CN-117237969-B - Correction method, device, equipment and readable storage medium

CN117237969BCN 117237969 BCN117237969 BCN 117237969BCN-117237969-B

Abstract

The application discloses a correction method, a correction device, correction equipment and a readable storage medium. After the questions to be corrected are determined, the large model is called, the stems of the questions to be corrected are input into the large model, so that the large model generates answer results of the questions to be corrected, correct answers of the questions to be corrected are determined based on the answer results of the questions to be corrected generated by the large model, and correction results of the questions to be corrected can be obtained by comparing answers of users with the correct answers of the questions to be corrected. Based on the scheme, the efficient and accurate correction can be realized by utilizing the powerful language understanding capability and text generating capability of the large model without depending on question bank resources.

Inventors

WU JIAJIA
ZHANG YINTIAN
CHEN MINGJUN
YIN BING

Assignees

科大讯飞股份有限公司

Dates

Publication Date: 20260505
Application Date: 20230821

Claims (11)

1. A method of modifying, the method comprising: determining a to-be-modified question, wherein the to-be-modified question comprises a question stem and an answer answered by a user; Calling a large model, and inputting the stem of the to-be-modified question into the large model so that the large model generates a response result of the to-be-modified question; Determining a correct answer of the to-be-modified questions based on the answer result of the to-be-modified questions generated by the large model; comparing the answer answered by the user with the correct answer of the to-be-modified question to obtain a modification result of the to-be-modified question; The determining the correct answer of the to-be-modified question based on the answer result of the to-be-modified question generated by the big model comprises the following steps: inputting the stem of the to-be-modified question and the answer result of the to-be-modified question generated by the large model into an answer model, and outputting a correct answer of the to-be-modified question by the answer model; The answer model comprises a big model and a small model, and is obtained by fixing parameters of the big model, taking a stem of a training question and an answer result of the training question generated by the big model as training samples, taking a correct answer of the marked training question as a training label, and adjusting and training the parameters of the small model.
2. The method of claim 1, wherein the determining the topic to be modified comprises: Acquiring a picture of a page to be modified, wherein the page to be modified comprises a plurality of titles; Identifying the picture of the page to be modified, and determining the question stem of each question and the answer which is answered by the user, wherein any question in the page to be modified is the question to be modified.
3. The method according to claim 2, wherein the obtaining a picture of the page to be modified comprises: acquiring a scanned picture of an original page as a picture of the page to be modified; Or alternatively Acquiring a photographed picture of an original page; And preprocessing the photographed picture of the original page to obtain the picture of the page to be modified.
4. The method of claim 3, wherein the preprocessing the photographed picture of the original page to obtain the picture of the page to be modified includes: carrying out page extraction on the photographed picture of the original page to obtain a page area picture; And carrying out enhancement processing on the page area picture to obtain the picture of the page to be modified.
5. The method of claim 4, wherein the performing the page extraction on the photographed image of the original page to obtain the page area image comprises: detecting page boundaries of the photographed pictures of the original pages to obtain effective page area pictures; And performing perspective correction processing on the effective page area picture to obtain a page area picture.
6. The method of claim 3, wherein identifying the picture of the page to be modified, determining the stem of each question and the answer to be answered by the user, comprises: converting the picture of the page to be modified into a first picture and a second picture, wherein the first picture comprises a printed text line in the picture of the page to be modified, and the second picture comprises a handwritten text line in the picture of the page to be modified; And performing text line division and recognition processing on the first picture and the second picture to obtain the stem of each question and the answer of the user.
7. The method of claim 6, wherein the text line dividing and identifying the first picture and the second picture to obtain the stem of each question and the answer of the user answer comprises: Performing text line division processing on the first picture to obtain a printed text line corresponding to each question; dividing the text line of the second picture based on the text line of the printing body corresponding to each question to obtain a handwriting text line corresponding to each question; And identifying the printed text line corresponding to each question to obtain a question stem of the question, and identifying the handwriting text line corresponding to the question to obtain an answer for the user of the question.
8. The method according to claim 2, wherein after obtaining the result of the modification of each of the questions to be modified in the page to be modified, the method further comprises: and displaying the correction results of the questions to be corrected in the page to be corrected in the picture of the page to be corrected.
9. An altering device, the device comprising: A to-be-modified question determination unit for determining a to-be-modified question, the questions to be modified comprise the stems and answers answered by the user; The large model calling unit is used for calling a large model, and inputting the stem of the to-be-modified question into the large model so that the large model generates a response result of the to-be-modified question; the correct answer determining unit is used for determining a correct answer of the to-be-modified question based on the answer result of the to-be-modified question generated by the large model; the answer comparison unit is used for comparing the answer answered by the user with the correct answer of the to-be-corrected question to obtain the correction result of the to-be-corrected question; The correct answer determining unit is specifically configured to input a stem of the to-be-corrected question and a answer result of the to-be-corrected question generated by the large model into an answer model, where the answer model outputs a correct answer of the to-be-corrected question, and the answer model includes the large model and a small model, and is obtained by fixing parameters of the large model, using the stem of the training question and the answer result of the training question generated by the large model as training samples, using a correct answer of the labeled training question as a training label, and adjusting and training parameters of the small model.
10. An altering device comprising a memory and a processor; the memory is used for storing programs; the processor being adapted to execute the program for carrying out the steps of the modifying method according to any one of claims 1 to 8.
11. A readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the modifying method according to any one of claims 1 to 8.

Description

Correction method, device, equipment and readable storage medium Technical Field The present application relates to the field of artificial intelligence, and more particularly, to a correction method, apparatus, device, and readable storage medium. Background In recent years, with the rapid development of artificial intelligence technology, the use of machines instead of manpower has become a hotspot direction for various industries. The education field is also gradually changed into a scene of interaction of a teacher, a machine and students by teaching the traditional teacher and the students one by one and one by many. However, when dealing with large-scale correction works, teachers are easily disturbed by subjective factors such as fatigue, personal preferences, etc., thereby affecting correction, especially the accuracy and objectivity of correction. Therefore, the machine is used for completing or assisting in completing the correction, so that the workload of manual correction is reduced, the correction, particularly the accuracy and objectivity of scoring, is improved, and the method has great significance to teaching processes. At present, the mode of completing or assisting in completing the correction by using a machine is to intelligently correct the operation by using a neural network model, the correction mode is to firstly photograph the stem text, then search a preset question bank for a correct answer corresponding to the stem text, further photograph a handwriting part of a response area of the stem text, and then compare the similarity between the handwriting part and the correct answer by using the neural network model so as to obtain a correction result. However, the method is limited by the magnitude and the perfection degree of the question bank resources, if the magnitude of the question bank resources is too large, the correction efficiency is low, and if the question bank resources are insufficient, the correction accuracy is low. Therefore, how to provide a correction method to improve correction efficiency and accuracy is a technical problem to be solved by those skilled in the art. Disclosure of Invention In view of the above, the present application proposes an altering method, apparatus, device and readable storage medium. The specific scheme is as follows: A method of modifying, the method comprising: determining a to-be-modified question, wherein the to-be-modified question comprises a question stem and an answer answered by a user; Calling a large model, and inputting the stem of the to-be-modified question into the large model so that the large model generates a response result of the to-be-modified question; Determining a correct answer of the to-be-modified questions based on the answer result of the to-be-modified questions generated by the large model; And comparing the answer answered by the user with the correct answer of the to-be-modified question to obtain the modification result of the to-be-modified question. Optionally, the determining the topic to be modified includes: Acquiring a picture of a page to be modified, wherein the page to be modified comprises a plurality of titles; Identifying the picture of the page to be modified, and determining the question stem of each question and the answer which is answered by the user, wherein any question in the page to be modified is the question to be modified. Optionally, the determining the correct answer of the to-be-modified question based on the answer result of the to-be-modified question generated by the big model includes: inputting the stem of the to-be-modified question and the answer result of the to-be-modified question generated by the large model into an answer model, and outputting a correct answer of the to-be-modified question by the answer model; The answer model comprises a big model and a small model, and is obtained by fixing parameters of the big model, taking a stem of a training question and an answer result of the training question generated by the big model as training samples, taking a correct answer of the marked training question as a training label, and adjusting and training the parameters of the small model. Optionally, the obtaining the picture of the page to be modified includes: acquiring a scanned picture of an original page as a picture of the page to be modified; Or alternatively Acquiring a photographed picture of an original page; And preprocessing the photographed picture of the original page to obtain the picture of the page to be modified. Optionally, the preprocessing the photographed image of the original page to obtain the image of the page to be modified includes: carrying out page extraction on the photographed picture of the original page to obtain a page area picture; And carrying out enhancement processing on the page area picture to obtain the picture of the page to be modified. Optionally, the performing page extraction on the photographed image of the original page