Marking method, device and equipment and readable storage medium

Abstract

This application discloses a grading method, apparatus, device, and readable storage medium. After determining the question to be graded, a large model is first invoked, and the question stem is input into the large model to generate the answer to the question. Then, based on the answer generated by the large model, the correct answer to the question is determined. By comparing the user's answer with the correct answer to the question, the grading result of the question can be obtained. Based on this solution, efficient and accurate grading can be achieved by utilizing the powerful language understanding and text generation capabilities of the large model without relying on question bank resources.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to a method, apparatus, device, and readable storage medium for editing. Background Technology

[0002] In recent years, with the rapid development of artificial intelligence technology, using machines to replace human labor has become a hot topic across various industries. The education sector has also evolved from traditional one-on-one or one-to-many teaching to a scenario involving interaction between teachers, machines, and students. However, when dealing with large-scale grading work, teachers are easily influenced by subjective factors such as fatigue and personal biases, affecting the accuracy and objectivity of grading. Therefore, utilizing machines to complete or assist in grading, thereby reducing the workload of manual grading and improving the accuracy and objectivity of grading, especially scoring, is of great significance to the teaching process.

[0003] Currently, the method of using machines to complete or assist in grading assignments involves intelligent grading using neural network models. This method typically involves first photographing the question text, then searching for the correct answer from a pre-set question bank, and further photographing the handwritten portion of the answer area. The neural network model then compares the similarity between the handwritten portion and the correct answer to obtain the grading result. However, this method is limited by the size and completeness of the question bank. If the question bank is too large, the grading efficiency is often low; if the question bank is insufficient, the grading accuracy is often low.

[0004] Therefore, how to provide a grading method to improve grading efficiency and accuracy has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of the above problems, this application proposes a method, apparatus, device, and readable storage medium for correction. The specific solution is as follows:

[0006] A method for grading, the method comprising:

[0007] Identify the questions to be graded, which include the question stem and the user's answers;

[0008] Call the large model and input the question stem of the question to be graded into the large model so that the large model can generate the answer result of the question to be graded;

[0009] Based on the answers to the questions to be graded generated by the large model, the correct answers to the questions to be graded are determined.

[0010] The user's answer is compared with the correct answer to the question to be graded to obtain the grading result of the question to be graded.

[0011] Optionally, determining the questions to be graded includes:

[0012] Obtain an image of the page to be graded, which contains multiple questions;

[0013] The images on the page to be graded are identified to determine the question stem and the user's answer for each question. Any question on the page to be graded is the question to be graded.

[0014] Optionally, determining the correct answer to the question to be graded based on the answers generated by the large model includes:

[0015] The question stem of the question to be graded and the answer result of the question to be graded generated by the large model are input into the answer model, and the answer model outputs the correct answer of the question to be graded.

[0016] The answering model includes a large model and a small model. The answering model is obtained by fixing the parameters of the large model, using the question stems of training questions and the answers to the training questions generated by the large model as training samples, and using the correct answers to the training questions as training labels, and adjusting the parameters of the small model.

[0017] Optionally, obtaining the image of the page to be modified includes:

[0018] Obtain a scanned image of the original page as the image of the page to be modified;

[0019] or,

[0020] Get a screenshot of the original page;

[0021] The image of the original page is preprocessed to obtain the image of the page to be modified.

[0022] Optionally, the preprocessing of the photographed image of the original page to obtain the image of the page to be modified includes:

[0023] The original page image is captured and the page area image is obtained;

[0024] The image of the page area is enhanced to obtain the image of the page to be modified.

[0025] Optionally, the step of extracting page area images from the photographed images of the original page includes:

[0026] Perform page boundary detection on the photographed image of the original page to obtain an image of the valid page area;

[0027] The effective page area image is subjected to perspective correction processing to obtain the page area image.

[0028] Optionally, the step of recognizing the image of the page to be corrected to determine the question stem and the user's answer for each question includes:

[0029] The image of the page to be corrected is converted into a first image and a second image. The first image includes printed text lines from the image of the page to be corrected, and the second image includes handwritten text lines from the image of the page to be corrected.

[0030] The first and second images are processed by text line segmentation and recognition to obtain the question stem and user answers for each question.

[0031] Optionally, the step of performing text line segmentation and recognition processing on the first image and the second image to obtain the question stem and user's answer for each question includes:

[0032] The first image is divided into text lines to obtain the printed text lines corresponding to each question.

[0033] Based on the printed text lines corresponding to each question, the second image is divided into text lines to obtain the handwritten text lines corresponding to each question.

[0034] For each question, the printed text line corresponding to the question is identified to obtain the question stem, and the handwritten text line corresponding to the question is identified to obtain the user's answer to the question.

[0035] Optionally, after obtaining the grading results for each of the questions to be graded on the page to be graded, the method further includes:

[0036] The grading results of each question on the page to be graded are displayed in the image on the page to be graded.

[0037] A correction device, the device comprising:

[0038] The question-to-be-graded question determination unit is used to determine the questions to be graded, which include the question stem and the user's answer.

[0039] The large model calling unit is used to call the large model, input the question stem of the question to be graded into the large model, so that the large model generates the answer result of the question to be graded;

[0040] The correct answer determination unit is used to determine the correct answer to the question to be graded based on the answer results generated by the large model.

[0041] The answer comparison unit is used to compare the user's answer with the correct answer to the question to be graded, and to obtain the grading result of the question to be graded.

[0042] Optionally, the unit for determining the questions to be graded includes:

[0043] The acquisition unit is used to acquire an image of the page to be graded, which contains multiple questions;

[0044] The recognition unit is used to recognize the images on the page to be corrected, determine the question stem and the user's answer for each question, and any question on the page to be corrected is the question to be corrected.

[0045] Optionally, the correct answer determination unit includes:

[0046] The question stem of the question to be graded and the answer result of the question to be graded generated by the large model are input into the answer model, and the answer model outputs the correct answer of the question to be graded.

[0047] The answering model includes a large model and a small model. The answering model is obtained by fixing the parameters of the large model, using the question stems of training questions and the answers to the training questions generated by the large model as training samples, and using the correct answers to the training questions as training labels, and adjusting the parameters of the small model.

[0048] Optionally, the acquisition unit includes:

[0049] The first acquisition unit is used to acquire a scanned image of the original page as an image of the page to be modified;

[0050] or,

[0051] The second acquisition unit is used to acquire a photograph of the original page;

[0052] The preprocessing unit is used to preprocess the photographed image of the original page to obtain the image of the page to be modified.

[0053] Optionally, the preprocessing unit includes:

[0054] The page extraction unit is used to extract the page from the photographed image of the original page to obtain a page area image;

[0055] An enhancement processing unit is used to enhance the image of the page area to obtain the image of the page to be modified.

[0056] Optionally, the page extraction unit includes:

[0057] The page boundary detection unit is used to perform page boundary detection on the photographed image of the original page to obtain an image of the effective page area.

[0058] A perspective correction processing unit is used to perform perspective correction processing on the effective page area image to obtain a page area image.

[0059] Optionally, the identification unit includes:

[0060] The conversion unit is used to convert the image of the page to be corrected into a first image and a second image, wherein the first image includes printed text lines from the image of the page to be corrected, and the second image includes handwritten text lines from the image of the page to be corrected.

[0061] The text line segmentation and recognition processing unit is used to perform text line segmentation and recognition processing on the first image and the second image to obtain the question stem and the user's answer for each question.

[0062] Optionally, the text line segmentation and recognition processing unit is specifically used for:

[0063] The first image is divided into text lines to obtain the printed text lines corresponding to each question.

[0064] Based on the printed text lines corresponding to each question, the second image is divided into text lines to obtain the handwritten text lines corresponding to each question.

[0065] For each question, the printed text line corresponding to the question is identified to obtain the question stem, and the handwritten text line corresponding to the question is identified to obtain the user's answer to the question.

[0066] Optionally, the device further includes:

[0067] The grading result display processing unit is used to display the grading results of each question to be graded on the page to be graded in the image of the page to be graded after obtaining the grading results of each question to be graded on the page to be graded.

[0068] A batching device, including a memory and a processor;

[0069] The memory is used to store programs;

[0070] The processor is used to execute the program and implement the various steps of the batching method described above.

[0071] A readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the batching method as described above.

[0072] By utilizing the above technical solution, this application discloses a grading method, apparatus, device, and readable storage medium. After determining the question to be graded, a large model is first invoked, and the question stem is input into the large model to generate the answer to the question. Then, the correct answer to the question is determined based on the answer generated by the large model. By comparing the user's answer with the correct answer to the question, the grading result of the question can be obtained. Based on this solution, efficient and accurate grading can be achieved by utilizing the powerful language understanding and text generation capabilities of the large model without relying on question bank resources. Attached Figure Description

[0073] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0074] Figure 1 This is a flowchart illustrating a correction method disclosed in an embodiment of this application;

[0075] Figure 2 This is a schematic diagram of a photograph of an original page disclosed in an embodiment of this application;

[0076] Figure 3 This is a schematic diagram of a page area image disclosed in an embodiment of this application;

[0077] Figure 4 This is a schematic diagram of an image of a page to be modified, as disclosed in an embodiment of this application.

[0078] Figure 5 This is a schematic diagram of a first image disclosed in an embodiment of this application;

[0079] Figure 6 This is a schematic diagram of a second image disclosed in an embodiment of this application;

[0080] Figure 7 This is a schematic diagram illustrating how the correction results are displayed on a page to be corrected, as disclosed in an embodiment of this application.

[0081] Figure 8 This is a schematic diagram of the structure of a correction device disclosed in an embodiment of this application;

[0082] Figure 9 This is a hardware structure block diagram of a correction device disclosed in an embodiment of this application. Detailed Implementation

[0083] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0084] The following embodiments will be used to describe the correction method provided in this application.

[0085] Reference Figure 1 , Figure 1 This is a flowchart illustrating a correction method disclosed in an embodiment of this application. The method may include:

[0086] Step S101: Determine the questions to be graded, which include the question stem and the user's answer.

[0087] In this application, the questions to be graded can be from any subject, such as mathematics or English, or various types of questions, such as fill-in-the-blank questions or problem-solving questions. Considering that grading work in the education field often involves grading test papers and practice pages, the questions to be graded can be determined from the pages to be graded (e.g., test papers, practice pages, etc.). As one possible implementation, an image of the page to be graded can be obtained. This page contains multiple questions. The image of the page to be graded is identified to determine the question stem and the user's answer for each question. Any question on the page to be graded is considered a question to be graded. It should be noted that the image of the page to be graded can be obtained by scanning the original page, or by taking a photograph of the original page and then preprocessing the photograph. The specific implementation methods for preprocessing the photograph of the original page and for identifying the image of the page to be graded to determine the question stem and the user's answer for each question will be described in detail in the following embodiments.

[0088] It should be noted that the question stem and the user's answer can be in text format.

[0089] Step S102: Call the large model and input the question stem of the question to be graded into the large model so that the large model generates the answer result of the question to be graded.

[0090] In this application, "large model" refers to a machine learning model with a large number of parameters (tens of billions or more) and a complex structure, possessing massive knowledge learning and generalization capabilities, and strong language understanding and text generation abilities. The large model described in this application can be a currently mature and widely used large model, such as the ChatGPT model, the PaLM (Pathways Language Model), the Pangu large model, and the Xinghuo cognitive large model. Alternatively, it can be a large model that has already been maturely applied, trained using questions and answers from a pre-set question bank, to improve the accuracy of its responses.

[0091] It should be noted that when inputting the question stem of the question to be graded into the large model, either text input or voice input can be used, and this application does not impose any restrictions on this.

[0092] Step S103: Based on the answer results of the questions to be graded generated by the large model, determine the correct answer of the questions to be graded.

[0093] It should be noted that the answers generated by the large model for the questions to be graded may not necessarily contain the correct answers. Even if the answers generated by the large model contain the correct answers, they may also contain some redundant information, which may affect the final grading results. Therefore, it is necessary to determine the correct answers to the questions based on the answers generated by the large model.

[0094] In this application, there are multiple ways to determine the correct answer to the question to be graded based on the answers generated by the large model. One possible implementation is to pre-train a neural network model, input the answers to the question to be graded generated by the large model into the neural network model, and then output the correct answer to the question to be graded. The training data can be the answers to training questions generated by the large model and the correct answers to those training questions. However, this training method cannot enable the neural network model to converge quickly.

[0095] As another possible implementation, a question-answering model can be constructed based on a large model. This model includes a large model and a small model, where the small model is distinguished from the large model by having fewer parameters and a simpler structure. After constructing the question-answering model, it can be trained to obtain a trained question-answering model. The training method can be to fix the parameters of the large model, using the stems of training questions and the answers generated by the large model as training samples, and the labeled correct answers to the training questions as training labels, while adjusting the parameters of the small model. This training method ensures that the performance of the large model remains unchanged, and that the small model can converge quickly.

[0096] The training questions can be from a preset question bank, and the correct answers to the training questions can be the answers to the questions saved in the preset question bank.

[0097] For ease of understanding, let's assume the training question is "Xiaoming has 10 books, lends 3 books to Xiaoli, how many books are left?" The large model generates the answer "7 books are left," and the correct answer is "7 books." Alternatively, let's assume the training question is "Xiaoming has 10 books, lends 3 books to Xiaoli, how many books are left?" The large model generates the answer "6 books are left," and the correct answer is "7 books."

[0098] Then, determining the correct answer to the question to be graded based on the answer results generated by the large model includes: inputting the question stem and the answer results generated by the large model into the answer model, and the answer model outputting the correct answer to the question to be graded.

[0099] Step S104: Compare the user's answer with the correct answer to the question to be graded to obtain the grading result of the question to be graded.

[0100] If the user's answer matches the correct answer to the question to be graded, the grading result of the question to be graded is determined to be correct; if the user's answer does not match the correct answer to the question to be graded, the grading result of the question to be graded is determined to be incorrect.

[0101] This embodiment discloses a grading method. After determining the question to be graded, a large-scale model is first invoked. The question stem is input into the large-scale model, which then generates the answer to the question. The correct answer to the question is then determined based on the answer generated by the large-scale model. By comparing the user's answer with the correct answer, the grading result of the question can be obtained. Based on this solution, efficient and accurate grading can be achieved by leveraging the powerful language understanding and text generation capabilities of the large-scale model without relying on question bank resources.

[0102] In another embodiment of this application, a specific implementation method for preprocessing the photographed image of the original page is described in detail, which may include the following steps:

[0103] Step S201: Extract the page from the photographed image of the original page to obtain the page area image;

[0104] As one possible implementation, page extraction is performed on the photographed image of the original page to obtain a page area image, including:

[0105] Perform page boundary detection on the photographed image of the original page to obtain an image of the valid page area;

[0106] The effective page area image is subjected to perspective correction processing to obtain the page area image.

[0107] In this application, a page boundary detection model can be pre-trained, and the effective page region image can be obtained by inputting the original page's photographic image into the page boundary detection model.

[0108] To better understand the difference between images of the page area and photos of the original page, please refer to... Figure 2 , Figure 2 This is a schematic diagram of a photograph of an original page disclosed in an embodiment of this application, with reference to... Figure 3 , Figure 3 This is a schematic diagram of a page area image disclosed in an embodiment of this application. Figure 2 and Figure 3 As shown, invalid page areas in the original page's photograph have been removed, and perspective correction has been applied to the valid page areas.

[0109] Step S202: Enhance the image of the page area to obtain the image of the page to be modified.

[0110] In this application, the page area image can be enhanced using an enhancement model. The enhancement process can remove lighting and shadows from the page area image, and the enhanced page area image is the image of the page to be corrected. It has higher clarity and can ensure the accuracy of subsequent recognition.

[0111] To better understand the difference between images in the page area and images on the page to be corrected, please refer to... Figure 4 , Figure 4 This is a schematic diagram of an image of a page to be modified, as disclosed in an embodiment of this application. Figure 4 As shown, its ratio Figure 3 The clarity is higher.

[0112] In another embodiment of this application, a detailed implementation method is provided for recognizing the image of the page to be corrected and determining the question stem and user answer for each question. This implementation method may include the following steps:

[0113] Step S301: Convert the image of the page to be corrected into a first image and a second image. The first image includes printed text lines from the image of the page to be corrected, and the second image includes handwritten text lines from the image of the page to be corrected.

[0114] In this application, a text line detection model supporting printed and handwritten attribute output can be trained by collecting a large amount of supervised data (annotated handwritten and printed attributes of text lines). Using this model, the image of the page to be corrected is transformed into a first image and a second image.

[0115] For ease of understanding, let's assume the image on the page to be corrected is... Figure 4 As shown, the first image obtained after the conversion is as follows: Figure 5 As shown in the second image Figure 6 As shown.

[0116] Step S302: Perform text line segmentation and recognition processing on the first image and the second image to obtain the question stem and user's answer for each question.

[0117] As one possible implementation, the step of performing text line segmentation and recognition processing on the first image and the second image to obtain the question stem and user answer for each question includes: performing text line segmentation processing on the first image to obtain printed text lines corresponding to each question; performing text line segmentation processing on the second image based on the printed text lines corresponding to each question to obtain handwritten text lines corresponding to each question; for each question, recognizing the printed text lines corresponding to the question to obtain the question stem, and recognizing the handwritten text lines corresponding to the question to obtain the user answer for the question.

[0118] In this application, the first image can be segmented into text lines using a question-segmentation model to obtain printed text lines corresponding to each question; and, based on the printed text lines corresponding to each question, the second image can be segmented into text lines to obtain handwritten text lines corresponding to each question. For each question, the printed text lines corresponding to the question are recognized to obtain the question stem, and the handwritten text lines corresponding to the question are recognized to obtain the user's answer to the question, specifically using an OCR recognition method.

[0119] As one possible implementation, after obtaining the grading results for each of the questions to be graded on the page to be graded, the method further includes:

[0120] The grading results for each question on the page to be graded are displayed on the image of the page. For example, the grading results for each question can be pasted onto the image of the page to be graded as "√" and "×" and displayed on the screen for the teacher's reference. For ease of understanding, refer to... Figure 7 , Figure 7 This is a schematic diagram illustrating how the correction results are displayed on the page to be corrected, as disclosed in an embodiment of this application.

[0121] The correction device disclosed in the embodiments of this application is described below. The correction device described below can be referred to in correspondence with the correction method described above.

[0122] Reference Figure 8 , Figure 8 This is a schematic diagram of a correction device disclosed in an embodiment of this application. Figure 8 As shown, the correction device may include:

[0123] The question-to-be-graded determination unit 11 is used to determine the questions to be graded, which include the question stem and the user's answer.

[0124] Large model calling unit 12 is used to call the large model, input the question stem of the question to be graded into the large model, so that the large model generates the answer result of the question to be graded;

[0125] The correct answer determination unit 13 is used to determine the correct answer to the question to be graded based on the answer results of the question to be graded generated by the large model.

[0126] The answer comparison unit 14 is used to compare the user's answer with the correct answer of the question to be graded, and obtain the grading result of the question to be graded.

[0127] As one possible implementation, the question-to-be-graded question determination unit includes:

[0128] The acquisition unit is used to acquire an image of the page to be graded, which contains multiple questions;

[0129] The recognition unit is used to recognize the images on the page to be corrected, determine the question stem and the user's answer for each question, and any question on the page to be corrected is the question to be corrected.

[0130] As one possible implementation, the correct answer determination unit includes:

[0131] The question stem of the question to be graded and the answer result of the question to be graded generated by the large model are input into the answer model, and the answer model outputs the correct answer of the question to be graded.

[0132] The answering model includes a large model and a small model. The answering model is obtained by fixing the parameters of the large model, using the question stems of training questions and the answers to the training questions generated by the large model as training samples, and using the correct answers to the training questions as training labels, and adjusting the parameters of the small model.

[0133] As one possible implementation, the acquisition unit includes:

[0134] The first acquisition unit is used to acquire a scanned image of the original page as an image of the page to be modified;

[0135] or,

[0136] The second acquisition unit is used to acquire a photograph of the original page;

[0137] The preprocessing unit is used to preprocess the photographed image of the original page to obtain the image of the page to be modified.

[0138] As one possible implementation, the preprocessing unit includes:

[0139] The page extraction unit is used to extract the page from the photographed image of the original page to obtain a page area image;

[0140] An enhancement processing unit is used to enhance the image of the page area to obtain the image of the page to be modified.

[0141] As one possible implementation, the page extraction unit includes:

[0142] The page boundary detection unit is used to perform page boundary detection on the photographed image of the original page to obtain an image of the effective page area.

[0143] A perspective correction processing unit is used to perform perspective correction processing on the effective page area image to obtain a page area image.

[0144] As one possible implementation, the identification unit includes:

[0145] The conversion unit is used to convert the image of the page to be corrected into a first image and a second image, wherein the first image includes printed text lines from the image of the page to be corrected, and the second image includes handwritten text lines from the image of the page to be corrected.

[0146] The text line segmentation and recognition processing unit is used to perform text line segmentation and recognition processing on the first image and the second image to obtain the question stem and the user's answer for each question.

[0147] As one possible implementation, the text line segmentation and recognition processing unit is specifically used for:

[0148] The first image is divided into text lines to obtain the printed text lines corresponding to each question.

[0149] Based on the printed text lines corresponding to each question, the second image is divided into text lines to obtain the handwritten text lines corresponding to each question.

[0150] For each question, the printed text line corresponding to the question is identified to obtain the question stem, and the handwritten text line corresponding to the question is identified to obtain the user's answer to the question.

[0151] As one possible implementation, the device further includes:

[0152] The grading result display processing unit is used to display the grading results of each question to be graded on the page to be graded in the image of the page to be graded after obtaining the grading results of each question to be graded on the page to be graded.

[0153] Reference Figure 9 , Figure 9 A hardware structure block diagram of a correction device provided in this application embodiment is shown below. Figure 9 The hardware structure of the correction device may include: at least one processor 1, at least one communication interface 2, at least one memory 3, and at least one communication bus 4;

[0154] In this embodiment of the application, the number of processor 1, communication interface 2, memory 3, and communication bus 4 is at least one, and processor 1, communication interface 2, and memory 3 communicate with each other through communication bus 4;

[0155] Processor 1 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0156] Memory 3 may include high-speed RAM, and may also include non-volatile memory, such as at least one disk storage device;

[0157] The memory stores a program, which the processor can call. The program is used for:

[0158] Identify the questions to be graded, which include the question stem and the user's answers;

[0159] Call the large model and input the question stem of the question to be graded into the large model so that the large model can generate the answer result of the question to be graded;

[0160] Based on the answers to the questions to be graded generated by the large model, the correct answers to the questions to be graded are determined.

[0161] The user's answer is compared with the correct answer to the question to be graded to obtain the grading result of the question to be graded.

[0162] Optionally, the refined and extended functions of the program can be found in the description above.

[0163] This application embodiment also provides a readable storage medium that can store a program suitable for execution by a processor, the program being used for:

[0164] Identify the questions to be graded, which include the question stem and the user's answers;

[0165] Call the large model and input the question stem of the question to be graded into the large model so that the large model can generate the answer result of the question to be graded;

[0166] Based on the answers to the questions to be graded generated by the large model, the correct answers to the questions to be graded are determined.

[0167] The user's answer is compared with the correct answer to the question to be graded to obtain the grading result of the question to be graded.

[0168] Optionally, the refined and extended functions of the program can be found in the description above.

[0169] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0170] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0171] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for grading, characterized in that, The method includes: Identify the questions to be graded, which include the question stem and the user's answers; Call the large model and input the question stem of the question to be graded into the large model so that the large model can generate the answer result of the question to be graded; Based on the answers to the questions to be graded generated by the large model, the correct answers to the questions to be graded are determined. The user's answer is compared with the correct answer to the question to be graded to obtain the grading result of the question to be graded; The determination of the correct answer to the question to be graded based on the answer results generated by the large model includes: The question stem of the question to be graded and the answer result of the question to be graded generated by the large model are input into the answer model, and the answer model outputs the correct answer of the question to be graded. The answering model includes a large model and a small model. The answering model is obtained by fixing the parameters of the large model, using the question stems of training questions and the answers to the training questions generated by the large model as training samples, and using the correct answers to the training questions as training labels, and adjusting the parameters of the small model.

2. The method according to claim 1, characterized in that, The process of determining the questions to be graded includes: Obtain an image of the page to be graded, which contains multiple questions; The images on the page to be graded are identified to determine the question stem and the user's answer for each question. Any question on the page to be graded is the question to be graded.

3. The method according to claim 2, characterized in that, The step of obtaining the image of the page to be modified includes: Obtain a scanned image of the original page as the image of the page to be modified; or, Get a screenshot of the original page; The image of the original page is preprocessed to obtain the image of the page to be modified.

4. The method according to claim 3, characterized in that, The step of preprocessing the photographed image of the original page to obtain the image of the page to be modified includes: The original page image is captured and the page area image is obtained; The image of the page area is enhanced to obtain the image of the page to be modified.

5. The method according to claim 4, characterized in that, The step of extracting page area images from the photographed images of the original page includes: Perform page boundary detection on the photographed image of the original page to obtain an image of the valid page area; The effective page area image is subjected to perspective correction processing to obtain the page area image.

6. The method according to claim 3, characterized in that, The process of recognizing the images on the page to be corrected, and determining the question stem and user's answer for each question, includes: The image of the page to be corrected is converted into a first image and a second image. The first image includes printed text lines from the image of the page to be corrected, and the second image includes handwritten text lines from the image of the page to be corrected. The first and second images are processed by text line segmentation and recognition to obtain the question stem and user answers for each question.

7. The method according to claim 6, characterized in that, The text line segmentation and recognition processing of the first and second images to obtain the question stem and user answer for each question includes: The first image is divided into text lines to obtain the printed text lines corresponding to each question. Based on the printed text lines corresponding to each question, the second image is divided into text lines to obtain the handwritten text lines corresponding to each question. For each question, the printed text line corresponding to the question is identified to obtain the question stem, and the handwritten text line corresponding to the question is identified to obtain the user's answer to the question.

8. The method according to claim 2, characterized in that, After obtaining the grading results for each of the questions to be graded on the page to be graded, the method further includes: The grading results of each question on the page to be graded are displayed in the image on the page to be graded.

9. A correction device, characterized in that, The device includes: The question-to-be-graded question determination unit is used to determine the questions to be graded, which include the question stem and the user's answer. The large model calling unit is used to call the large model, input the question stem of the question to be graded into the large model, so that the large model generates the answer result of the question to be graded; The correct answer determination unit is used to determine the correct answer to the question to be graded based on the answer results generated by the large model. The answer comparison unit is used to compare the user's answer with the correct answer of the question to be graded, and to obtain the grading result of the question to be graded. Specifically, the correct answer determination unit is used to input the question stem of the question to be graded and the answer result of the question to be graded generated by the large model into the answer model, and the answer model outputs the correct answer of the question to be graded. The answer model includes the large model and the small model. The answer model is obtained by fixing the parameters of the large model, using the question stem of the training question and the answer result of the training question generated by the large model as training samples, and using the labeled correct answer of the training question as training label, and adjusting the parameters of the small model for training.

10. A correction device, characterized in that, Including memory and processor; The memory is used to store programs; The processor is configured to execute the program to implement the various steps of the correction method as described in any one of claims 1 to 8.

11. A readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the correction method as described in any one of claims 1 to 8.

Images