by the student client into a target composition text and then input it into the composition scoring model to obtain a multi-dimensional scoring result, and then replace the placeholder in the prompt word template with the target composition text and the multi-dimensional scoring result to form a prompt word text for driving the model to generate multi-dimensional composition improvement suggestions. Finally, the prompt word text is input into the composition feedback model built based on the second largest language model, and the generated multi-dimensional composition improvement suggestions are fed back to the student client.

8. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, it can implement the method for automated essay scoring and feedback based on a large model as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the method for automated essay scoring and feedback based on a large model as described in any one of claims 1 to 7 is implemented.

10. A computer electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to implement the method for automated essay scoring and feedback based on a large model as described in any one of claims 1 to 7 when executing the computer program.

Images