Answer content correction method, system and related device

Abstract

The invention discloses an answer content correction method and system and a related device, and the method comprises the steps: obtaining a target image corresponding to a current answer sheet, and determining at least one target partition corresponding to the target image through an image recognition large model; on the basis of the position information of the target partition in the target image, recognition prompt information matched with the target partition is generated and determined, the recognition prompt information is input into the recognition large model, and a recognition result matched with the target image is generated and obtained through the prompt information; and based on the identification result, generating and obtaining a target correction result matched with the answer content in the identification result. By means of the method, the efficiency and accuracy of correcting the answer content can be improved.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method, system and related device for correcting answer content. Background Art

[0002] In the field of education, answer correction is an important part of the teaching process. Traditional manual correction methods have many drawbacks. With the continuous expansion of the scale of education and the in-depth development of educational informatization, automatic correction technology has emerged. The current automatic correction technology mainly relies on OCR recognition technology to identify the answer sheet, and judge whether the corresponding answer content is correct based on the recognition content. Due to handwriting differences, this method is prone to errors during recognition, resulting in poor correction results.

[0003] In view of this, how to propose an efficient and accurate method for correcting answer content has become an urgent problem to be solved. Summary of the invention

[0004] The main technical problem solved by the present application is to provide a method, system and related devices for correcting answer content, which can improve the efficiency and accuracy of correcting answer content.

[0005] In order to solve the above technical problems, a technical solution adopted in the present application is: to provide a method for correcting answer content, including: obtaining a target image corresponding to the current answer sheet, and using an image recognition model to determine at least one target partition corresponding to the target image; based on the position information of the target partition in the target image, generating recognition prompt information for determining the match of the target partition, inputting the recognition prompt information into the recognition model, and using the prompt information to generate a recognition result for matching the target image; based on the recognition result, generating a target correction result that matches the answer content in the recognition result.

[0006] In order to solve the above technical problems, another technical solution adopted by the present application is: to provide a system for correcting answer content, including: an acquisition module, used to obtain the target image corresponding to the current answer sheet, and use the recognition large model to determine at least one target partition corresponding to the target image; a recognition module, used to generate recognition prompt information matching the target partition based on the position information of the target partition in the target image, and input the recognition prompt information into the recognition large model to generate a recognition result matching the target image; a processing module, used to generate a target correction result matching the answer content in the recognition result based on the recognition result. In order to solve the above technical problems, another technical solution adopted by the present application is: to provide an electronic device, including: a memory and a processor coupled to each other, the memory storing program instructions, and the processor executing the program instructions to implement the answer content correction method mentioned in the above technical solution.

[0007] In order to solve the above technical problems, another technical solution adopted in the present application is: providing a computer-readable storage medium on which program instructions are stored, and when the program instructions are executed by a processor, the method for correcting answer content mentioned in the above technical solution is implemented.

[0008] The beneficial effects of the present application are as follows: Different from the prior art, the answer content correction method proposed in the present application detects at least one corresponding target partition from the target image corresponding to the current answer sheet by using a large recognition model, and then generates corresponding recognition prompt information according to the position information of the target partition in the target image. The large recognition model is combined with the semantic information of the recognition prompt information to identify and obtain a recognition result matching the target image from the target image, thereby improving the accuracy of obtaining the recognition result. Furthermore, the answer content is automatically corrected according to the recognition result to improve the efficiency of correction. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them: Figure 1 It is a flowchart of an implementation method of the first embodiment of the method for correcting the content of the answer of the application; Figure 2 yes Figure 1 Step S101 corresponds to a flow chart of another implementation method; Figure 3 yes Figure 1 Step S102 corresponds to a flow chart of another implementation method; Figure 4 yes Figure 1 Step S102 corresponds to a flowchart of another implementation method; Figure 5 yes Figure 1 Step S103 corresponds to a flow chart of another implementation method; Figure 6 It is a structural diagram corresponding to an implementation method of the system for correcting the answer content of this application; Figure 7 It is a structural schematic diagram of an embodiment of the electronic device of the present application; Figure 8 It is a structural schematic diagram of an implementation method of a computer-readable storage medium of the present application. DETAILED DESCRIPTION

[0010] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments, and different embodiments can be adaptively combined. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0011] See also Figure 1 , Figure 1 This is a flow chart of an implementation method of a method for correcting the content of answers in this application, which method includes: S101: Obtain a target image corresponding to the current answer sheet, and use a recognition model to determine at least one target partition corresponding to the target image.

[0012] In one embodiment, the current answer sheet to be corrected is imaged to obtain a corresponding target image, and the target image is region-detected using a recognition model to obtain at least one target partition corresponding to the target image.

[0013] Specifically, the target image is detected by using the recognition model to obtain at least one corresponding target partition, and the position information corresponding to each target partition is determined. The target partition is at least one framed area determined according to the layout structure of the content in the target image.

[0014] In one implementation scenario, the above-mentioned recognition model is a large language model with better data analysis capabilities. By inputting the target image into the recognition model and prompting the recognition model to perform partition detection, the target partition output by the recognition model is obtained.

[0015] In a specific application scenario, the above-mentioned large language model may include but is not limited to deep neural networks (DNNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory networks (LSTM) and generative pre-trained Transformer models, etc. No specific restrictions are made on the specific construction and deployment of the large language model.

[0016] S102: Based on the position information of the target partition in the target image, generate recognition prompt information of the target partition matching, input the recognition prompt information into the recognition large model, and generate the recognition result of the target image matching.

[0017] In one embodiment, matching recognition prompt information is generated according to the location information corresponding to each target partition. The recognition prompt information is input into the recognition model, and the recognition prompt information is semantically analyzed using the recognition model to obtain a recognition result from the target image. The recognition result includes at least one of the question stem information and the answer content.

[0018] S103: Based on the recognition result, generate a target correction result that matches the answer content in the recognition result.

[0019] In one embodiment, after the recognition result is obtained, a reference answer matching the answer content in the recognition result is obtained, and the corresponding target correction result is obtained according to the matching reference answer and answer content.

[0020] Specifically, it is determined whether the answer content is consistent with the matched reference answer. If consistent, the generated target correction result indicates that the answer to the answer content is correct. If inconsistent, the generated target correction result indicates that the answer to the answer content is wrong.

[0021] In another embodiment, for the answer content in the recognition result, a candidate character corresponding to at least one recognized character is determined according to the recognized characters in the answer content. The corresponding recognized characters are replaced with the candidate characters to obtain reference content corresponding to the answer content. A reference answer matching the answer content is obtained, and a corresponding target correction result is obtained according to the reference answer, the answer content and the corresponding reference content.

[0022] Specifically, candidate characters are other characters similar to corresponding recognized characters. Corresponding reference content is obtained by replacing recognized characters with candidate characters, thereby reducing the impact of recognition errors caused by handwriting habits on target correction results and improving the flexibility of correction.

[0023] In a specific application scenario, when the handwritten answer content in the recognition result includes "stard", in order to avoid the influence of recognition errors caused by writing habits on the correction result, some recognized characters in "stard" are replaced to obtain the corresponding reference content "stand". The answer content "stard" and the reference content "stand" are compared with the corresponding reference answers respectively, and when either the answer content or the reference content is consistent with the reference answer, the target correction result of the corresponding question is determined to be correct. Otherwise, the target correction result of the corresponding question is determined to be wrong.

[0024] In another embodiment, the implementation process of step S103 may further include: obtaining a reference answer that matches the answer content. Based on the answer content and the reference answer, generating correction prompt information, inputting the correction prompt information into the intelligent analysis model, and generating a target correction result that matches the answer content.

[0025] Specifically, a pre-set correction prompt template is obtained, and the answer content and reference answer are input into the corresponding position in the correction prompt template to obtain correction prompt information. The correction prompt information is input into the intelligent analysis model, and the intelligent analysis model is used to perform semantic recognition on the correction prompt information and generate the target correction result. For example, the pre-built correction prompt template is "The standard reference answer is: <stdanswer>, the user's answer is: <useranswer>, give the target correction result corresponding to the answer content", and obtain the correction prompt information by inputting the answer content and reference answers into the corresponding area of ​​the correction prompt template.

[0026] In one implementation scenario, the intelligent analysis model is also a large language model with better data analysis capabilities. By inputting the correction prompt information into the intelligent analysis model, the intelligent analysis model generates the target correction result according to the semantic information corresponding to each answer content and reference answer.

[0027] In another implementation scenario, the above-mentioned intelligent analysis model is the same model as the recognition large model mentioned in the above-mentioned corresponding implementation mode, so that after acquiring the target image, the target image is input into the recognition large model, and finally the recognition large model outputs the corresponding target correction result, which greatly improves the efficiency of correction.

[0028] The answer content grading method proposed in the present application detects at least one corresponding target partition from the target image corresponding to the current answer sheet by using a large recognition model, and then generates corresponding recognition prompt information according to the position information of the target partition in the target image. The large recognition model is combined with the semantic information of the recognition prompt information to obtain a recognition result matching the target image from the target image, thereby improving the accuracy of obtaining the recognition result. Furthermore, the answer content is automatically graded according to the recognition result to improve the efficiency of the grading.

[0029] See also Figure 2 , Figure 2 yes Figure 1 Step S101 in the flowchart corresponds to another embodiment. Specifically, the implementation process of step S101 includes: S201: Capture the current answer sheet to obtain a target image.

[0030] In one implementation, after the current answer sheet is acquired, an acquisition device is used to acquire an image of the current answer sheet to obtain a corresponding target image.

[0031] In one implementation scenario, the acquisition device includes a camera. By setting the current answer sheet at a corresponding position and triggering an acquisition instruction, the camera on the acquisition device is used to acquire a target image corresponding to the current answer sheet.

[0032] In another embodiment, an initial image is acquired by capturing the current answer sheet, wherein the initial image is matched with a corresponding initial length and initial width.

[0033] Specifically, the current answer sheet is collected by using a collection device to obtain a corresponding initial image, and an initial length and an initial width corresponding to the initial image are obtained, wherein the initial length is greater than or equal to the initial width.

[0034] Furthermore, a reference correction length matching the initial length and a reference correction width matching the initial width are obtained, and the initial image is resized based on the reference correction length and the reference correction width to obtain an adjusted target image.

[0035] Specifically, a preset reference correction length and reference correction width are obtained. The initial width of the initial image is compared with the reference correction width. In response to the initial width being greater than the reference correction width, a first correction coefficient is determined, and the initial length and initial width of the initial image are corrected using the first correction coefficient, that is, the initial length and initial width of the initial image are multiplied by the first correction coefficient respectively to obtain a first corrected image, so that the width of the first corrected image is equal to the reference correction width. Further, the length of the first corrected image is compared with the reference correction width. In response to the length of the first corrected image being greater than the reference correction length, a second correction coefficient is determined, and the length and width of the first corrected image are corrected using the second correction coefficient, that is, the length and width of the first corrected image are multiplied by the second correction coefficient respectively to obtain a target image, so that the length of the target image is equal to the reference correction length. The initial image is resized by using the determined reference correction length and reference correction width, so as to facilitate unified processing of target images of different answer sheets.

[0036] It should be noted that, in the above correction process, if the initial width of the initial image is less than or equal to the reference correction width, the initial image is not preliminarily adjusted, that is, the initial image is used as the first correction image. And if the length of the first correction image is less than or equal to the reference correction length, the first correction image is directly used as the target image.

[0037] In another embodiment, in order to improve the image quality of the initial image, the initial image is preprocessed after being acquired.

[0038] Specifically, the preprocessing process includes performing morphological processing on the initial image to remove noise in the initial image. Alternatively, the preprocessing process also includes performing bilinear difference processing on the initial image to improve the smoothness of the initial image.

[0039] S202: Using the trained region detection module to obtain image coding features corresponding to the target image.

[0040] In one embodiment, a trained region detection module is obtained, and a target image corresponding to the current answer sheet is input into the region detection module, so that the target image is encoded by the region detection module to obtain image encoding features corresponding to the target image. By obtaining the image encoding features, the subsequent recognition model can be used to analyze the target image.

[0041] In one implementation scenario, the region detection module is trained using a plurality of first training samples, so that the trained region detection module has the ability to extract image features from a target image.

[0042] S203: Input the image coding features into the recognition model to generate at least one target partition corresponding to the target image; wherein the target partition is matched with corresponding position information, and the position information includes reference coordinates corresponding to the target partition.

[0043] In one embodiment, corresponding partition prompt information is generated based on image coding features, and the partition prompt information is input into the recognition model. After the partition prompt information is analyzed using the recognition model, at least one target partition matching the target image and the location information corresponding to each target partition are generated.

[0044] Specifically, a pre-built partition prompt template is obtained. After the image coding feature is obtained, the image coding feature is input into the corresponding position in the partition prompt template to generate corresponding partition prompt information, which is used to prompt the recognition model to detect at least one target partition from the target image according to the image coding feature.

[0045] In a specific application scenario, the partition prompt template is "perform partition detection on the target partition according to the <image coding feature> of the target image, and output at least one corresponding target partition and the location information of each target partition". The "<image coding feature>" in the partition prompt template is replaced with the image coding feature corresponding to the target image to obtain the partition prompt information.

[0046] In another embodiment, before using the recognition large model to determine the target partition, the recognition large model is fine-tuned using a plurality of second training samples to optimize the effect of the fine-tuned recognition large model in partition detection. Each second training sample is matched with a corresponding training label, which includes a pre-labeled box selection area and the coordinates of the four vertices of each box selection area.

[0047] The above scheme uses the recognition model to output the corresponding target partition according to the partition prompt information, so as to help obtain the recognition result in the target image by combining the target partition recognition and reduce the probability of content interference between different questions.

[0048] In another embodiment, the recognition large model includes a partitioning module, based on which step S101 includes: obtaining a trained partitioning module, inputting the target image corresponding to the current answer sheet into the partitioning module, using the partitioning module to detect at least one target partition in the target image, and determining the location information corresponding to each target partition.

[0049] Specifically, the trained partitioning module includes an image encoding submodule and an image decoding submodule coupled to each other. After the target image is input into the partitioning module, the image encoding submodule in the partitioning module processes the target image to obtain image encoding features corresponding to the target image. The image encoding features are input into the image decoding submodule, so that the layout structure of the content in the target image is analyzed by the image decoding submodule to output the target partition in the target image and the position information corresponding to each target partition.

[0050] In one implementation scenario, the partition module is trained using multiple third training samples. Each third training sample is matched with a corresponding training label, which includes a pre-marked box selection area and the coordinates of the four vertices of each box selection area. The partition module is trained using multiple third training samples so that the trained partition module has the ability to detect and obtain the corresponding target partition from the target image. Among them, the specific structure of the partition module can refer to the existing neural network model structure, which will not be elaborated in detail here.

[0051] See also Figure 3 , Figure 3 yes Figure 1 Step S102 in the flowchart corresponds to another embodiment. Specifically, the implementation process of step S102 includes: S301: Generate identification prompt information of the target partition match based on the position information corresponding to the target partition.

[0052] In one implementation, identification prompt information is generated according to reference coordinates corresponding to the target partition.

[0053] In a specific application scenario, a pre-built recognition prompt template is obtained. For each target partition, the corresponding reference coordinates are filled into the corresponding position in the recognition prompt template to obtain the corresponding recognition prompt information. For example, the pre-built recognition prompt template is "According to the <reference position> of the target partition, the content of the target partition is detected and the corresponding recognition result is output". Then when the reference coordinates corresponding to the target partition A are , , and When the above four reference coordinates are used to replace "<reference position>" in the recognition prompt template, the prompt information corresponding to the target partition A is obtained.

[0054] In another embodiment, in response to obtaining the image coding features corresponding to the target image in the above-mentioned corresponding embodiments, in order to improve the efficiency of obtaining the recognition results, the implementation process of the above-mentioned step S301 can also be: generating recognition prompt information according to the image coding features corresponding to the target image and the reference coordinates corresponding to the target partition.

[0055] In a specific application scenario, a pre-built recognition prompt template is obtained. For each target partition, the corresponding reference coordinates are filled into the corresponding position in the recognition prompt template to obtain the corresponding recognition prompt information. For example, the pre-built recognition prompt template is "according to <image coding features> and <reference position> of the target partition, the target partition is detected for content, and the corresponding recognition result is output", then the image coding features of the target image and the reference coordinates of the target partition are filled into the corresponding position in the recognition prompt template to obtain the recognition prompt information.

[0056] S302: Input the recognition prompt information into the recognition large model, and use the recognition large model to generate a structured recognition result that matches the target partition; wherein the recognition result includes recognition content that matches at least one recognition category.

[0057] In one embodiment, the recognition prompt information is input into the recognition model, so that the recognition model analyzes the recognition prompt information and obtains a matching structured recognition result from the target image. The recognition result includes recognition content that matches at least one recognition category. The recognition categories include printed categories, handwritten categories, answer area categories, image categories, and table categories.

[0058] The recognition result obtained by the recognition large model includes the recognition content matching the recognition category, so that the recognition large model distinguishes the answer content in the recognition content according to the recognition category. For example, the recognition content corresponding to the handwriting category is used as the answer content, or the recognition content corresponding to the answer area category is used as the answer content.

[0059] The above scheme inputs the recognition prompt information into the recognition model so that the recognition model combines the positional association relationship between different target partitions and the semantic information of the content in each target partition to obtain the recognition results that match each target partition respectively, thereby reducing the probability of identifying the content in other target partitions as the recognition results corresponding to the current target partition and improving the recognition accuracy.

[0060] See also Figure 4 , Figure 4 yes Figure 1 Step S102 in the flowchart corresponds to another implementation method. Specifically, the implementation process of step S102 may also include: S401: Acquire a target recognition pattern that matches a target image.

[0061] In one embodiment, a target recognition pattern matching the target image is predetermined, and the target recognition pattern is used to represent an output pattern of the recognition result.

[0062] Specifically, the target recognition mode includes a small question output mode and a large question output mode. When the target recognition mode is a small question output mode, the corresponding recognition result is outputted for each small question in the subsequent content recognition of the target image. Alternatively, when the target recognition mode is a large question output mode, the corresponding recognition result is outputted for each large question in the subsequent content recognition of the target image.

[0063] In one implementation scenario, when the target recognition mode is the big question output mode, when outputting the recognition result corresponding to the current big question, the association information between the small questions in the current big question is combined to improve the accuracy of the recognition result. When the target recognition mode is the small question output mode, when outputting the recognition result corresponding to the current big question, there is no need to combine the association information between different small questions, thereby improving the efficiency of obtaining the recognition result. The above association information includes semantic association information and position association information.

[0064] Optionally, in other implementations, the target recognition mode also includes a row output mode, that is, when the current target recognition mode is the row output mode, when the target image is output for content recognition, the corresponding recognition result is output in units of each row.

[0065] S402: Generate recognition prompt information based on the target recognition mode and the position information corresponding to the target partition.

[0066] In one embodiment, recognition prompt information is generated according to the target recognition mode, the reference coordinates corresponding to each target partition and the target image. The recognition prompt information is used to prompt the recognition model to recognize the recognition result corresponding to the target image according to the target recognition mode and the reference coordinates corresponding to each target partition obtained by detection.

[0067] S403: Input the recognition prompt information into the recognition large model to generate a structured recognition result matching the target image; wherein the recognition result includes recognition content matching at least one recognition category.

[0068] In one embodiment, the recognition prompt information is input into the recognition model, so that the recognition model analyzes the recognition prompt information and obtains a matching structured recognition result from the target image. The recognition result includes recognition content that matches at least one recognition category. Among them, the recognition category includes a printed category, a handwritten category, an answer area category, an image category, and a table category, etc. The specific implementation process can refer to the corresponding implementation method above.

[0069] The above scheme helps to improve the flexibility of obtaining recognition results by obtaining the target recognition mode and generating recognition prompt information in combination with the target recognition mode.

[0070] See also Figure 5 , Figure 5 yes Figure 1 Step S103 in the flowchart corresponds to another embodiment. Specifically, the implementation process of step S103 includes: S501: Obtain a reference answer that matches the answer content, and based on the reference answer, obtain an initial correction result that matches the answer content.

[0071] In one embodiment, after obtaining the recognition result, when the recognition result includes the answer content, a reference answer matching the answer content is obtained. According to the matching reference answer and answer content, a corresponding initial correction result is obtained.

[0072] Specifically, when the answer content is consistent with the corresponding reference answer, the corresponding initial correction result is generated as "correct". Otherwise, the corresponding initial correction result is generated as "wrong" and the corresponding score is deducted.

[0073] In another embodiment, when the recognition result includes the answer content, a candidate character corresponding to at least one recognized character is determined based on the recognized characters in the answer content. The corresponding recognized characters are replaced with the candidate characters to obtain reference content corresponding to the answer content. A reference answer matching the answer content is obtained, and a corresponding initial correction result is obtained based on the reference answer, the answer content, and the corresponding reference content. The specific implementation process can refer to the corresponding implementation method described above, and will not be elaborated in detail here.

[0074] S502: Based on the initial correction result, obtain the diversion judgment result.

[0075] In one embodiment, when the initial correction result is "correct", a diversion judgment result is generated to determine whether the initial correction result meets the preset condition. Alternatively, when the initial correction result is "wrong", a diversion judgment result is generated to determine whether the initial correction result does not meet the preset condition.

[0076] S503: In response to the diversion judgment result determining that the initial correction result meets a preset condition, the initial correction result is used as a target correction result.

[0077] In one implementation, when it is determined that the initial correction result meets the preset condition, the initial correction result is directly used as the target correction result.

[0078] S504: In response to the diversion judgment result determining that the initial correction result does not meet the preset conditions, based on the answer content and the reference answer, generate correction prompt information, input the correction prompt information into the intelligent analysis model, and generate a target correction result that matches the answer content.

[0079] In one embodiment, when it is determined that the initial correction result does not meet the preset conditions, the answer content and the reference answer are input into the intelligent analysis model, so that the intelligent analysis model can be used to perform semantic analysis on the answer content and the reference answer respectively, and obtain the target correction result that matches the answer content based on the semantics.

[0080] In one implementation scenario, a preset correction prompt template is obtained, and the answer content and reference answers are input into the corresponding positions in the correction prompt template to obtain correction prompt information. The correction prompt information is input into the intelligent analysis model, and the intelligent analysis model is used to perform semantic recognition on the correction prompt information and generate the target correction result.

[0081] In a specific application scenario, the pre-built task template is "The question information is: <content>, the standard reference answer is: <stdanswer>, the user's answer is: <useranswer>, give the target correction result corresponding to the answer content", by inputting the answer content, the question stem information corresponding to the answer content and the reference answer into the corresponding area of ​​the task template, the task text is obtained.

[0082] In another embodiment, in order to improve the accuracy of grading, the process of using the intelligent analysis model in the above corresponding embodiment to generate the target correction result can be: using the intelligent analysis model to obtain the matching target correction result according to the answer content, the question stem information matching the answer content, the reference answer matching the answer content, and the problem-solving process corresponding to the reference answer.

[0083] Specifically, the intelligent analysis model is prompted to generate a problem-solving process that matches the above-mentioned reference answer, and generates a target grading result by combining the problem-solving process, the answer content, the question stem information corresponding to the answer content, and the reference answer corresponding to the answer content.

[0084] In a specific application scenario, the pre-built task template is "Complete the following questions and give the solution process. The question stem information is: <content>, the standard reference answer is: <stdanswer>, the user's answer is: <useranswer>, combined with the problem-solving process, give the target correction result corresponding to the answer content", by inputting the answer content, the stem information matching the answer content and the reference answer into the corresponding area of ​​the task template, the task text is obtained.

[0085] In the above scheme, for the initial grading results that do not meet the preset conditions, the intelligent analysis model is called to obtain the target grading results based on at least the answer content and the semantics of the reference answer, thereby reducing the probability of grading errors.

[0086] In another embodiment, the implementation process of the above step S502 may further include: obtaining feature information matching the answer content, wherein the feature information includes at least one of the stem information matching the answer content, the number of characters corresponding to the reference answer, and the language category.

[0087] Specifically, the question stem information and the reference answer matching the answer content are obtained, the number of characters corresponding to the reference answer and the language category corresponding to the reference answer are determined, and at least one of the answer content, the matching question stem information, the matching reference answer, the number of characters corresponding to the reference answer and the language category corresponding to the reference answer is used as feature information. The above-mentioned language categories include Chinese, English and Japanese.

[0088] Optionally, when the language category corresponding to the reference answer is English, the feature information may also include character case information corresponding to the reference answer.

[0089] Furthermore, the initial correction result and the feature information are input into the trained diversion judgment model to obtain the diversion judgment result output by the diversion judgment model.

[0090] Specifically, the initial correction result and the above-mentioned characteristic information are input into the trained diversion judgment model, so that the initial correction result and the characteristic information are analyzed by the diversion judgment model, and the diversion judgment result is output.

[0091] In one implementation scenario, the diversion judgment model is a pre-trained language model, which is trained using multiple second training samples. The specific structure of the diversion judgment model can refer to the existing BERT model structure.

[0092] The above scheme automatically judges the credibility of the initial grading results by training the diversion judgment model to decide whether it is necessary to call the intelligent analysis model for further correction, thereby improving the accuracy of judging the answer content.

[0093] See also Figure 6 , Figure 6 1 is a schematic diagram of a structure of a system for correcting the answer content of the present application corresponding to an embodiment. Specifically, the system for correcting the answer content includes an acquisition module 10, an identification module 20 and a processing module 30 coupled to each other.

[0094] Specifically, the acquisition module 10 is used to acquire the target image corresponding to the current answer sheet, and use the recognition model to determine at least one target partition corresponding to the target image.

[0095] The recognition module 20 is used to generate recognition prompt information of the target partition matching based on the position information of the target partition in the target image, input the recognition prompt information into the recognition large model, and generate a recognition result of the target image matching.

[0096] The processing module 30 is used to generate a target correction result that matches the answer content in the recognition result based on the recognition result.

[0097] In one embodiment, the acquisition module 10 acquires the target image corresponding to the current answer sheet, and uses the recognition large model to determine at least one target partition corresponding to the target image, including: collecting the current answer sheet to obtain the target image; using the trained area detection module to acquire the image coding features corresponding to the target image; inputting the image coding features into the recognition large model to generate at least one target partition corresponding to the target image; wherein the target partition is matched with corresponding position information, and the position information includes the reference coordinates corresponding to the target partition.

[0098] In one embodiment, please continue to refer to Figure 6 The answer content correction system proposed in this application also includes an image adjustment module 40 coupled to the acquisition module 10, and the image adjustment module 40 is used to: obtain an initial image obtained by capturing the current answer sheet; wherein the initial image is matched with a corresponding initial length and initial width; obtain a reference correction length matching the initial length, and a reference correction width matching the initial width, and based on the reference correction length and the reference correction width, resize the initial image to obtain an adjusted target image.

[0099] In one embodiment, the recognition module 20 generates recognition prompt information for target partition matching based on the position information of the target partition in the target image, inputs the recognition prompt information into the recognition large model, and generates a recognition result for target image matching, including: obtaining a target recognition pattern that matches the target image; generating recognition prompt information based on the target recognition pattern and the position information corresponding to the target partition; inputting the recognition prompt information into the recognition large model, and generating a structured recognition result for target image matching; wherein the recognition result includes recognition content that matches at least one recognition category.

[0100] In one embodiment, the processing module 30 generates a target correction result that matches the answer content in the recognition result based on the recognition result, including: obtaining a reference answer that matches the answer content; generating correction prompt information based on the answer content and the reference answer, and inputting the correction prompt information into the intelligent analysis model to generate a target correction result that matches the answer content.

[0101] In one embodiment, the processing module 30 generates a target correction result that matches the answer content in the recognition result based on the recognition result, including: obtaining a reference answer that matches the answer content, and obtaining an initial correction result that matches the answer content based on the reference answer; obtaining a diversion judgment result based on the initial correction result; in response to the diversion judgment result determining that the initial correction result meets the preset conditions, using the initial correction result as the target correction result; in response to the diversion judgment result determining that the initial correction result does not meet the preset conditions, generating correction prompt information based on the answer content and the reference answer, inputting the correction prompt information into the intelligent analysis model, and generating a target correction result that matches the answer content.

[0102] In one embodiment, please continue to refer to Figure 6 The answer content grading system proposed in this application also includes a diversion module 50 coupled to the processing module 30, and the diversion module 50 is used to obtain feature information matching the answer content; wherein the feature information includes at least one of the question stem information matching the answer content, the number of characters corresponding to the reference answer, and the language category; the initial grading result and the feature information are input into the trained diversion judgment model to obtain the diversion judgment result output by the diversion judgment model.

[0103] See also Figure 7 , Figure 7 : is a structural diagram of an embodiment of an electronic device of the present application. The electronic device includes: a memory 60 and a processor 70 coupled to each other. The memory 60 stores program instructions, and the processor 70 is used to execute the program instructions to implement the method mentioned in any of the above embodiments. Specifically, the electronic device includes but is not limited to: a desktop computer, a laptop computer, a tablet computer, a server, etc., which are not limited here. In addition, the processor 70 can also be called a CPU (Center Processing Unit). The processor 70 may be an integrated circuit chip with signal processing capabilities. The processor 70 can also be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field-programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. In addition, the processor 70 can be implemented by an integrated circuit chip.

[0104] See also Figure 8 , Figure 8 It is a structural diagram of an embodiment of a computer-readable storage medium of the present application. The computer-readable storage medium 80 stores program instructions 90 that can be executed by a processor. When the program instructions 90 are executed by the processor, the method mentioned in any of the above embodiments is implemented.

[0105] In the several embodiments provided in the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0106] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0107] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0108] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0109] The above description is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.< / useranswer> < / stdanswer> < / content> < / useranswer> < / stdanswer> < / content> < / useranswer> < / stdanswer>

Claims

1. A method for correcting answer content, characterized in that: include: Obtaining a target image corresponding to the current answer sheet, and determining at least one target partition corresponding to the target image using a recognition model; Based on the position information of the target partition in the target image, generate recognition prompt information of the target partition matching, input the recognition prompt information into the recognition large model, and generate the recognition result of the target image matching; Based on the recognition result, a target correction result matching the answer content in the recognition result is generated.

2. The method according to claim 1, characterized in that The step of obtaining a target image corresponding to the current answer sheet and determining at least one target partition corresponding to the target image by using a recognition model includes: Capturing the current answer sheet to obtain the target image; Using the trained region detection module to obtain image coding features corresponding to the target image; The image coding features are input into the recognition model to generate at least one target partition corresponding to the target image; wherein the target partition is matched with the corresponding position information, and the position information includes the reference coordinates corresponding to the target partition.

3. The method according to claim 2, characterized in that The collecting the current answer sheet to obtain the target image includes: Acquire an initial image obtained by capturing the current answer sheet; wherein the initial image is matched with a corresponding initial length and initial width; A reference correction length matching the initial length and a reference correction width matching the initial width are obtained, and the initial image is resized based on the reference correction length and the reference correction width to obtain the adjusted target image.

4. The method according to claim 1, characterized in that: The step of generating recognition prompt information of the target partition matching based on the position information of the target partition in the target image, inputting the recognition prompt information into the recognition large model, and generating the recognition result of the target image matching includes: Acquire a target recognition pattern that matches the target image; Generate the recognition prompt information based on the target recognition mode and the position information corresponding to the target partition; The recognition prompt information is input into the recognition large model to generate a structured recognition result matching the target image; wherein the recognition result includes recognition content matching at least one recognition category.

5. The method according to claim 1, characterized in that The step of generating a target correction result matching the answer content in the recognition result based on the recognition result includes: Obtaining a reference answer that matches the question content; Based on the answer content and the reference answer, correction prompt information is generated, and the correction prompt information is input into the intelligent analysis model to generate the target correction result matching the answer content.

6. The method according to claim 1, characterized in that The step of generating a target correction result matching the answer content in the recognition result based on the recognition result includes: Obtaining a reference answer that matches the answer content, and obtaining an initial correction result that matches the answer content based on the reference answer; Based on the initial correction result, obtaining a diversion judgment result; In response to the diversion judgment result determining that the initial correction result meets a preset condition, taking the initial correction result as the target correction result; In response to the diversion judgment result determining that the initial correction result does not meet the preset conditions, correction prompt information is generated based on the answer content and the reference answer, and the correction prompt information is input into the intelligent analysis model to generate the target correction result matching the answer content.

7. The method according to claim 6, characterized in that The obtaining of the diversion judgment result based on the initial correction result includes: Acquire feature information matching the answer content; wherein the feature information includes at least one of the question stem information matching the answer content, the number of characters corresponding to the reference answer, and the language category; The initial correction result and the feature information are input into the trained diversion judgment model to obtain the diversion judgment result output by the diversion judgment model.

8. A system for correcting answer content, characterized in that: include: An acquisition module, used for acquiring a target image corresponding to the current answer sheet, and determining at least one target partition corresponding to the target image by using a recognition large model; A recognition module, used for generating recognition prompt information of the target partition matching based on the position information of the target partition in the target image, inputting the recognition prompt information into the recognition large model, and generating a recognition result of the target image matching; A processing module is used to generate a target correction result that matches the answer content in the recognition result based on the recognition result.

9. An electronic device, characterized in that: include: A memory and a processor are coupled to each other, wherein the memory stores program instructions, and the processor is used to execute the program instructions to implement the method for correcting answer content as described in any one of claims 1-7.

10. A computer-readable storage medium having program instructions stored thereon, characterized in that: When the program instructions are executed by the processor, the method for correcting the answer content as described in any one of claims 1-7 is implemented.

Images