File review method, device and equipment and storage medium
By constructing picture category identification and review rules for environmental protection law enforcement case files, combining the content extraction model to extract case files and calculate review scores, the problems of inefficient and insufficient accuracy of traditional reviews are solved, and more efficient and fair case file review is achieved.
Patent Information
- Application Number
- CN202411560816.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-04
- Publication Date
- 2025-02-11
AI Technical Summary
Traditional environmental law enforcement case reviews rely on manual labor, are inefficient and easily affected by subjective judgments. The existing technologies such as OCR, rule engines and machine learning models still lack accuracy and flexibility.
By obtaining business case files, splitting them into multiple case files pictures, and constructing case files categories based on business knowledge, and using case file classification model to identify picture categories. Then, the review rules for each picture category are constructed, and the case content is extracted based on the content extraction model, the review score is calculated and the review results are generated.
It improves the accuracy and efficiency of case review, reduces manual intervention, and enhances the fairness and transparency of the review results.
Smart Images
Figure CN119418357A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a case review method, device, equipment and storage medium. Background Art
[0002] With the enhancement of environmental awareness and the improvement of laws and regulations, the management and review of environmental law enforcement files have become increasingly important. Traditional environmental law enforcement file reviews mainly rely on manual work, which is not only inefficient but also easily affected by subjective judgment, resulting in inconsistent and inaccurate review results.
[0003] The review of environmental law enforcement files is usually implemented using OCR technology, rule engines or machine learning models. OCR technology can convert the text in the file images into editable text format, but it is limited by the image quality and text layout. When dealing with complex layouts and blurred images, its accuracy and efficiency still need to be improved. For rule engines, traditional rule engines can classify and judge text information according to preset rules, but when faced with new or unforeseen situations, they lack flexibility and the ability to handle complex situations. For machine learning models, some preliminary machine learning models have been applied to text classification and information extraction, but they usually require a large amount of labeled data for training and perform poorly when dealing with unstructured data. Summary of the invention
[0004] The main technical problem solved by the present invention is that case review relying on manual work is inefficient and easily affected by subjective judgment, and the accuracy of case review using OCR technology, rule engine or machine learning model is not high enough.
[0005] According to the first aspect, an embodiment provides a case review method, comprising:
[0006] Obtaining business files under the target business, and splitting the business files into multiple file images;
[0007] Constructing a plurality of case file categories according to the business knowledge of the target business, and inputting the plurality of case file categories and the plurality of case file pictures into a preset case file classification model to obtain a picture category of each of the case file pictures;
[0008] Constructing review rules corresponding to each of the picture categories, inputting the review rules and the multiple case file pictures into a preset content extraction model, and obtaining the case file content of each of the case file pictures;
[0009] The review rules are used to calculate the review score of the case file content, and the case review result of the business case file is generated according to the review score.
[0010] In some embodiments, constructing review rules corresponding to each of the image categories includes:
[0011] Constructing multiple scoring items under each of the image categories, and constructing scoring rules under different scoring items; wherein the scoring rules include screening whether the case file images contain target content, scoring according to the content screening results, and judging whether the format of the target content meets the preset format requirements, and scoring according to the format judgment results;
[0012] For each of the picture categories, the priorities among different scoring items under the picture category are defined, and the review rules corresponding to the picture category are constructed according to the priority of each scoring item and the scoring rules under each scoring item.
[0013] In some embodiments, the case file content includes content in the case file picture corresponding to the scoring item.
[0014] In some embodiments, the inputting the plurality of case file categories and the plurality of case file images into a preset case file classification model to obtain the image category of each of the case file images includes:
[0015] Inputting the multiple case file categories and the multiple case file pictures into a preset case file classification model, and outputting a first picture category of each of the case file pictures;
[0016] Using a preset text recognition algorithm to identify the text content of the case file image, and determining a second image category of each of the case file images according to the multiple case file categories and the text content;
[0017] The picture category of each of the case file pictures is determined based on the first picture category of each of the case file pictures and the second picture category of each of the case file pictures.
[0018] In some embodiments, the case classification model and the content extraction model are large language models.
[0019] In some embodiments, the constructing of multiple case file categories based on the business knowledge of the target business includes:
[0020] A corresponding case file category is constructed according to the business format and business content in the business knowledge, wherein the business format includes a table format, and the business content includes a case file source, a case file number and a case file description.
[0021] In some embodiments, the business case file is in PDF format; splitting the business case file into multiple case file images includes: splitting the business case file into multiple case file images according to a preset PDF splitting rule.
[0022] According to the second aspect, an embodiment provides a case review device, including:
[0023] A case file splitting module is used to obtain the business case files under the target business and split the business case files into multiple case file images;
[0024] A case file classification module, used to construct a plurality of case file categories according to the business knowledge of the target business, and input the plurality of case file categories and the plurality of case file pictures into a preset case file classification model to obtain a picture category of each of the case file pictures;
[0025] A content extraction module, used to construct review rules corresponding to each of the picture categories, input the review rules and the multiple case file pictures into a preset content extraction model, and obtain the case file content of each of the case file pictures;
[0026] The case review module is used to calculate the review score of the case content using the review rules, and generate the case review result of the business case according to the review score.
[0027] According to the third aspect, an embodiment provides a case review device, including:
[0028] Memory, used to store programs;
[0029] A processor is used to implement the case review method by executing the program stored in the memory.
[0030] According to the fourth aspect, an embodiment provides a computer-readable storage medium, on which a program is stored, and the program can be executed by a processor to implement a case review method.
[0031] According to the case review method, device, equipment and computer-readable storage medium of the above-mentioned embodiment, the case classification model is used to classify multiple case images split from the business case, so that the identification of the image category of the case image is more accurate. The review rules corresponding to each image category are constructed so that the review rules are more suitable for different image categories. Therefore, the case content of the case image extracted by inputting the review rules and the case content into the content extraction model is more accurate. At the same time, by calculating the review score of the case content, the case review result of the business case is generated according to the review score, thereby improving the efficiency of the case review. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 A flowchart of the case review method of an embodiment of the present application;
[0033] Figure 2A flowchart of an embodiment of inputting a plurality of case file categories and a plurality of case file pictures into a preset case file classification model to obtain a picture category of each case file picture;
[0034] Figure 3 A flowchart of constructing review rules corresponding to each picture category in an embodiment;
[0035] Figure 4 The present invention is a schematic diagram of the structure of a case review device according to an embodiment. DETAILED DESCRIPTION
[0036] The present invention is further described in detail below by specific embodiments in conjunction with the accompanying drawings. Wherein similar elements in different embodiments adopt associated similar element numbers. In the following embodiments, many detailed descriptions are for making the present application better understood. However, those skilled in the art can easily recognize that some features can be omitted in different situations, or can be replaced by other elements, materials, methods. In some cases, some operations related to the present application are not shown or described in the specification, this is to avoid the core part of the present application being overwhelmed by too much description, and for those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations according to the description in the specification and the general technical knowledge in the art.
[0037] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various implementations. At the same time, the steps or actions in the method description can also be interchanged or adjusted in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a required sequence, unless otherwise specified that a certain sequence must be followed.
[0038] The serial numbers of the components in this document, such as "first", "second", etc., are only used to distinguish the objects described and do not have any order or technical meaning. The "connection" and "coupling" mentioned in this application, unless otherwise specified, include direct and indirect connections (couplings).
[0039] Please refer to Figure 1 In an embodiment of the present invention, a case review method is provided, including steps S10 to S40, which are described in detail below.
[0040] Step S10: Obtain the business case files under the target business, and split the business case files into multiple case file images.
[0041] In some embodiments, when the target business is an environmental law enforcement business, an environmental law enforcement case file under the environmental law enforcement business is obtained as a business case file. When the business case file is in PDF format, the business case file can be split into multiple case file images according to a preset PDF splitting rule.
[0042] In some embodiments, the business case files are split according to the page numbers marked in the business case files, and the business case files with different page numbers are respectively used as separate case file images. For example, the business case file with page number 1 is used as the first case file image, and the business case file with page number 2 is used as the second case file image, and so on, until the business case files are completely split, thereby obtaining multiple case file images.
[0043] In some embodiments, since multiple case file images are separate image pages, the case file images can be directly reviewed during the case review, thereby improving the efficiency of subsequent case review.
[0044] Step S20: construct multiple case file categories based on the business knowledge of the target business, and input the multiple case file categories and multiple case file images into a preset case file classification model to obtain the image category of each case file image.
[0045] In some embodiments, multiple case file categories are constructed based on the business knowledge of the target business, including:
[0046] Construct corresponding case file categories based on business formats and business contents in business knowledge.
[0047] In some embodiments, the business format includes a table format, and the business content includes a case file source, a case file number, and a case file description. Multiple case file categories corresponding to the target business can be constructed through different business formats and business contents.
[0048] In some embodiments, under the environmental law enforcement business, for the case filing approval form, the business format is a table format, and the business content includes the source of the case, the case filing number, the party information, the case resolution and the approval signature, etc. In addition, the title also contains words such as case filing approval form or case filing approval form. For the case file category of administrative penalty decision, the business format is a non-table format, and the business content includes descriptions of the violated law, the penalty amount and the means of relief when dissatisfied with the decision.
[0049] Please refer to Figure 2 In some embodiments, step S20 inputs multiple case file categories and multiple case file pictures into a preset case file classification model to obtain the picture category of each case file picture, including steps S21 to S23, which are described in detail below.
[0050] Step S21: Input multiple case file categories and multiple case file images into a preset case file classification model, and output the first image category of each case file image.
[0051] In some embodiments, the case classification model is a large language model (LLM). The case classification constructed based on business knowledge and the large language model are combined to output the first image category of each case image. Among them, the large language model can better classify the case images, which helps the subsequent case review work to be more targeted and more efficient.
[0052] Step S22: Use a preset text recognition algorithm to identify the text content of the case file image, and determine the second image category of each case file image based on multiple case file categories and text content.
[0053] In some embodiments, the text recognition algorithm is an optical character recognition algorithm (OCR). The text recognition algorithm can directly identify the text content of the case file image, and determine the second image category of the case file image with the help of the case file category and text content constructed according to business knowledge.
[0054] Step S23: Determine the picture category of each case file picture based on the first picture category of each case file picture and the second picture category of each case file picture.
[0055] In some embodiments, the first image category and the second image category are determined by different image classification methods. When the first image category and the second image category are the same, the same category is used as the image category of the case file image. When the first image category and the second image category have some of the same categories, the same category is used as the image category of the case file image. When the first image category and the second image category are completely different, the case file image is reclassified. Therefore, the image category of each case file image is determined based on the first image category of each case file image and the second image category of each case file image, and the image category of each case file image obtained thereby is more accurate.
[0056] In some embodiments, when the first image category of the case file image is the case filing approval form, and the second image category of the case file image is also the case filing approval form, the image category of the case file image is determined to be the case filing approval form. When the first image category of the case file image is the case filing approval form, and the second image category of the case file image is the case filing approval form or the administrative penalty decision, the image category of the case file image is determined to be the case filing approval form. When the first image category of the case file image is the case filing approval form, and the second image category of the case file image is the administrative penalty decision, the case file image is reclassified.
[0057] Step S30: Construct review rules corresponding to each image category, input the review rules and multiple case images into a preset content extraction model, and obtain the case content of each case image.
[0058] Please refer to Figure 3 In some embodiments, step S30 constructs review rules corresponding to each image category, including steps S31 to S32, which are described in detail below.
[0059] Step S31: construct multiple scoring items under each image category, and construct scoring rules under different scoring items.
[0060] In some embodiments, the multiple scoring items under each image category include "name of the ecological environment authority, document name, document number, approval opinion and signature of the person in charge of the ecological environment authority, and date", "basic information of the parties, including the name or name, address, unified social credit code or ID number of the parties", "opinions and signature of the person in charge, date, approval opinion and signature of the person in charge of the agency, date", etc.
[0061] In some embodiments, the scoring rules include screening whether the case file image contains the target content, scoring according to the content screening results, and judging whether the format of the target content meets the preset format requirements, and scoring according to the format judgment results. For example, for the scoring item "name of the ecological environment competent department, document name, document number, approval opinion and signature of the person in charge of the ecological environment competent department, and date", when all the contents in the scoring item are present, the score given is 2 points.
[0062] In some embodiments, the scoring criteria also include screening whether the case file image contains content that must be included, for example, for a case file of the type of penalty decision, it must include the document number, or determine whether the content that must be included complies with the established format, for example, whether the document number in the penalty decision complies with the established format. At the same time, the scoring criteria also include determining whether there are logical errors or contradictions between different scoring items, for example, the "violated laws and regulations" and the "legal and regulatory basis for the penalty" in the penalty decision should be the same.
[0063] Step S32: For each picture category, define the priority between different scoring items under the picture category, and construct the review rules corresponding to the picture category according to the priority of each scoring item and the scoring rules under each scoring item.
[0064] In some embodiments, there are priorities between different scoring items, and the priorities of different scoring items in different picture categories are different. For example, in the picture category of case filing approval form, the priority of the scoring item "opinion and signature, date of the person in charge of the case handling, approval opinion and signature, date of the person in charge of the case handling organization" is higher than the priority of the picture category of administrative penalty decision. Therefore, in the subsequent scoring, the scoring items with higher priority will be scored, and according to the actual situation, in the subsequent scoring, the scoring of some scoring items with lower priority may be abandoned or the weight of these scoring items in the scoring may be reduced.
[0065] In some embodiments, the review rules and multiple case file images are input into a preset content extraction model to obtain the case file content of each case file image, wherein the case file content includes the content corresponding to the scoring item in the case file image. By extracting the case file content of each case file image, the scoring items to be scored can be screened in advance, thereby improving the efficiency of subsequent case file scoring.
[0066] In some embodiments, the content extraction model is a large language model. Through the deep learning ability of the large language model, it is possible to more accurately identify and extract key information from the case file image, especially for complex or ambiguous texts, the large language model has a better recognition rate than the traditional text recognition algorithm. At the same time, the case file content is automatically extracted through the large language model, reducing the workload of manual reading and data entry, and improving work efficiency.
[0067] Step S40: Calculate the review score of the case content using the review rules, and generate the case review result of the business case based on the review score.
[0068] In some embodiments, the review score of the case file content can be calculated using the review rules. Since the case file images contain different case file contents, different weights can be assigned to different case file contents. The review score of each case file image can be calculated based on the review score of the case file content and the weight of the case file content. The review score of each case file image can be summarized, and the summarized review score can be compared with the score ranges corresponding to different review results to obtain the case review result of the business case file.
[0069] In some embodiments, the review score of the case file content is calculated through the review rules, and different case file images are judged with consistent standards to reduce the difference in judgment caused by human factors. And the case file review results of the business case file are generated according to the review score to improve the fairness and transparency of the review.
[0070] Please refer to Figure 4In some embodiments, a case review device includes a case splitting module 10, a case classification module 20, a content extraction module 30 and a case review module 40, which are described in detail below.
[0071] The case file splitting module 10 is used to obtain the business case files under the target business and split the business case files into multiple case file images.
[0072] In some embodiments, when the target business is an environmental law enforcement business, an environmental law enforcement case file under the environmental law enforcement business is obtained as a business case file. When the business case file is in PDF format, the business case file can be split into multiple case file images according to a preset PDF splitting rule.
[0073] In some embodiments, the case file splitting module 10 splits the business case file according to the page numbers marked in the business case file, and uses the business case files with different page numbers as separate case file images. For example, the business case file with page number 1 is used as the first case file image, and the business case file with page number 2 is used as the second case file image, and so on, until the business case file is split completely, thereby obtaining multiple case file images.
[0074] In some embodiments, since multiple case file images are separate image pages, the case file images can be directly reviewed during the case review, thereby improving the efficiency of subsequent case review.
[0075] The case file classification module 20 is used to construct multiple case file categories according to the business knowledge of the target business, and input the multiple case file categories and multiple case file pictures into a preset case file classification model to obtain the picture category of each case file picture.
[0076] In some embodiments, the case classification module 20 constructs multiple case categories based on the business knowledge of the target business, including:
[0077] Construct corresponding case file categories based on business formats and business contents in business knowledge.
[0078] In some embodiments, the business format includes a table format, and the business content includes a case file source, a case file number, and a case file description. Multiple case file categories corresponding to the target business can be constructed through different business formats and business contents.
[0079] In some embodiments, under the environmental law enforcement business, for the case filing approval form, the business format is a table format, and the business content includes the source of the case, the case filing number, the party information, the case resolution and the approval signature, etc. In addition, the title also contains words such as case filing approval form or case filing approval form. For the case file category of administrative penalty decision, the business format is a non-table format, and the business content includes descriptions of the violated law, the penalty amount and the means of relief when dissatisfied with the decision.
[0080] Please refer back to Figure 2 In some embodiments, the case file classification module 20 inputs multiple case file categories and multiple case file pictures into a preset case file classification model to obtain the picture category of each case file picture, which is described in detail below.
[0081] In some embodiments, the case file classification module 20 inputs multiple case file categories and multiple case file images into a preset case file classification model, and outputs a first image category of each case file image. The text content of the case file image is identified using a preset text recognition algorithm, and the second image category of each case file image is determined based on the multiple case file categories and the text content. The image category of each case file image is determined based on the first image category of each case file image and the second image category of each case file image.
[0082] In some embodiments, the case classification model is a large language model (LLM). The case classification constructed based on business knowledge and the large language model are combined to output the first image category of each case image. Among them, the large language model can better classify the case images, which helps the subsequent case review work to be more targeted and more efficient.
[0083] In some embodiments, the text recognition algorithm is an optical character recognition algorithm (OCR). The text recognition algorithm can directly identify the text content of the case file image, and determine the second image category of the case file image with the help of the case file category and text content constructed according to business knowledge.
[0084] In some embodiments, the first image category and the second image category are determined by different image classification methods. When the first image category and the second image category are the same, the same category is used as the image category of the case file image. When the first image category and the second image category have some of the same categories, the same category is used as the image category of the case file image. When the first image category and the second image category are completely different, the case file image is reclassified. Therefore, the image category of each case file image is determined based on the first image category of each case file image and the second image category of each case file image, and the image category of each case file image obtained thereby is more accurate.
[0085] In some embodiments, when the first image category of the case file image is the case filing approval form, and the second image category of the case file image is also the case filing approval form, the image category of the case file image is determined to be the case filing approval form. When the first image category of the case file image is the case filing approval form, and the second image category of the case file image is the case filing approval form or the administrative penalty decision, the image category of the case file image is determined to be the case filing approval form. When the first image category of the case file image is the case filing approval form, and the second image category of the case file image is the administrative penalty decision, the case file image is reclassified.
[0086] The content extraction module 30 is used to construct review rules corresponding to each image category, input the review rules and multiple case images into a preset content extraction model, and obtain the case content of each case image.
[0087] Please refer back to Figure 3 In some embodiments, the content extraction module 30 constructs review rules corresponding to each image category, which is described in detail below.
[0088] In some embodiments, the content extraction module 30 constructs multiple scoring items under each picture category, and constructs scoring rules under different scoring items. For each picture category, the priority between different scoring items under the picture category is defined, and the review rules corresponding to the picture category are constructed according to the priority of each scoring item and the scoring rules under each scoring item.
[0089] In some embodiments, the multiple scoring items under each image category include "name of the ecological environment authority, document name, document number, approval opinion and signature of the person in charge of the ecological environment authority, and date", "basic information of the parties, including the name or name, address, unified social credit code or ID number of the parties", "opinions and signature of the person in charge, date, approval opinion and signature of the person in charge of the agency, date", etc.
[0090] In some embodiments, the scoring rules include screening whether the case file image contains the target content, scoring according to the content screening results, and judging whether the format of the target content meets the preset format requirements, and scoring according to the format judgment results. For example, for the scoring item "name of the ecological environment competent department, document name, document number, approval opinion and signature of the person in charge of the ecological environment competent department, and date", when all the contents in the scoring item are present, the score given is 2 points.
[0091] In some embodiments, the scoring criteria also include screening whether the case file image contains content that must be included, for example, for a case file of the type of penalty decision, it must include the document number, or determine whether the content that must be included complies with the established format, for example, whether the document number in the penalty decision complies with the established format. At the same time, the scoring criteria also include determining whether there are logical errors or contradictions between different scoring items, for example, the "violated laws and regulations" and the "legal and regulatory basis for the penalty" in the penalty decision should be the same.
[0092] In some embodiments, there are priorities between different scoring items, and the priorities of different scoring items in different picture categories are different. For example, in the picture category of case filing approval form, the priority of the scoring item "opinion and signature, date of the person in charge of the case handling, approval opinion and signature, date of the person in charge of the case handling organization" is higher than the priority of the picture category of administrative penalty decision. Therefore, in the subsequent scoring, the scoring items with higher priority will be scored, and according to the actual situation, in the subsequent scoring, the scoring of some scoring items with lower priority may be abandoned or the weight of these scoring items in the scoring may be reduced.
[0093] In some embodiments, the content extraction module 30 inputs the review rules and multiple case files images into a preset content extraction model to obtain the case file content of each case file image, wherein the case file content includes the content corresponding to the scoring item in the case file image. By extracting the case file content of each case file image, the scoring items to be scored can be screened in advance, thereby improving the efficiency of subsequent case file scoring.
[0094] In some embodiments, the content extraction model is a large language model. Through the deep learning ability of the large language model, it is possible to more accurately identify and extract key information from the case file image, especially for complex or ambiguous texts, the large language model has a better recognition rate than the traditional text recognition algorithm. At the same time, the case file content is automatically extracted through the large language model, reducing the workload of manual reading and data entry, and improving work efficiency.
[0095] The case review module 40 is used to calculate the review score of the case content using the review rules, and generate the case review result of the business case according to the review score.
[0096] In some embodiments, the case review module 40 can calculate the review score of the case file content using the review rules. Since the case file images contain different case file contents, different weights can be assigned to different case file contents. The review score of each case file image can be calculated based on the review score of the case file content and the weight of the case file content. The review score of each case file image can be summarized, and the summarized review score can be compared with the score ranges corresponding to different review results to obtain the case review result of the business case.
[0097] In some embodiments, the case review module 40 calculates the review score of the case content through the review rules, and judges different case images with consistent standards to reduce the judgment differences caused by human factors. And the case review results of the business case are generated according to the review score to improve the fairness and transparency of the review.
[0098] In some embodiments, a case review device includes: a memory for storing programs; and a processor for implementing a case review method by executing the programs stored in the memory.
[0099] Those skilled in the art will appreciate that all or part of the functions of the various methods in the above-mentioned embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, and the storage medium can include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to implement the above-mentioned functions. For example, the program is stored in the memory of the device, and when the program in the memory is executed by the processor, all or part of the above-mentioned functions can be implemented. In addition, when all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can also be stored in a storage medium such as a server, another computer, disk, optical disk, flash disk or mobile hard disk, and can be downloaded or copied and saved in the memory of the local device, or the system of the local device is updated, and when the program in the memory is executed by the processor, all or part of the functions in the above-mentioned embodiments can be implemented.
[0100] The above specific examples are used to illustrate the present invention, which is only used to help understand the present invention and is not intended to limit the present invention. For those skilled in the art, according to the concept of the present invention, some simple deductions, modifications or substitutions can be made.
Claims
1. A case review method, characterized in that: include: Obtaining business files under the target business, and splitting the business files into multiple file images; Constructing a plurality of case file categories according to the business knowledge of the target business, and inputting the plurality of case file categories and the plurality of case file pictures into a preset case file classification model to obtain a picture category of each of the case file pictures; Constructing review rules corresponding to each of the picture categories, inputting the review rules and the multiple case file pictures into a preset content extraction model, and obtaining the case file content of each of the case file pictures; The review rules are used to calculate the review score of the case file content, and the case review result of the business case file is generated according to the review score.
2. The case review method according to claim 1, characterized in that: The constructing of review rules corresponding to each of the image categories includes: Constructing multiple scoring items under each of the image categories, and constructing scoring rules under different scoring items; wherein the scoring rules include screening whether the case file images contain target content, scoring according to the content screening results, and judging whether the format of the target content meets the preset format requirements, and scoring according to the format judgment results; For each of the picture categories, the priorities among different scoring items under the picture category are defined, and the review rules corresponding to the picture category are constructed according to the priority of each scoring item and the scoring rules under each scoring item.
3. The case review method according to claim 2, characterized in that: The case file content includes the content corresponding to the scoring item in the case file picture.
4. The case review method according to claim 1, characterized in that: The step of inputting the plurality of case file categories and the plurality of case file pictures into a preset case file classification model to obtain a picture category of each of the case file pictures comprises: Inputting the multiple case file categories and the multiple case file pictures into a preset case file classification model, and outputting a first picture category of each of the case file pictures; Using a preset text recognition algorithm to identify the text content of the case file image, and determining a second image category of each of the case file images according to the multiple case file categories and the text content; The picture category of each of the case file pictures is determined based on the first picture category of each of the case file pictures and the second picture category of each of the case file pictures.
5. The case review method according to claim 1, characterized in that: The case classification model and the content extraction model are large language models.
6. The case review method according to claim 1, characterized in that: The multiple case file categories are constructed according to the business knowledge of the target business, including: A corresponding case file category is constructed according to the business format and business content in the business knowledge, wherein the business format includes a table format, and the business content includes a case file source, a case file number and a case file description.
7. The case review method according to claim 1, characterized in that: The business case file is in PDF format; splitting the business case file into multiple case file images includes: splitting the business case file into multiple case file images according to preset PDF splitting rules.
8. A case review device, characterized in that: include: A case file splitting module is used to obtain the business case files under the target business and split the business case files into multiple case file images; A case file classification module, used to construct a plurality of case file categories according to the business knowledge of the target business, and input the plurality of case file categories and the plurality of case file pictures into a preset case file classification model to obtain a picture category of each of the case file pictures; A content extraction module, used to construct review rules corresponding to each of the picture categories, input the review rules and the multiple case file pictures into a preset content extraction model, and obtain the case file content of each of the case file pictures; The case review module is used to calculate the review score of the case content using the review rules, and generate the case review result of the business case according to the review score.
9. A case review device, characterized in that: include: Memory, used to store programs; A processor, configured to implement the case review method as described in any one of claims 1 to 7 by executing the program stored in the memory.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, which can be executed by a processor to implement the case review method as described in any one of claims 1-7.