Form reexamination method and device, electronic equipment and storage medium
By combining RPA, large language model and RAG technology, the form review is automated and intelligent, solving the problems of inefficient and high cost of manual review, and improving the efficiency and accuracy of review.
Patent Information
- Application Number
- CN202510139811.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-08
- Publication Date
- 2025-05-09
AI Technical Summary
In the prior art, the form review process requires multiple manual reviews, which consumes a lot of labor and time costs.
RPA, large language model and RAG technology are used to automatically download and identify forms, extract feature information, retrieve relevant regulations, and complete review.
Through the automated and intelligent form review process, business efficiency is improved, labor costs are saved, and the accuracy of review results is ensured.
Smart Images
Figure CN119963137A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of form processing technology, and in particular to a form review method and device, electronic equipment, and storage medium. Background Art
[0002] Forms are one of the important carriers of information in today's society. Various forms record various daily activities and business transactions. However, while they bring convenience to business, they also bring a lot of tedious form review work to relevant business personnel. Even in some more stringent scenarios, multiple manual reviews are usually required to avoid errors.
[0003] In related technologies, the punishment of violations by relevant business personnel is a scenario that requires multiple manual reviews. Usually, it is necessary to refer to the relevant employee violation punishment regulations to make reasonable punishments for the employee's illegal behavior and conduct corresponding reviews. This process will consume a lot of manpower and time costs. Summary of the invention
[0004] The embodiments of the present application provide a form review method and device, an electronic device, and a storage medium to improve business efficiency and save labor costs.
[0005] The present application embodiment adopts the following technical solutions:
[0006] In a first aspect, an embodiment of the present application provides a form review method, wherein the method comprises:
[0007] Download and recognize the form based on RPA to obtain the text information in the form;
[0008] Extracting first feature information of the form from the text information according to a pre-selected large language model;
[0009] According to the first characteristic information of the form, obtaining second reference information corresponding to the first characteristic information of the form based on RAG retrieval;
[0010] The form is reviewed based on the second reference information.
[0011] In some embodiments, the obtaining, based on the RAG search according to the first characteristic information of the form, second reference information corresponding to the first characteristic information of the form includes:
[0012] Based on the RAG, a regulation knowledge base is established in advance by splitting multiple regulations related to form review into each complete and independent regulation as a block;
[0013] According to the first characteristic information of the form, based on the RAG, the second reference information corresponding to the first characteristic information of the form is retrieved in the regulation knowledge base,
[0014] in,
[0015] The first characteristic information includes basic attribute information of the form;
[0016] The second reference information is used to indicate which regulation the basic attribute information is related to.
[0017] In some embodiments, completing the review of the form according to the second reference information includes:
[0018] Determining whether the second reference information is consistent with the original review result of the form;
[0019] If they are inconsistent, the third reference information corresponding to the first characteristic information of the form is retrieved from the regulation knowledge base based on the RAG,
[0020] in,
[0021] The third reference information is used to characterize the degree of association between the basic attribute information and the regulations.
[0022] In some embodiments, the method further comprises:
[0023] The result returned by the large language model in the RAG is determined by multi-recall fusion and discrimination to obtain the final result.
[0024] In some embodiments, the extracting the first feature information of the form from the text information according to the pre-selected large language model includes:
[0025] According to the pre-selected large language model and prompt words, key field information is extracted from the text information, and the key field information includes at least one of the following: employee name, department, job title, description of violation facts, and penalty type.
[0026] In some embodiments, the downloading and identifying the form based on RPA to obtain text information in the form includes:
[0027] Automatically log in to the business system based on RPA, query the corresponding penalty list according to the query time period, then view the penalty details one by one, download the target form in the detail attachment, and save it to a fixed directory;
[0028] The form document is converted into the text information based on RPA.
[0029] In some embodiments, the method further comprises:
[0030] Based on RPA, the review results of the forms will be summarized and an accountability re-examination ledger will be generated.
[0031] In a second aspect, an embodiment of the present application further provides a form review device, wherein the device comprises:
[0032] Download recognition module, used to download and recognize the form based on RPA to obtain the text information in the form;
[0033] An extraction module, used for extracting first feature information of the form from the text information according to a pre-selected large language model;
[0034] A retrieval module, configured to retrieve, based on the RAG, the first characteristic information of the form and obtain second reference information corresponding to the first characteristic information of the form;
[0035] A review module is used to complete the review of the form according to the second reference information.
[0036] In a third aspect, an embodiment of the present application further provides an electronic device, comprising: a processor; and a memory arranged to store computer executable instructions, wherein the executable instructions, when executed, cause the processor to perform the above method.
[0037] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, which stores one or more programs. When the one or more programs are executed by an electronic device including multiple application programs, the electronic device executes the above method.
[0038] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: first, the form is downloaded and identified based on RPA to obtain the text information in the form. Further, the first feature information of the form is extracted from the text information based on the pre-selected large language model, and the second reference information corresponding to the first feature information of the form is retrieved based on RAG according to the first feature information of the form. Finally, the review of the form is completed based on the second reference information. Through the above method, RPA and large model technology are introduced to realize the automation and intelligence of form review, replacing manual repetitive and tedious work. Then, when using RAG technology, a unique knowledge base segmentation method is adopted for this scenario, so as to more accurately match the segments in the knowledge base. Through the above method, the accuracy of the review results is guaranteed while improving business efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0040] Figure 1 This is a flow chart of the form review method in the embodiment of the present application;
[0041] Figure 2 This is a schematic diagram of the structure of the form review device in the embodiment of the present application;
[0042] Figure 3 This is a schematic diagram of the structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION
[0043] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.
[0044] Generative artificial intelligence (AIGC) technology has gradually become the focus of widespread public attention, and large language models (abbreviated as large models) as the most critical core technology of AIGC have become one of the most important development strategies of major AI companies. Large models refer to models with large parameters trained on large-scale massive corpora. They are usually based on Transformer network structures and have emerged with powerful contextual learning capabilities and wide versatility in natural language processing. RAG is an advanced framework that combines retrieval and generation technologies. Its core idea is to improve the relevance and accuracy of text generated by AI systems by combining retrieval with large language models. RPA is an automation technology that uses software robots to simulate and execute repetitive tasks and business processes of humans on computer systems. Its goal is to reduce manual operations, improve work efficiency and reduce error rates by automating manual, repetitive and rule-based tasks.
[0045] Traditional form review usually uses manual review, which is complicated, tedious and prone to errors.
[0046] Manual form review has the problems of low efficiency, strong subjectivity of reviewers and easy errors. The form review method in the embodiment of this application applies RAG, big model and RPA technology to this scenario, and uses AI review and original manual review cross-validation to ensure the accuracy of form review, thereby freeing manpower from complex and tedious matters.
[0047] In addition, RAG technology based on large models is also more widely used, but traditional RAG technology usually divides the files to be searched into blocks with a fixed number of characters when creating a knowledge base, then extracts the embedding vector, and finally performs a similarity search in the knowledge base for the input prompt word. This block division method has certain limitations when dealing with scenarios such as form review or re-examination. RAG technology based on large models usually divides the files to be searched into blocks with a fixed number of characters, but because this segmentation method does not consider the semantic integrity and independence of each block, it may cause the target text content to be incomplete or redundant, making the results returned by the large model inaccurate. In the embodiment of the present application, when using RAG technology, a unique knowledge base segmentation method is adopted for this scenario, that is, in order to more accurately match the block (a certain penalty regulation), the knowledge base document is segmented into each complete regulation according to certain regular matching rules.
[0048] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.
[0049] The present application embodiment provides a form review method, such as Figure 1 As shown, a flow chart of a form review method in an embodiment of the present application is provided, and the method at least includes the following steps S110 to S140:
[0050] Step S110: Download and identify the form based on RPA to obtain text information in the form.
[0051] RPA is used to download the form and identify the content in the form. The image information can be converted into text information based on the content in the form.
[0052] It can be understood that RPA is a process robot automation technology that uses software robots to simulate and execute repetitive tasks and business processes of humans on computer systems. The goal of RPA is to reduce manual operations, improve work efficiency and reduce error rates by automating manual, repetitive and rule-based tasks.
[0053] Step S120: extracting first feature information of the form from the text information according to a pre-selected large language model.
[0054] The pre-selected large language model can be a mature pre-trained large model in the relevant technology, and is not specifically limited in the embodiments of the present application. Feature information in the form is extracted from the text information according to the large language model, such as the employee name, department, job title, fact description, etc. in the business form.
[0055] It can be understood that a large model refers to a language model with large-scale parameters and trained on a massive database of expectations. It usually adopts a Transformer network structure and has strong contextual learning capabilities and wide versatility.
[0056] Step S130: According to the first characteristic information of the form, obtain second reference information corresponding to the first characteristic information of the form based on RAG retrieval.
[0057] Based on the first characteristic information of the form, the retrieval module of the RAG framework is used to perform regulation retrieval to obtain the second reference information corresponding to the first characteristic information of the form. It can be understood that the second reference information is information related to the characteristic information, such as violated regulations.
[0058] It can be understood that RAG is a retrieval-enhanced generation, which is an advanced framework that combines retrieval and generation techniques. Its core idea is to improve the relevance and accuracy of text generated by AI systems by combining retrieval with large language models.
[0059] Step S140: review the form based on the second reference information.
[0060] Based on the second reference information, the form can be reviewed and evaluated, and a decision can be made as to whether manual intervention is required for re-review.
[0061] Through the above method, a technical solution combining RPA and big models was used to replace manual work to complete the form review task, changing the traditional manual-based business process, improving work efficiency, and saving labor costs.
[0062] Through the above method, a unique knowledge base segmentation method is adopted, that is, the knowledge base is segmented into each complete regulation according to certain regular matching rules. Or in order to obtain the source of the retrieval recall results more accurately, the two different types of penalty documents are not segmented, but retrieved as a whole, thereby greatly improving the retrieval recall rate. Different from the related technology, the RAG technology based on the large model usually divides the files to be retrieved into blocks with a fixed number of characters, but because this segmentation method does not consider the semantic integrity and independence of each block, it may cause the target text content to be incomplete or redundant, which makes the results returned by the large model inaccurate.
[0063] Different from related technologies, traditional form review usually adopts manual review, which is complicated, cumbersome and prone to errors. First, download and identify the form based on RPA to obtain the text information in the form. Further, the first feature information of the form is extracted from the text information based on the pre-selected large language model, and the second reference information corresponding to the first feature information of the form is obtained based on RAG retrieval based on the first feature information of the form. Finally, the review of the form is completed based on the second reference information.
[0064] Different from the related technologies, the RAG technology based on large models usually divides the files to be retrieved into blocks with a fixed number of characters. However, since this segmentation method does not consider the semantic integrity and independence of each block, it may cause the target text content to be incomplete or redundant, making the results returned by the large model inaccurate. Through the above method, a unique knowledge base segmentation method is adopted for this scenario when using RAG technology, so as to more accurately match the blocks in the knowledge base.
[0065] In one embodiment of the present application, according to the first characteristic information of the form, based on RAG retrieval, obtaining the second reference information corresponding to the first characteristic information of the form includes: based on the RAG, pre-splitting multiple regulations related to form review into each complete and independent regulation as a block, and establishing a regulations knowledge base; according to the first characteristic information of the form, based on the RAG, retrieving the second reference information corresponding to the first characteristic information of the form in the regulations knowledge base, wherein the first characteristic information includes basic attribute information of the form; the second reference information is used to indicate which regulation the basic attribute information is related to.
[0066] The review scenario of the form is explained in detail by taking the example of making reasonable punishment for the employee's illegal behavior facts and conducting corresponding review in reference to the relevant regulations on employee violation penalties.
[0067] The form obtains the type of penalty based on the violation facts, and uses RAG's technical solution. RAG technology provides enhanced retrieval to screen and return answers for large model questions and answers. Considering that the large model may have unstable answers and easily produce "AI hallucinations", the answers returned by the large model are irrelevant to the questions. When creating a knowledge base, RAG technology usually divides the files to be searched into blocks with a fixed number of characters, then extracts the embedded vectors, and finally performs a similarity search in the knowledge base for the input prompt words. Specifically in this embodiment, the employee violation penalty regulations are first registered in advance to the vector database. Unlike the previous way of building a knowledge base, this application does not segment the document knowledge according to a fixed number of characters, but manually splits the penalty regulations into complete and independent regulations as a block, thereby ensuring the accuracy of the vector database retrieval. In other words, each regulation is used as a block to establish a knowledge base.
[0068] Furthermore, the prompt word is input to call the large language model to obtain which penalty regulation the employee has violated. The prompt word example is as follows:
[0069] "Which rule of the 'Baseline Standards for Accountability for Employee Violations' does the following employee behavior violate? Based on the employee's description, find the following fields: type of violation, bottom line standard for accountability, basis for accountability. Output in json format is required. The employee's behavior description is as follows: The person in charge violated regulations during the post-loan inspection of Ma, and the specific violation facts are as follows: 1. Post-loan management is not in place. The post-loan inspection in May 2022 found that the borrower's repayment ability was insufficient, and effective collection and disposal measures were not taken in time."
[0070] If the result returned by the large model is consistent with the original audit result of the form, then it means that the review result is consistent with the original audit result, and you can enter the review ledger generation stage. If the returned result is inconsistent with the original audit result of the form, then enter the further optimization processing step.
[0071] It can be understood that the above prompt word Prompt is only an example and is not used to limit the protection scope of the embodiments of the present application.
[0072] In one embodiment of the present application, the review of the form is completed based on the second reference information, including: determining whether the second reference information is consistent with the original review result of the form; if inconsistent, retrieving the third reference information corresponding to the first characteristic information of the form in the regulations knowledge base based on the RAG, wherein the third reference information is used to characterize the degree of association between the basic attribute information and the regulations.
[0073] It can be understood that the third reference information used to characterize the degree of association between the basic attribute information and the regulations can characterize whether the basic attribute information and the regulations are in serious violation or minor violation.
[0074] Obtain the source of the regulations based on the violation facts. In order to obtain the source of the regulations based on the violation facts, the retrieval module of the RAG framework is continued to be used for regulation retrieval, and the regulation knowledge base is not divided into blocks according to a fixed number of characters, but the two types of regulations are searched as two independent blocks (need to judge whether they are minor or serious). The goal is to retrieve whether they are minor or serious, whether manual operations are performed, and whether they are not divided or handled as a whole block, which will result in less interference information; such as serious violation regulations and minor violation regulations, thereby maximizing the accuracy of retrieval recall.
[0075] It should be noted that the above-mentioned regulation knowledge base can be divided into blocks in the same way as when "calling the large language model to obtain which penalty regulation the employee has violated". In other words, the premise of dividing into two or more categories is to divide the knowledge base documents into each complete regulation.
[0076] In one embodiment of the present application, the method further includes: using a multi-recall fusion discrimination method to obtain a final result decision on the result returned by the large language model in the RAG.
[0077] Preferably, in order to ensure the accuracy of the form review, in the embodiment of the present application, the determination result returned by the large language model in the RGA technology is not used as the review result, but a multi-recall fusion discrimination method is used to decide the final result.
[0078] For example, in the case of recalling the Top 3, if two or more results appear in the knowledge base of serious violation regulations, it is judged to involve serious violations; otherwise, it is judged to be a minor violation.
[0079] Taking into account that there is generally no problem of misjudging minor violations as serious violations in the initial review of the form, the present invention mainly focuses on solving the problem of misjudging serious violations as minor violations. According to the multi-recall fusion discrimination method, if it is judged as a serious violation, the result is a manual review; otherwise, the result is directly a minor treatment. In other words, in the process of the initial review, there is less of a problem of judging minor violations as serious, but there may be a situation where serious violations are judged as minor. The key to the review is to determine whether the serious violation was judged as minor in the initial review. Therefore, a multi-recall fusion discrimination method is adopted.
[0080] Through the above method, in order to ensure the accuracy of form review.
[0081] In one embodiment of the present application, the extracting the first feature information of the form from the text information according to a pre-selected large language model includes: extracting key field information from the text information according to a pre-selected large language model and a prompt word, and the key field information includes at least one of the following: employee name, department, job title, description of violation facts, and type of punishment.
[0082] In the review of the employee violation penalty form, key field information needs to be extracted from the form as the basic information extraction result, such as employee name, department, job title, violation fact description, penalty type, etc. A large model is used to extract key information from the text information obtained in the previous step in the form of prompt word question and answer to obtain a structured parsing result.
[0083] Preferably, the pre-selected large model uses the open source Qianwen 7B model, and the prompt words used are as follows:
[0084] "Please extract the following field information from the "Opinion on Handling Illegal Behavior". The required fields are as follows: name, gender, work unit, department, title or position, specific violation facts, basis for handling opinions, violation points, and reduction of performance income. Among them, the department and title or position fields may be '-'. Please return according to the actual results. The output is required to be in json format."
[0085] In one embodiment of the present application, the RPA-based downloading and identification of the form to obtain the text information in the form includes: automatically logging into the business system based on the RPA, querying the corresponding penalty list according to the query time period, and then viewing the penalty details one by one, and downloading the target form in the details attachment, and saving it to a fixed directory; converting the form document into the text information based on the RPA.
[0086] Use RPA technology to automatically log in to the business system, query the corresponding penalty list according to the query time period, then check the penalty details one by one, and download the "Employee Violation Penalty Opinion Form" form in the details attachment and save it to a fixed directory. Then call the OCR component of RPA to convert form documents in pdf, png, jpg and other formats into text information. Preferably, text detection is implemented using the open source DBNet model, and text recognition is implemented using the classic SVTR model, and finally the input document or image is converted into text information.
[0087] In one embodiment of the present application, the method further includes: summarizing the review results of the form based on RPA and generating an accountability re-inspection ledger.
[0088] The RPA technology is used to summarize the results and generate an accountability re-inspection ledger (form). Then, after reading each item, the corresponding accountability re-inspection process is automatically queried in the business system according to the inspection form number, and the basic accountability information of the corresponding process and whether the accountability is in place are completed, and the process is automatically submitted.
[0089] The present application embodiment also provides a form review device 200, such as Figure 2 As shown, a schematic diagram of the structure of a form review device in an embodiment of the present application is provided, wherein the form review device 200 at least includes: a download identification module 210, an extraction module 220, a retrieval module 230 and a review module 240, wherein:
[0090] In one embodiment of the present application, the download identification module 210 is specifically used to: download and identify the form based on RPA to obtain text information in the form.
[0091] RPA is used to download the form and identify the content in the form. The image information can be converted into text information based on the content in the form.
[0092] It can be understood that RPA is a process robot automation technology that uses software robots to simulate and execute repetitive tasks and business processes of humans on computer systems. The goal of RPA is to reduce manual operations, improve work efficiency and reduce error rates by automating manual, repetitive and rule-based tasks.
[0093] In one embodiment of the present application, the extraction module 220 is specifically used to extract the first feature information of the form from the text information according to a pre-selected large language model.
[0094] The pre-selected large language model can be a mature pre-trained large model in the relevant technology, and is not specifically limited in the embodiments of the present application. Feature information in the form is extracted from the text information according to the large language model, such as the employee name, department, job title, fact description, etc. in the business form.
[0095] It can be understood that a large model refers to a language model with large-scale parameters and trained on a massive database of expectations. It usually adopts a Transformer network structure and has strong contextual learning capabilities and wide versatility.
[0096] In one embodiment of the present application, the retrieval module 230 is specifically used to: retrieve the second reference information corresponding to the first characteristic information of the form based on the RAG according to the first characteristic information of the form.
[0097] Based on the first characteristic information of the form, the retrieval module of the RAG framework is used to perform regulation retrieval to obtain the second reference information corresponding to the first characteristic information of the form. It can be understood that the second reference information is information related to the characteristic information, such as violated regulations.
[0098] It can be understood that RAG is a retrieval-enhanced generation, which is an advanced framework that combines retrieval and generation techniques. Its core idea is to improve the relevance and accuracy of text generated by AI systems by combining retrieval with large language models.
[0099] In one embodiment of the present application, the review module 240 is specifically used to complete the review of the form according to the second reference information.
[0100] Based on the second reference information, the form can be reviewed and evaluated, and a decision can be made as to whether manual intervention is required for re-review.
[0101] In one embodiment of the present application, the retrieval module 230 is also used to
[0102] Based on the RAG, a regulation knowledge base is established in advance by splitting multiple regulations related to form review into each complete and independent regulation as a block;
[0103] According to the first characteristic information of the form, based on the RAG, the second reference information corresponding to the first characteristic information of the form is retrieved in the regulation knowledge base,
[0104] in,
[0105] The first characteristic information includes basic attribute information of the form;
[0106] The second reference information is used to indicate which regulation the basic attribute information is related to.
[0107] In one embodiment of the present application, the review module 240 is also used to
[0108] Determining whether the second reference information is consistent with the original review result of the form;
[0109] If they are inconsistent, the third reference information corresponding to the first characteristic information of the form is retrieved from the regulation knowledge base based on the RAG,
[0110] in,
[0111] The third reference information is used to characterize the degree of association between the basic attribute information and the regulations.
[0112] In one embodiment of the present application, the review module 240 is also used to
[0113] The result returned by the large language model in the RAG is determined by multi-recall fusion and discrimination to obtain the final result.
[0114] In one embodiment of the present application, the extraction module 220 is also used to
[0115] According to the pre-selected large language model and prompt words, key field information is extracted from the text information, and the key field information includes at least one of the following: employee name, department, job title, description of violation facts, and penalty type.
[0116] In one embodiment of the present application, the download identification module 210 is also used to
[0117] Automatically log in to the business system based on RPA, query the corresponding penalty list according to the query time period, then view the penalty details one by one, download the target form in the detail attachment, and save it to a fixed directory;
[0118] The form document is converted into the text information based on RPA.
[0119] It can be understood that the above-mentioned form review device can implement each step of the form review method provided in the above-mentioned embodiment. The relevant explanations about the form review method are applicable to the form review device and will not be repeated here.
[0120] Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. Figure 3 At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. The memory may include a memory, such as a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. Of course, the electronic device may also include hardware required for other services.
[0121] The processor, network interface and memory can be interconnected through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.
[0122] The memory is used to store the program. Specifically, the program may include a program code, and the program code includes a computer operation instruction. The memory may include a memory and a non-volatile memory, and provides instructions and data to the processor.
[0123] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a form review device at the logical level. The processor executes the program stored in the memory and is specifically used to perform the following operations:
[0124] Download and recognize the form based on RPA to obtain the text information in the form;
[0125] Extracting first feature information of the form from the text information according to a pre-selected large language model;
[0126] According to the first characteristic information of the form, obtaining second reference information corresponding to the first characteristic information of the form based on RAG retrieval;
[0127] The form is reviewed based on the second reference information.
[0128] The above application Figure 1The method performed by the form review device disclosed in the illustrated embodiment can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.
[0129] The electronic device may also perform Figure 1 The method executed by the form review device in Figure 1 The functions of the illustrated embodiment will not be described in detail in the embodiments of the present application.
[0130] The present application also provides a computer-readable storage medium, which stores one or more programs, wherein the one or more programs include instructions, which, when executed by an electronic device including multiple application programs, enable the electronic device to execute Figure 1 The method performed by the form review device in the illustrated embodiment is specifically used to perform:
[0131] Download and recognize the form based on RPA to obtain the text information in the form;
[0132] Extracting first feature information of the form from the text information according to a pre-selected large language model;
[0133] According to the first characteristic information of the form, obtaining second reference information corresponding to the first characteristic information of the form based on RAG retrieval;
[0134] The form is reviewed based on the second reference information.
[0135] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0136] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0137] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0139] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0140] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.
[0141] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
[0142] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.
[0143] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.
[0144] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.
Claims
1. A form review method, wherein: The method comprises: Download and recognize the form based on RPA to obtain the text information in the form; Extracting first feature information of the form from the text information according to a pre-selected large language model; According to the first characteristic information of the form, obtaining second reference information corresponding to the first characteristic information of the form based on RAG retrieval; The form is reviewed based on the second reference information.
2. The method of claim 1, wherein: The obtaining, based on the RAG search according to the first characteristic information of the form, second reference information corresponding to the first characteristic information of the form includes: Based on the RAG, a regulation knowledge base is established in advance by splitting multiple regulations related to form review into each complete and independent regulation as a block; According to the first characteristic information of the form, based on the RAG, the second reference information corresponding to the first characteristic information of the form is retrieved in the regulation knowledge base, in, The first characteristic information includes basic attribute information of the form; The second reference information is used to indicate which regulation the basic attribute information is related to.
3. The method of claim 2, wherein: The step of completing the review of the form according to the second reference information includes: Determining whether the second reference information is consistent with the original review result of the form; If they are inconsistent, the third reference information corresponding to the first characteristic information of the form is retrieved from the regulation knowledge base based on the RAG, in, The third reference information is used to characterize the degree of association between the basic attribute information and the regulations.
4. The method according to claim 3, further comprising: The result returned by the large language model in the RAG is determined by multi-recall fusion and discrimination to obtain the final result.
5. The method of claim 1, wherein: The step of extracting the first feature information of the form from the text information according to the pre-selected large language model includes: According to the pre-selected large language model and prompt words, key field information is extracted from the text information, and the key field information includes at least one of the following: employee name, department, job title, description of violation facts, and penalty type.
6. The method of claim 1, wherein: The downloading and identifying of the form based on RPA to obtain text information in the form includes: Automatically log in to the business system based on RPA, query the corresponding penalty list according to the query time period, then view the penalty details one by one, download the target form in the detail attachment, and save it to a fixed directory; The form document is converted into the text information based on RPA.
7. The method according to any one of claims 1 to 6, further comprising: Based on RPA, the review results of the form will be summarized to generate an accountability re-inspection ledger.
8. A form review device, wherein: The device comprises: Download recognition module, used to download and recognize the form based on RPA to obtain the text information in the form; An extraction module, used for extracting first feature information of the form from the text information according to a pre-selected large language model; A retrieval module, configured to retrieve, based on the RAG, the first characteristic information of the form and obtain second reference information corresponding to the first characteristic information of the form; A review module is used to complete the review of the form according to the second reference information.
9. An electronic device, comprising: processor; as well as A memory arranged to store computer executable instructions, which when executed cause the processor to perform the method of any one of claims 1 to 7.
10. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including a plurality of application programs, causes the electronic device to execute any one of the methods of claims 1 to 7.