Business document processing method and device, equipment, medium and program product

Abstract

The invention provides a business document processing method which can be applied to the technical field of artificial intelligence and financial science and technology, and particularly relates to application of a large model in the field of financial science and technology. The business processing method comprises the steps that optical character recognition is carried out on an obtained target business document to obtain target business data, and the target data comprises a business text and corresponding structure position information; content identification is carried out based on the business text, a scene identifier corresponding to the target business document is determined, and the scene identifier is used for representing the content architecture of the target business document; retrieving in a pre-configured rule library based on the scene identifier to obtain a plurality of first cue words, the plurality of first cue words being used for representing an analysis rule of the target business document; and guiding the large model to extract the business text based on the analysis rule and the structure position information by using the plurality of first cue words to obtain target business information. The invention further provides a business document processing device and equipment, a medium and a program product.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and financial technology, in particular to the application of large models in the field of financial technology, and more particularly to a business document processing method and device, equipment, medium and program product. BACKGROUND

[0002] With the development of technology, business document processing has shown a trend of dataization and standardization, for example, the processing of enterprise annuity plan documents. However, in actual business operations, the text in the business document is directly extracted by optical character recognition, and the form is filled by regular matching keywords, but for the cross-row complex structure such as "right and interest attribution ratio table" in the enterprise annuity plan document, the analysis accuracy needs to be improved.

[0003] In addition, due to the differences in the content architecture of business documents in different cities, it is necessary to redevelop analysis rules for different regions when analyzing business documents, which has poor adaptability and needs to be improved in terms of analysis efficiency. SUMMARY

[0004] In view of the above problems, the present application provides a business document processing method and device, equipment, medium and program product for improving the processing efficiency of business documents.

[0005] According to a first aspect of the present application, a business document processing method is provided, which comprises: performing optical character recognition on an obtained target business document to obtain target business data, the target data comprising business text and corresponding structural position information; performing content recognition based on the business text to determine a scene identifier corresponding to the target business document, the scene identifier being used to represent the content architecture of the target business document; retrieving a plurality of first prompt words in a preconfigured rule library based on the scene identifier, the plurality of first prompt words being used to represent the analysis rules of the target business document; and using the plurality of first prompt words to guide a large model to extract the business text based on the analysis rules and the structural position information to obtain target business information.

[0006] According to an embodiment of the present application, the content recognition based on the business text to determine the scene identifier corresponding to the target business document comprises: extracting a plurality of module titles in the business text, and sequentially matching the plurality of module titles with a pre-set content general framework using a regular expression to obtain a matching feature vector, the content general framework comprising a general content architecture and corresponding standard module titles; and obtaining the scene identifier based on the cumulative value of the matching results in the matching feature vector.

[0007] According to an embodiment of the present application, the plurality of first prompt words are used to represent a plurality of information types to be extracted and corresponding structural position information, and a hierarchical order of the plurality of information types to be extracted is predefined between the plurality of first prompt words; by using the plurality of first prompt words, the large model is guided to extract the target business information from the business text based on the parsing rule and the structural position information, including: by using the plurality of first prompt words, the large model is guided to organize the parsing rule into a task sequence according to the hierarchical order; by using the task sequence, the large model is guided to sequentially perform field extraction on the business text according to the plurality of information types to be extracted and the corresponding structural position information, to obtain the target business information.

[0008] According to an embodiment of the present application, the method further includes: based on a preset second prompt word, guiding the large model to verify the target business information, the second prompt word being used to indicate a structural verification reasoning step and a business verification reasoning step of the target business information.

[0009] According to an embodiment of the present application, the second prompt word includes a plurality of first sub-prompt words and a plurality of second sub-prompt words, and based on the preset second prompt word, guiding the large model to verify the target business information includes: based on the plurality of first sub-prompt words, guiding the large model to call a preset standard field list, and based on the standard field list, performing the structural verification reasoning step on the target business information to identify missing information in the target information; based on the plurality of second sub-prompt words, guiding the large model to perform the business verification reasoning step on the target business information according to a predefined business rule.

[0010] According to an embodiment of the present application, the method further includes: guiding the large model to perform data verification on the target business information based on a preset field dependency tree, the field dependency tree being used to indicate a data verification basis of the target business information; in a case where the data verification is passed, guiding the large model to output the target business information in a hierarchical structure of the field dependency tree, the hierarchical structure being used to indicate an entry order of the target business information.

[0011] According to an embodiment of the present application, the rule library includes parsing rules indexed by scene identifier tags, and based on the scene identifier, the plurality of first prompt words are obtained by searching in the preconfigured rule library, including: based on a corresponding relationship between the scene identifier and the scene identifier tag, performing index matching in the rule library to obtain the corresponding parsing rule; performing structural integrity verification on the parsing rule, and converting the parsing rule that passes the verification into the plurality of first prompt words for output.

[0012] According to an embodiment of the present application, the data extraction state of the large model on the target business data is determined; in the case that the data extraction state is failed, an extraction trigger instruction is issued to guide the large model to perform the extraction operation on the target business data again, wherein the number of times of issuing the extraction trigger instruction is less than a preset threshold; in the case that the number of times of issuing is equal to the preset threshold and the data extraction state is failed, a clearing instruction is issued after a preset time threshold to guide the large model to clear the target business data; in the case that the data extraction state is successful, the clearing instruction is issued after the preset time threshold to guide the large model to clear the target business data.

[0013] The second aspect of the present application provides a business document processing apparatus, comprising: a text recognition module, configured to perform optical character recognition on an obtained target business document to obtain target business data, wherein the target data comprises business text and corresponding structural position information; a scene identifier determination module, configured to determine a scene identifier corresponding to the target business document based on content recognition of the business text, wherein the scene identifier is used to represent the content architecture of the target business document; a retrieval module, configured to retrieve a plurality of first prompt words in a preconfigured rule library based on the scene identifier, wherein the plurality of first prompt words are used to represent the parsing rules of the target business document; and an information extraction module, configured to guide a large model to extract the business text based on the parsing rules and the structural position information by using the plurality of first prompt words to obtain target business information.

[0014] The third aspect of the present application provides an electronic device, comprising: one or more processors; a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0015] The fourth aspect of the present application further provides a computer-readable storage medium having a computer program or instructions stored thereon, wherein the computer program or instructions are executed by a processor to implement the steps of the above method.

[0016] The fifth aspect of the present application further provides a computer program product comprising a computer program or instructions, wherein the computer program or instructions are executed by a processor to implement the steps of the above method.

[0017] According to the business document processing method, apparatus, device, medium and program product provided by the present application, the optical character recognition is used to preliminarily recognize the business document, realize text extraction and physical position positioning; the scene identifier is used to represent the content architecture of the business document, the parsing rules matched with the content architecture are dynamically loaded based on the scene identifier in the rule library, and the problem of differences in content architectures of business documents in different cities is solved; and the semantic understanding ability of the large model is combined to solve the difficult problem of complex table parsing. BRIEF DESCRIPTION OF DRAWINGS

[0018] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0019] Figure 1 The illustrations depict application scenarios of business document processing methods, apparatus, devices, media, and program products according to embodiments of this application.

[0020] Figure 2 A flowchart illustrating a business document processing method according to an embodiment of this application is shown schematically.

[0021] Figure 3 A flowchart illustrating the determination of a scene identifier according to an embodiment of this application is shown schematically;

[0022] Figure 4 A flowchart illustrating business text extraction according to an embodiment of this application is shown schematically;

[0023] Figure 5 A flowchart illustrating the verification of target service information according to an embodiment of this application is shown in the illustration.

[0024] Figure 6 A flowchart illustrating the output of target service information according to an embodiment of this application is shown schematically.

[0025] Figure 7 A flowchart illustrating the process of obtaining a first prompt word according to an embodiment of this application is shown schematically;

[0026] Figure 8 This schematically illustrates a structural block diagram of a business document processing apparatus according to an embodiment of the present application; and

[0027] Figure 9 A block diagram schematically illustrates an electronic device suitable for implementing a business document processing method according to an embodiment of this application. Detailed Implementation

[0028] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0029] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0030] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0031] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0032] The embodiments of this application provide a business document processing method, which pre-configures a large number of parsing rules in a rule base, uses scene identifiers to represent the content structure of business documents, and dynamically loads corresponding parsing rules based on scene identifiers to improve the efficiency of business document parsing; and combines optical character recognition with a large model to achieve complementarity between physical location information and semantic information, thereby improving the accuracy of business document parsing.

[0033] Figure 1 The illustration shows an application scenario diagram of the business document processing method according to an embodiment of this application.

[0034] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0035] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0036] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0037] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0038] It should be noted that the business document processing method provided in this application embodiment can generally be executed by server 105. Correspondingly, the business document processing device provided in this application embodiment can generally be located in server 105. The business document processing method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the business document processing device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0039] It should be understood that Figure 1 The number of terminal devices, networks 104, and servers 105 shown is merely illustrative. Any number of terminal devices, networks, and servers can be included depending on implementation needs.

[0040] It should be noted that the methods and apparatus provided in this application can be used in the field of fintech, or in any field other than fintech. The application fields of the methods and apparatus provided in this application are not limited.

[0041] The following will be based on Figure 1 The described scene, through Figures 2-7 The business document processing method according to the embodiments of this application will be described in detail.

[0042] Figure 2 A flowchart illustrating a business document processing method according to an embodiment of this application is shown.

[0043] like Figure 2 As shown, the business document processing method of this embodiment includes operations S210 to S240.

[0044] In operation S210, optical character recognition is performed on the acquired target business document to obtain target business data.

[0045] According to the embodiments of this application, the target business document is the banking business document to be processed, including enterprise annuity plan documents, enterprise supplementary pension insurance plan documents, insurance contracts, etc. The target data includes business text and corresponding structural location information. Optical character recognition, or OCR (Optical Character Recognition) technology, is a technology that uses computer technology to convert text information in images and portable documents into editable and searchable text formats.

[0046] For example, the target business document can be a corporate annuity plan document, which is a formal written document formulated by the company regarding the corporate annuity system, and is mostly in image or portable document format; the target data includes the corporate annuity plan text and the corresponding structural location information; for example, by performing OCR recognition on the acquired portable document format corporate annuity plan document, editable corporate annuity plan text and the structural location information corresponding to the text can be obtained, such as the equity vesting table position (x1, y1, x2, y2).

[0047] In operation S220, content recognition is performed based on business text to determine the scene identifier corresponding to the target business document.

[0048] According to embodiments of this application, a scenario identifier is used to represent the content architecture of a target business document.

[0049] Considering that the content structure of business documents may differ in different regions and from different customers, and that the content structure can also be referred to as the content format, in order to take into account the impact of the content structure on the parsing when parsing the target business document, according to the embodiments of this application, a scenario identifier is used to represent the content structure of the target business document, so as to dynamically filter the subsequent parsing rules based on the scenario identifier, thereby improving the adaptability of the parsing rules to the content structure and the parsing efficiency.

[0050] Specifically, if the content architecture in the target business document is general, the scenario identifier is 0; if the content architecture in the target business document is special, the scenario identifier is 1.

[0051] For example, if the target business document is a corporate annuity plan document and the content structure of the corporate annuity plan document is general, then the scenario identifier is 0.

[0052] When operating S230, the system searches the pre-configured rule base based on the scene identifier and obtains multiple first prompt words.

[0053] According to an embodiment of this application, the rule base refers to a database containing various parsing rules. Different parsing rules correspond to different scenario identifiers, that is, different parsing rules correspond to different content architectures. Parsing rules corresponding to different content architectures are pre-configured in the rule base to parse target business documents with different content architectures in a targeted manner. During parsing, the parsing rules can be dynamically retrieved according to the scenario identifier, without the need for special targeted development, thus improving parsing efficiency. Multiple first prompt words are used to represent the parsing rules of the target business document.

[0054] In operation S240, multiple first prompt words are used to guide the large model to extract business text based on parsing rules and structural location information, thereby obtaining the target business information.

[0055] According to an embodiment of this application, the first prompt word is a pre-defined text instruction used to guide the large model to extract target business information. It includes explicit constraints on the type of information to be extracted, structural location information, and hierarchical order between the types of information to be extracted, as well as guidance on the input, output, and how to process the input to obtain the output of the large model. The large model can be a large language model, which refers to a large model specifically used to process business documents in the banking business document processing scenario and extract target business information according to preset rules under the guidance of the prompt word. Its input can be target business data, and its output can be target business information.

[0056] For example, if the target business document is a corporate annuity plan document, then the target business information refers to business information related to the corporate annuity plan, including participants, corporate contributions, individual contributions, the method of allocating the balance of the public account to the individual account, transfers and payments, and the attribution of rights.

[0057] Based on this, the synergistic effect of large models and OCR recognition enables the extraction of target business information through the complementarity of physical structure information and semantic information, thereby improving the accuracy of parsing.

[0058] Considering the differences in the content format of business documents across different cities and customers, in order to ensure the accuracy of scene identifier recognition, in some embodiments, the scene identifier can be determined based on the uploaded business document. Figure 3 A flowchart illustrating the determination of scene identifiers according to an embodiment of this application is shown schematically.

[0059] like Figure 3 As shown, determining the scene identifier includes operations S310 to S320.

[0060] In operation S310, multiple module titles are extracted from the business text. Regular expressions are then used to match these multiple module titles sequentially with a preset general content framework to obtain a matching feature vector.

[0061] According to embodiments of this application, the general content framework includes a general content architecture and corresponding standard module titles, and the matching feature vector includes the matching results of all module titles.

[0062] Specifically, first, the title recognition regular expression is used to extract the title from the business text, clean up the numbering and formatting symbols in the title and remove duplicates to obtain the core text content as the module title; then, the module title is matched sequentially with the regular expression pattern corresponding to each standard module title until a matching standard module title is found; finally, a matching feature vector is generated based on the matching results, for example, a successful match corresponds to a feature value of 0, and a failed match corresponds to a feature value of 1; the regular expression pattern is a pre-defined regular expression for the standard module title, allowing other characters between keywords and the existence of keyword variations.

[0063] For example, if the target business document is a corporate annuity plan document, the general content framework includes the general content structure of the corporate annuity plan and the corresponding standard module titles. For example, in parallel hierarchical order, it includes general provisions, participants, fundraising and allocation, account management, equity ownership, fund management, benefit calculation and payment methods, plan changes and termination, organization management and supervision, and supplementary provisions. The business text is the corporate annuity plan text, and the titles in the corporate annuity plan text are in chapter format. The chapter titles are extracted using keywords such as Chapter 1 and Section 3 through title recognition regular expressions.

[0064] In operation S320, the scene identifier is obtained based on the accumulated value of the matching results in the matching feature vector.

[0065] According to an embodiment of this application, if the cumulative value of the matching result in the matching feature vector is 0, then the business document is completely matched with the general content framework, the content structure of the business document is general, and the scenario identifier is 0; if the cumulative value of the matching result in the matching feature vector is not 0, then it indicates that the business document is not completely matched with the general content framework, the content structure of the business document is special, and the scenario identifier is 1.

[0066] Based on this, by using chapter titles to reflect the content structure of business documents, the corresponding scenario identifiers can be determined according to the content structure of the business documents to be processed, ensuring a strong correlation between scenario identifiers and business documents, and providing accurate and representative basic data for the selection of parsing rules.

[0067] In some embodiments, the pre-configured rule base stores a variety of parsing rules corresponding to special case identifiers. In order to achieve accurate parsing of business documents, when the business documents and the general content framework do not match completely, similarity calculation is performed based on the enterprise attributes and enterprise location and the preset special case identifier tags to obtain the final scenario identifier.

[0068] According to an embodiment of this application, the special scenario identifier label marks common special enterprises and regions.

[0069] Figure 4 A flowchart illustrating business text extraction according to an embodiment of this application is shown.

[0070] Multiple first prompt words are used to represent multiple types of information to be extracted and their corresponding structural position information, and the hierarchical order among the multiple types of information to be extracted is predefined among the multiple first prompt words; such as Figure 4 As shown, according to an embodiment of this application, operation S240 includes operations S410 to S420.

[0071] When operating S410, multiple first cue words are used to guide the large model to organize the parsing rules into a task sequence according to the hierarchical order.

[0072] According to embodiments of this application, multiple first prompt words can correspond to the reasoning steps in the parsing rules according to the hierarchical order among the types of information to be extracted.

[0073] For example, if the target business document is an enterprise annuity business document and its corresponding scenario identifier is 0, then the multiple first prompt words can be: extracting participant-related information, participants are in Chapter 2 of the enterprise annuity plan, extracting fund-related information, the fund part is in Chapter 3 Fund Raising and Allocation, extracting equity-related information, equity is in Chapter 5 Equity Attribution, extracting transfer and payment-related information, transfer and payment are in Chapter 4 Account Management, Chapter 6 Fund Management, and Chapter 7 Benefit Calculation and Payment Methods; the business handler role can be the business personnel handling bank enterprise annuity business.

[0074] According to embodiments of this application, the process of guiding a large model to organize parsing rules into a task sequence according to a hierarchical order using the multiple first prompt words may include: First, the large model receives multiple first prompt words; then, the large model sorts the received first prompt words according to the hierarchical order among multiple types of information to be extracted predefined among the multiple first prompt words to generate a preliminary task sequence, for example, sorting them according to the chapter order reflected in the first prompt words; then, the large model converts the sorted multiple first prompt words into a task sequence according to task number-task objective-operation steps.

[0075] For example, if the target business document is a corporate annuity business document, then the task example in the task sequence can be: Task 1 - Determine the participation conditions of participants - Extract the participation conditions from the "Participants Chapter" of the corporate annuity plan document.

[0076] When operating S420, the task sequence guides the large model to sequentially extract fields from the business text based on multiple information types to be extracted and their corresponding structural location information, thereby obtaining the target business information.

[0077] According to an embodiment of this application, the type of information to be extracted corresponds to the task type in the task sequence, and the structural location information corresponds to the chapter. The large model executes the tasks sequentially according to the task order in the task sequence and extracts the corresponding business information in the corresponding chapter. At the same time, during the extraction process, the structural location information of the text obtained by OCR recognition provides a positioning benchmark for the large model, making it easier for it to find the corresponding chapter in a timely manner.

[0078] Based on this, the semantic understanding capability of the large model is combined with the physical location information provided by OCR recognition to extract target business information, providing an accurate physical location benchmark for the extraction task of the large model and improving the task execution efficiency of the large model.

[0079] In some embodiments, when pre-training a large model, role indicator words and predefined business rules are used to guide the large model to establish the role perspective of a business handler.

[0080] According to an embodiment of this application, business rules are used to represent business processing steps and corresponding required business information. When the big model receives role prompts and business rules, it automatically learns relevant knowledge so that during the processing of the target business document, the big model takes the business handler role as the core perspective and transforms the abstract parsing rules into an executable task sequence that fits the actual business scenario, ensuring that the task logic is consistent with the business processing process and avoiding the disconnect between technical extraction and actual operation.

[0081] In some embodiments, the method further includes: guiding a large model to verify corporate annuity business information based on a preset second prompt word.

[0082] According to an embodiment of this application, the second prompt word is used to indicate the structure verification reasoning step and the business verification reasoning step of the enterprise annuity business information.

[0083] The structural and business verification reasoning steps specified in the second prompt guide the large model to perform structural and business verification on the obtained corporate annuity business information, ensuring the completeness and rationality of the extracted information.

[0084] Figure 5 A flowchart illustrating the verification of target service information according to an embodiment of this application is shown.

[0085] like Figure 5As shown, in some embodiments, the second prompt word includes multiple first sub-prompt words and multiple second sub-prompt words. Based on the preset second prompt word, the large model is guided to verify the enterprise annuity business information, including operations S510 to S520.

[0086] When operating S510, based on multiple first sub-prompt words, the large model is guided to call a pre-set list of standard fields, and based on the list of standard fields, structural verification and reasoning steps are performed on the corporate annuity business information to identify missing information in the corporate annuity business information.

[0087] According to embodiments of this application, the standard field list indicates the fields that must be included in the business document, and multiple first sub-prompt words indicate the role positioning and structural verification reasoning steps of the large model. For example, the standard field list corresponding to the business document is called, and the target business information is checked to see if all standard fields are included based on the standard field list. Fields with the same or similar semantics can be regarded as the same field. If a field is not mentioned and the field has no default value, an empty value is returned.

[0088] For example, if the target business document is a corporate annuity plan document, then the business text is the corporate annuity plan text. First, the large model receives multiple first sub-prompt words and verifies the corporate annuity business information field by field based on the standard field list. If a certain field is not mentioned and the field has no default value, then an empty value is returned. If the empty value field is indispensable information, then the corresponding supplementary requirements are output. For example, if the empty value is the contribution base of the insured person, the supplementary requirement is to verify the base according to the salary table and fill it in.

[0089] When operating S520, based on multiple second sub-prompt words, the large model is guided to perform business verification and reasoning steps on the target business information according to predefined business rules.

[0090] According to an embodiment of this application, multiple second sub-prompt words are used to instruct the large model to perform business verification and reasoning steps. Specifically, after receiving multiple second sub-prompt words, the large model calls predefined business rules under the instruction of the multiple second sub-prompt words and performs business verification and reasoning steps on the target business information according to the business rules.

[0091] For example, the target business information is enterprise annuity business information, and the business verification reasoning steps may include: for the equity vesting table in the enterprise annuity business information, verify its vesting ratio, and verify whether the vesting ratio is 100% when the working years are at the highest level.

[0092] By verifying both the structure and the business operations, the accuracy and completeness of the enterprise annuity business information extracted from the large model can be assessed, thereby improving the accuracy of the verification.

[0093] Figure 6A flowchart illustrating the display of target service information according to an embodiment of this application is shown schematically.

[0094] like Figure 6 As shown, in some embodiments, the method further includes operations S610 to S620.

[0095] When operating S610, guide the large model to perform data validation on the target business information based on the preset field dependency tree.

[0096] According to an embodiment of this application, the field dependency tree is used to indicate the data verification basis of the target business information, and the field dependency tree represents a hierarchical dependency relationship based on business logic.

[0097] Here, the field dependency tree is pre-stored in the system in JSON format.

[0098] For example, for a corporate annuity plan document, the field dependency tree includes parent nodes for participants, parent nodes for fundraising and allocation, parent nodes for equity ownership, and parent nodes for transfer and payment. Business rules, calculation parameters, and supplementary instructions related to the parent nodes are used as their child nodes, and dependency relationships are established according to the business process. For example, the parent node for participants has participation condition nodes and exit condition nodes. Only after the information of the parent node for participants is determined can the contribution amount be calculated.

[0099] According to an embodiment of this application, firstly, inputting a prompt word guides the large model to call a preset field dependency tree corresponding to the target business document. Then, the large model checks against the field dependency tree whether there are missing values ​​in the fields of the target business information, whether there are duplicate extractions or misalignments in the fields of the target business information.

[0100] For example, the target business information is enterprise annuity plan business information. The process of guiding the large model to perform data verification on the enterprise annuity business information based on the preset field dependency tree can be as follows: First, input prompt words guide the large model to call the preset field dependency tree corresponding to the enterprise annuity plan document. Then, the large model checks the field dependency tree to see if there are missing values, duplicate extractions, misaligned extractions, etc. in the fields of the enterprise annuity business information. For example, in the enterprise payment section, if the payment method is identified as Method 1, but the subfields of the extracted information are filled with both Method 1 and Method 2, then the actual verification return should only return Method 1.

[0101] When operating S620, if the data verification passes, guide the large model to output the target business information in a hierarchical structure of field dependency tree.

[0102] According to an embodiment of this application, the hierarchical structure is used to indicate the input order of the target business information, and the extracted target business information is backfilled into the corresponding field node of the field dependency tree based on the information type corresponding to each field in the target business information.

[0103] For example, if the target business information is enterprise annuity business information, then the extracted enterprise annuity business information will be backfilled to the corresponding field node of the field dependency tree based on the information type corresponding to each field in the enterprise annuity business information; for example, the payment frequency will be backfilled to the payment frequency node under the fund collection and allocation parent node, and the individual payment base type will be backfilled to the payment frequency node under the fund collection and allocation parent node.

[0104] Based on this, the field dependency tree is used to provide an intuitive validation basis and data display method for large models, making implicit field relationships explicit and structured, which is convenient for large models to understand.

[0105] Figure 7 A flowchart illustrating the process of obtaining a first prompt word according to an embodiment of this application is shown.

[0106] like Figure 7 As shown, operation S230 includes operations 710 to S720.

[0107] When operating S710, the corresponding parsing rules are obtained by indexing and matching in the rule base based on the correspondence between scene identifiers and scene identifier labels.

[0108] According to an embodiment of this application, the rule base includes multiple parsing rules indexed by scene identifier tags. The correspondence between scene identifiers and parsing rules can be clearly defined through scene identifier tags. Based on this, the response speed and processing efficiency of dynamically filtering and verifying rules based on scene identifiers can be improved.

[0109] Specifically, the system searches the rule base using scene identifiers as search criteria, matches the corresponding scene identifier tags in the rule base, and filters out parsing rules where the scene identifier tags match the scene identifiers.

[0110] When operating the S720, the parsing rules are structurally validated, and the validated parsing rules are converted into multiple first prompt words for output.

[0111] The parsing rules are structurally validated to ensure they contain the necessary information to guide the large model in information extraction. Specifically, the parsing rules are fuzzy matched based on a pre-defined list of required fields to verify whether they contain all required fields. If they do, the validation passes, and the parsing rules are converted into multiple first prompt words according to a pre-defined prompt word format.

[0112] For example, the cosine similarity between a field in the parsing rule and each of the required fields in the preset list of required fields is calculated, and a match is considered successful if the cosine similarity is greater than the similarity threshold.

[0113] In some embodiments, the method further includes: determining the data extraction status of the target business data by the large model; issuing an extraction trigger instruction to guide the large model to re-execute the extraction operation on the target business data if the data extraction status fails, wherein the number of times the extraction trigger instruction is issued is less than a preset threshold; issuing a clearing instruction after a preset time threshold if the number of times the instruction is issued equals the preset threshold and the data extraction status fails, to guide the large model to clear the target business data; and issuing a clearing instruction after a preset time threshold if the data extraction status is successful, to guide the large model to clear the target business data.

[0114] According to an embodiment of this application, the data extraction status is used to indicate the completion status of the target business data extraction task performed by the large model.

[0115] For example, the preset threshold is 3, and the preset time threshold is 7 days; that is, if the large model successfully executes the extraction task, the relevant task data will be cleared after 7 days, and if the large model fails to execute the extraction task multiple times, the relevant task data will be cleared after 7 days to avoid zombie tasks occupying system resources.

[0116] Based on the above-described business document processing method, this application also provides a business document processing apparatus. The following will be combined with... Figure 8 The device is described in detail.

[0117] Figure 8 A schematic block diagram of a business document processing apparatus 800 according to an embodiment of this application is shown.

[0118] like Figure 8 As shown, the business document processing device 800 of this embodiment includes a text recognition module 810, a scene identifier determination module 820, a retrieval module 830, and an information extraction module 840.

[0119] The text recognition module 810 is used to perform optical character recognition on the acquired target business document to obtain target business data, which includes business text and corresponding structural location information. In one embodiment, the text recognition module 810 can be used to perform the operation S210 described above, which will not be repeated here.

[0120] The scene identifier determination module 820 is used to perform content recognition based on business text and determine the scene identifier corresponding to the target business document. The scene identifier is used to represent the content structure of the target business document. In one embodiment, the scene identifier determination module 820 can be used to perform the operation S220 described above, which will not be repeated here.

[0121] The retrieval module 830 is used to retrieve data from a pre-configured rule base based on a scenario identifier, obtaining multiple first prompt words. These first prompt words represent the parsing rules of the target business document. In one embodiment, the retrieval module 830 can be used to perform the operation S230 described above, which will not be repeated here.

[0122] The information extraction module 840 is used to guide the large model to extract business text based on parsing rules and structural location information using multiple first prompt words, thereby obtaining the target business information. In one embodiment, the information extraction module 840 can be used to perform the operation S240 described above, which will not be repeated here.

[0123] According to an embodiment of this application, the business document processing device 800 further includes a verification module. The verification module is used to guide a large model to verify enterprise annuity business information based on a preset second prompt word. The second prompt word indicates the structural verification reasoning steps and business verification reasoning steps of the enterprise annuity business information. In one embodiment, the retrieval module can be used to perform the operations S510~S520 described above.

[0124] According to an embodiment of this application, the business document processing device 800 further includes a display module, which is used to perform the operations S610~S620 described above.

[0125] According to an embodiment of this application, the business document processing device 800 further includes a task management module, which is used to determine the data extraction status of the target business data by the large model; if the data extraction status fails, an extraction trigger instruction is issued to guide the large model to re-execute the extraction operation on the target business data, wherein the number of times the extraction trigger instruction is issued is less than a preset threshold; if the number of times the instruction is issued is equal to the preset threshold and the data extraction status fails, a clearing instruction is issued after a preset time threshold to guide the large model to clear the target business data; if the data extraction status is successful, a clearing instruction is issued after a preset time threshold to guide the large model to clear the target business data.

[0126] According to an embodiment of this application, the scene identifier determination module 820 includes a title extraction submodule and a determination submodule. The title extraction submodule is used to extract multiple module titles from the enterprise annuity plan text, and use regular expressions to match the multiple module titles with a preset general content framework in sequence to obtain a matching feature vector. The determination submodule is used to obtain the scene identifier based on the accumulated value of the matching results in the matching feature vector.

[0127] According to an embodiment of this application, the retrieval module 830 includes a matching submodule and a verification submodule. The matching submodule is used to perform index matching in the rule base based on the correspondence between scene identifiers and scene identifier tags to obtain the corresponding parsing rules. The verification submodule is used to perform structural integrity verification on the parsing rules and convert the verified parsing rules into multiple first prompt words for output.

[0128] According to an embodiment of this application, the information extraction module 840 includes a task determination submodule and an extraction submodule. The task determination submodule is used to guide the large model to organize the parsing rules into a task sequence in a hierarchical order using multiple first prompt words. The extraction submodule is used to guide the large model to organize the parsing rules into a task sequence in a hierarchical order using multiple first prompt words.

[0129] According to embodiments of this application, any multiple modules among the text recognition module 810, scene identifier determination module 820, retrieval module 830, and information extraction module 840 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the text recognition module 810, scene identifier determination module 820, retrieval module 830, and information extraction module 840 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any one of the three implementation methods, or in a suitable combination of any of them. Alternatively, at least one of the text recognition module 810, scene identifier determination module 820, retrieval module 830, and information extraction module 840 may be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0130] Figure 9 A block diagram schematically illustrates an electronic device suitable for implementing a business document processing method according to an embodiment of this application.

[0131] like Figure 9As shown, an electronic device 900 according to an embodiment of this application includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage portion 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0132] RAM 903 stores various programs and data required for the operation of electronic device 900. Processor 901, ROM 902, and RAM 903 are interconnected via bus 904. Processor 901 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 902 and / or RAM 903. It should be noted that the programs may also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0133] According to embodiments of this application, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to a bus 904. The electronic device 900 may also include one or more of the following components connected to the input / output (I / O) interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 910 as needed so that computer programs read from it can be installed into the storage section 908 as needed.

[0134] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0135] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 902 and / or RAM 903 and / or one or more memories other than ROM 902 and RAM 903 described above.

[0136] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this application.

[0137] When the computer program is executed by the processor 901, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0138] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 909, and / or installed from a removable medium 911. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0139] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0140] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0141] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0142] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A business document processing method, characterized in that, The method includes: The acquired target business document is subjected to optical character recognition to obtain target business data, which includes business text and corresponding structural location information. Based on the business text, content recognition is performed to determine the scene identifier corresponding to the target business document. The scene identifier is used to represent the content structure of the target business document. Based on the scenario identifier, a search is performed in a pre-configured rule base to obtain multiple first prompt words, which are used to represent the parsing rules of the target business document; Using the multiple first prompt words, the large model is guided to extract the business text based on the parsing rules and the structural location information to obtain the target business information.

2. The method according to claim 1, characterized in that, The step of performing content recognition based on the business text to determine the scene identifier corresponding to the target business document includes: Extract multiple module titles from the business text, and use regular expressions to match the multiple module titles with a preset general content framework in sequence to obtain a matching feature vector. The general content framework includes a general content structure and corresponding standard module titles. The scene identifier is obtained based on the accumulated value of the matching results in the matching feature vector.

3. The method according to claim 1, characterized in that, The plurality of first prompt words are used to represent a plurality of information types to be extracted and the corresponding structural position information, and a hierarchical order of the plurality of information types to be extracted is predefined among the plurality of first prompt words; The step of using the multiple first prompt words to guide the large model to extract the business text based on the parsing rules and the structural position information, and obtaining the target business information, includes: Using the multiple first prompt words, the large model is guided to organize the parsing rules into a task sequence according to the hierarchical order; Using the task sequence, the large model is guided to sequentially extract fields from the business text according to the multiple types of information to be extracted and the corresponding structural location information, thereby obtaining the target business information.

4. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Based on a preset second prompt word, the large model is guided to verify the target business information. The second prompt word is used to indicate the structural verification reasoning steps and the business verification reasoning steps of the target business information.

5. The method according to claim 4, characterized in that, The second prompt word includes multiple first sub-prompt words and multiple second sub-prompt words. The step of guiding the large model to verify the target business information based on the preset second prompt word includes: Based on the multiple first sub-prompt words, the large model is guided to call a pre-set list of standard fields, and the structural verification and reasoning steps are performed on the target business information based on the list of standard fields to identify missing information in the target business information; Based on the multiple second sub-prompt words, the large model is guided to perform the business verification and reasoning steps on the target business information according to predefined business rules.

6. The method according to any one of claims 1 to 3, characterized in that, The method further includes: The large model is guided to perform data verification on the target business information based on a preset field dependency tree, wherein the field dependency tree is used to indicate the basis for data verification of the target business information; If the data verification passes, the large model is guided to output the target business information in the hierarchical structure of the field dependency tree, and the hierarchical structure is used to indicate the input order of the target business information.

7. The method according to claim 1, characterized in that, The rule base includes parsing rules indexed by scene identifier tags. The retrieval of multiple first prompt words based on the scene identifiers within the pre-configured rule base includes: Based on the correspondence between the scene identifier and the scene identifier label, an index matching is performed in the rule base to obtain the corresponding parsing rule; The parsing rules are subjected to structural integrity verification, and the parsing rules that pass the verification are converted into the plurality of first prompt words for output.

8. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Determine the data extraction status of the target business data by the large model; If the data extraction status fails, an extraction trigger command is issued to guide the large model to re-execute the extraction operation on the target business data, wherein the number of times the extraction trigger command is issued is less than a preset threshold. If the number of times the data is sent is equal to the preset threshold and the data extraction status is failed, a clearing command is issued after a preset time threshold to guide the large model to clear the target business data. If the data extraction status is successful, the clearing command is issued after the preset time threshold to guide the large model to clear the target business data.

9. A business document processing device, characterized in that, The device includes: The text recognition module is used to perform optical character recognition on the acquired target business document to obtain target business data, wherein the target data includes business text and corresponding structural location information. The scene identifier determination module is used to perform content recognition based on the business text and determine the scene identifier corresponding to the target business document. The scene identifier is used to represent the content structure of the target business document. The retrieval module searches in a pre-configured rule base based on the scenario identifier to obtain multiple first prompt words, which are used to represent the parsing rules of the target business document; The information extraction module is used to guide the large model to extract the business text based on the parsing rules and the structural location information using the multiple first prompt words, so as to obtain the target business information.

10. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 8.

12. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 8.

Images