Artificial intelligence-based approval process attachment pre-auditing method and system

Abstract

The invention provides an approval process attachment pre-auditing method and system based on artificial intelligence, and the method comprises the steps: obtaining a business field and an uploading attachment in a form in response to a submission operation of a user in an approval process, and transmitting the business field and the uploading attachment to an AI processing platform; based on the business field, determining and executing a corresponding workflow, the execution of the workflow including traversing the attachments, judging the document type of each attachment based on the business field and the content of the attachments, and executing structured extraction of key parameters on the document type meeting a preset condition; receiving and displaying an extraction result; receiving a confirmation or modification instruction of the user; in response to the confirmation instruction, controlling the flow to flow to the next node; or in response to the modification instruction, allowing the user to modify the form or the attachment and re-triggering the pre-auditing. According to the invention, the AI capability is deeply integrated into the approval process, automatic and precise attachment pre-auditing is realized, the approval efficiency and accuracy are greatly improved, a man-machine collaborative intelligent closed loop is formed through an initiator confirmation link, and the user experience is optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of enterprise informatization and artificial intelligence, and in particular to an approval process attachment pre-examination method and system based on artificial intelligence. BACKGROUND

[0002] In modern enterprise operation, the approval process is the core link to support the compliance and efficient operation of various business activities. Whether it is procurement, reimbursement, project establishment, or business scenarios such as contract signing and project change, the process initiator is usually required to fill out a structured form and upload supporting attachments (such as contracts, price lists, invoices, project plans, etc.). These attachments are mostly unstructured documents (PDF, Word, Excel, etc.), which contain a large amount of key business parameters and information. In the traditional process, the approver needs to manually open the attachments and check whether the contents in the attachments are consistent with the information filled in the form and whether they comply with the company's regulations and policies. This process not only consumes time and effort, but also highly depends on the experience, meticulousness and available time of the approver. With the increasing degree of enterprise informatization and the expansion of business scale, the number and complexity of approval processes have increased dramatically, and relying on pure manual attachment review has become a key bottleneck restricting the efficiency and quality of the process.

[0003] Currently, some technologies have attempted to solve the problem of document content recognition and extraction, such as optical character recognition (OCR) technology which can convert scanned documents into editable text, and natural language processing (NLP) technology which can extract structured information from text. However, these technologies mostly exist as independent tools or external services and have not been deeply integrated with the enterprise's internal approval process system. Existing approval systems usually only have file uploading and storage functions and cannot automatically trigger document parsing tasks at process nodes or provide real-time feedback of extraction results to initiators or approvers. Therefore, even if external artificial intelligence (AI) services are introduced, manual downloading of attachments, calling of tools, extraction of information and comparison are still required, and the process is fragmented and cannot be applied on a large scale. This separation of "system to system" and "AI to AI" makes it difficult for enterprises to build an intelligent approval closed loop with event-driven, automatic flow and human-machine collaboration, and it is still unable to truly realize the automation and intelligent upgrade of the approval process, and the fundamental problems of low efficiency, easy errors and poor experience in manual review have not been solved. SUMMARY

[0004] In view of this, the present application provides an approval process attachment pre-examination method and system based on artificial intelligence to solve the problems of AI capability and approval process system fragmentation, low efficiency and errors in manual review of attachments, and poor experience caused by repeated communication between initiators and approvers due to attachment problems.

[0005] In one aspect, the present application provides an artificial intelligence-based approval process attachment pre-examination method, the method comprising: In response to a submission operation initiated by a user in an approval process, at least one business field in a form corresponding to the approval process and at least one attachment uploaded by the user are obtained, and the business field and the attachment are sent to an artificial intelligence processing platform; Based on the received business field, a corresponding pre-configured workflow is determined and executed in the artificial intelligence processing platform; wherein executing the pre-configured workflow comprises: traversing the attachments, judging the document type of each attachment based on the business field and the attachment content, and for the document type that meets the preset condition, performing structured extraction of key parameters; Receiving the structured extraction result returned by the artificial intelligence processing platform, and returning the extraction result to the approval process system to display to the user; Receiving a confirmation instruction or a modification instruction made by the user based on the extraction result; In response to the confirmation instruction, the approval process is controlled to flow to the next approval node; or in response to the modification instruction, the user is allowed to modify the form or the attachment, and the pre-examination method is retriggered based on the modified form or attachment.

[0006] In some embodiments of the present application, judging the document type of each attachment based on the business field and the attachment content comprises: The business field is used as a routing condition to select a pre-configured document judgment model; The attachment content is input into the document judgment model to obtain the type identification of the attachment; wherein the document judgment model is a large language model, the prompt word containing the business field is input into the large language model, and the output format is constrained to realize document type judgment.

[0007] In some embodiments of the present application, for the document type that meets the preset condition, the structured extraction of key parameters is performed, comprising: The type identification is used as a selection condition to call a parameter extraction model corresponding to the document type; The attachment content is input into the parameter extraction model to extract the pre-defined field and output as a structured data format; wherein the parameter extraction model is a large language model, the prompt word containing the document content is input into the large language model, and the output format is constrained to realize structured parameter extraction.

[0008] In some embodiments of the present application, the pre-configured workflow realizes branch logic through a selector node, and the selector node routes the processing flow to one of a plurality of different business branches according to the value of the business field.

[0009] In some embodiments of the present application, after the structured extraction result is generated, the method further comprises: comparing and analyzing the parameters contained in the structured extraction result with the field values filled in the form by the user for consistency; returning and displaying the comparison and analysis result as part of the extraction result to the user.

[0010] In some embodiments of the present application, the modification instruction made by the user based on the extraction result is received, the user is allowed to modify the form or the attachment, and the pre-examination method is retriggered, comprising: in response to the modification instruction, controlling the approval process to back up to the initiation node; receiving the modification operation and resubmission operation of the user on the form or the attachment at the initiation node; in response to the resubmission operation, re-executing the pre-examination method.

[0011] In some embodiments of the present application, the steps of the method implemented in the artificial intelligence processing platform are defined by a visual workflow composed of multiple nodes, including at least a logic node for branch selection, a loop node for processing multiple attachments, and a model node for calling an artificial intelligence model service.

[0012] In another aspect, the present application provides an artificial intelligence-based approval process attachment pre-examination system, which is used to implement the method according to any one of the above embodiments, and the system comprises: a form module for providing an approval template, receiving form data input by a user, and uploading an attachment file; an event triggering and interface module coupled with the form module, for automatically packaging business fields in the form and the attachment file after listening to a process submission event, and calling an interface of an artificial intelligence processing platform to send data; a result display and interaction module for receiving and displaying the structured extraction result returned by the artificial intelligence processing platform, and providing an interactive interface for confirmation and modification; a process control engine for controlling the flow direction of the approval process according to the instruction made by the user through the interactive interface; the artificial intelligence processing platform, which is communicatively connected to the event triggering and interface module, for executing a corresponding preconfigured workflow according to the received business fields to complete the document type judgment and key parameter extraction of the attachment, and returning the result.

[0013] In another aspect, the present application provides a computer readable storage medium having stored thereon computer programs / instructions which, when executed by a processor, implement the steps of the method of any one of the above mentioned aspects The present application provides an artificial intelligence-based approval process attachment pre-examination method and system, which realizes revolutionary improvement on the traditional manual examination mode by deeply and seamlessly embedding artificial intelligence capability into the existing approval process, and significantly improves the intelligent level of enterprise operation.

[0014] Further, the present application constructs an intelligent processing closed loop based on dynamic driving of business fields. The system can automatically schedule corresponding artificial intelligence workflow according to different business scenarios, accurately identify the type of unstructured attachments and extract key information, and convert the complex content that needs to be checked by manual into instant verifiable structured data. This process not only liberates the approver from the tedious document review work, greatly shortens the process time, but also ensures the uniformity and objectivity of the examination standard through the machine executing the preset rules, effectively avoids the problems of wrong examination and missed examination caused by human negligence or experience difference, and greatly improves the approval quality and compliance.

[0015] Further, the present application establishes an instant feedback and correction mechanism with the initiator as the core. By directly presenting the artificial intelligence extraction result to the initiator for confirmation before submission, data problems can be discovered and corrected in time at the starting point of the process, completely changing the long and inefficient cycle of "submission-rejection-communication-resubmission" in the traditional mode. This pre-checking design changes passive error correction after the fact into proactive prevention beforehand, greatly optimizing the operation experience of the initiator, reducing the communication cost between roles, and realizing the double leap of process efficiency and user experience. Finally, the present application forms a high-efficiency, reliable and self-optimizing human-computer collaborative approval system, providing a practical solution for enterprise process automation and intelligentization.

[0016] Additional advantages, objects, and features of the application will be set forth in part by the description that follows, and will in part be apparent to those of ordinary skill in the art upon examination of the following or can be learned from practice of the application. The advantages and features of the application can be realized and obtained by means of the instrumentalities and combinations pointed out in the appended claims.

[0017] Those skilled in the art will understand that the objects and advantages of the present application can be realized and obtained by means of the instrumentalities specifically pointed out in the appended claims and combinations thereof. The objects and advantages of the present application will become more fully understood from the following detailed description, particularly when reviewed in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings: Figure 1 A flowchart of the method for pre-inspecting an attachment based on an artificial intelligence in an approval process according to an embodiment of the present application.

[0019] Figure 2 A flowchart of the method for pre-inspecting an attachment based on an artificial intelligence in an approval process according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the objectives, technical solutions, and advantages of the present application clearer, the present application will be described in further detail below with reference to the embodiments and drawings. Here, the illustrative embodiments of the present application and their descriptions are used to explain the present application, but are not intended to limit the present application.

[0021] It should also be noted that, in order to avoid obscuring the present application due to unnecessary details, only structures and / or processing steps closely related to the solutions according to the present application are shown in the drawings, and other details not closely related to the present application are omitted.

[0022] It should be emphasized that the terms "comprises / comprising" when used in this text, refer to the presence of the stated features, elements, steps or components, but do not exclude the presence or addition of one or more other features, elements, steps or components.

[0023] It should also be noted that, if not specifically stated, the term "connected" in this text can not only mean direct connection, but also indirect connection with an intermediate.

[0024] In the following, embodiments of the present application will be described with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts or the same or similar steps.

[0025] It should be emphasized here that the step designations mentioned in the following are not intended to limit the order of the steps, but it should be understood that the steps can be performed in the order mentioned in the embodiments, or in a different order from the embodiments, or several steps can be performed simultaneously.

[0026] In order to solve the problems of the prior art, such as the AI capability being separated from the approval process system, the low efficiency and errors of manual review of attachments, and the poor experience caused by repeated communication between the initiator and the approver due to attachment problems, the present application provides a method for pre-inspecting an attachment based on an artificial intelligence in an approval process, as shown in Figure 1 The method includes the following steps S101-S105: Step S101: In response to a submission operation initiated by a user in an approval process, at least one business field in a form corresponding to the approval process and at least one attachment uploaded by the user are acquired, and the business field and the attachment are sent to an artificial intelligence processing platform.

[0027] Step S102: Based on the received business field, a corresponding preconfigured workflow is determined and executed in the artificial intelligence processing platform. The execution of the preconfigured workflow includes: traversing the attachment, judging the document type of each attachment based on the business field and the attachment content, and for the document type meeting the preset condition, performing structured extraction of key parameters.

[0028] Step S103: The structured extraction result returned by the artificial intelligence processing platform is received, and the extraction result is returned to the approval process system to be displayed to the user.

[0029] Step S104: A confirmation instruction or a modification instruction made by the user based on the extraction result is received.

[0030] Step S105: In response to the confirmation instruction, the approval process is controlled to flow to a next approval node; or in response to the modification instruction, the user is allowed to modify the form or the attachment, and the pre-inspection method is retriggered based on the modified form or the attachment.

[0031] As shown in FIG. 1, it is a flowchart of an approval process attachment pre-inspection method based on artificial intelligence. Figure 2

[0032] In step S101, in response to a submission operation initiated by a user in an approval process, at least one business field in a form corresponding to the approval process and at least one attachment uploaded by the user are acquired, and the business field and the attachment are sent to an artificial intelligence processing platform.

[0033] This step is the triggering and data preparation stage of the intelligent pre-inspection process. The core purpose is to automatically capture the key event when the user completes the form filling and attachment uploading and performs the submission action, and accurately extract the two types of core data, i.e., structured business fields and unstructured attachment files, which are necessary for subsequent AI processing from the current approval process instance, complete data encapsulation, and reliably transmit them to the external artificial intelligence (AI) processing platform. This step realizes the event-driven data bridging between the traditional manual approval system and the intelligent processing service.

[0034] In some embodiments, step S101 mainly includes four parts of event listening and capturing, target data extraction, data standardization encapsulation, and cross-platform interface calling, which are completed by the form module, the event triggering and interface module, and the underlying process control engine in the corresponding system. Specifically: ​First, the process control engine predefines key operation events including "submit" for each approval process template. When a user completes all mandatory fields and uploads attachments in the approval form interface and clicks the "submit" button, the process control engine will immediately generate a standard process submission event.

[0035] Subsequently, a specially configured event trigger and interface module continuously monitors this event. This module is deeply integrated with the process control engine to ensure that it can capture this event in milliseconds and reliably trigger subsequent automated data processing chains as soon as the user completes the submission operation.

[0036] After the event is triggered, the event trigger and interface module extracts data from the currently active approval process instance according to pre-defined configuration rules.

[0037] The module accurately locates and extracts at least one key business field from the user-filled form according to the rules. These fields are usually the core identifiers that determine the subsequent AI processing logic, such as business type, cooperation method, configuration requirements, etc. When extracting, the module reads the values of these fields in the database.

[0038] At the same time, the module obtains the storage identifiers (such as file IDs) of all the user-uploaded attachment files in this process or directly reads the file binary stream. The attachments may include PDF, Word, Excel, and other unstructured documents in various formats.

[0039] Since the extracted business fields and attachment information are heterogeneous (text values and files), in order to transmit them through a network interface, the event trigger and interface module will be encapsulated according to the pre-defined data structure that is strictly compatible with the AI processing platform Application Programming Interface (API).

[0040] In some embodiments, encapsulation usually generates a structured request data object, such as a JSON object. In this object: Business fields are mapped to specific key-value pairs, such as "biz_type": "first configuration".

[0041] Attachment information is processed as a list of file IDs or embedded by Base64 encoding the file content through multipart / form-data.

[0042] After data encapsulation is complete, the event trigger and interface module will initiate a synchronous or asynchronous call to the AI processing platform external API through pre-set network configurations such as API endpoint address, request method, authentication key, etc.

[0043] The call sends the encapsulated request data object as a payload. For example, the call to the AI platform’s “document upload interface” transmits the attachment, and the subsequent call to the “workflow trigger interface” is accompanied by the business field and file ID.

[0044] The success or failure of the interface call, as well as the initial response of the AI platform (such as receiving a successful confirmation), is recorded by the module to ensure traceability of data transmission.

[0045] In step S102, based on the received business field, the corresponding preconfigured workflow in the artificial intelligence processing platform is determined and executed. Wherein, the execution of the preconfigured workflow includes: traversing the attachment, judging the document type of each attachment based on the business field and the attachment content, and executing the structured extraction of key parameters for the document type that meets the preset condition.

[0046] This step is the core processing stage of the intelligent pre-examination process, which occurs inside the artificial intelligence (AI) processing platform. Its core purpose is to receive data from the examination and approval system, dynamically schedule and execute the corresponding AI processing pipeline according to the specific business scenario, to complete the task of intelligently identifying, filtering and extracting key business parameters from unstructured attachments. This step deeply integrates general AI capabilities with specific approval business rules, achieving scene-based, accurate and automated processing logic.

[0047] In some embodiments, step S102 mainly includes three parts of receiving data and workflow routing, executing preconfigured workflow, and generating final output, which are implemented in the AI workflow platform, involving API access layer, rule configuration engine, workflow execution engine, document parsing component and collaborative work of various AI model services. Specifically: The API access layer of the AI processing platform receives and verifies the request data from the examination and approval system, and parses the business field and attachment information.

[0048] The rule configuration engine (or workflow scheduler) of the platform takes the received business field values, such as “business type: standard factor factoring financing” and “cooperation mode: bank credit”, as routing keys.

[0049] The engine matches multiple workflow templates previously preconfigured for different business scenarios. According to the combination of business fields, a corresponding preconfigured workflow instance is uniquely determined and activated. For example, “cooperation mode = bank credit” will trigger a workflow specifically designed to process credit-related documents (such as three-party agreement, investor information table, confirmation letter).

[0050] The activated workflow is an automatic execution sequence visualized by multiple functional nodes. This visualized arrangement through drag-and-drop selectors, loopers, model calls, and other nodes enables business or technical personnel to intuitively and flexibly configure and adjust AI review rules and processes to adapt to different business scenarios without writing complex code, greatly improving the maintainability, scalability, and business adaptability of the system.

[0051] In some embodiments, executing the preconfigured workflow includes the following steps: (1) Attachment traversal and content parsing: After the workflow is started, it first enters a loop node. This node traverses all the attachments in the request.

[0052] For each attachment in the loop, the workflow calls a link reading plug-in or a similar document parsing engine to parse the content of the attachments in formats such as PDF, Word, Excel, etc. into pure text or structured text data that can be processed by AI models (2) Document type judgment based on business logic: The workflow realizes the core branch through a selector node. This node first determines which business processing branch (such as the bank credit branch) should be entered according to the incoming business field (such as the cooperation method).

[0053] Within the specific branch, in order to accurately locate the target file from multiple attachments, the workflow calls a document judgment model. Preferably, the document judgment model is usually a large language model. The workflow inputs the parsed text of the current attachment together with related business field information into the model through a designed prompt word. For example, the prompt word can be: You are a document classifier. The current business scenario is "bank credit". Please judge whether the following document content is a "three-party agreement" or a similar legal agreement file. If it is, output the number "1"; if not, output the number "0". The document content is as follows:

Fill in the parsed text of the attachment

[0054] The model outputs a type identifier (such as numbers 1, 2, 3, or a Boolean value) based on its understanding of the content and the business context. For example, in the bank credit branch, the model identifies the attachment as "1-three-party agreement", "2-investor information table", "3-confirmation letter", or "4-other".

[0055] (3) Targeted key parameter extraction: The subsequent selector node in the workflow will make a judgment based on the type identifier obtained in the previous step. Only when the attachment is identified as the preset target document type (such as 1, 2, 3), will the process enter the corresponding parameter extraction link.

[0056] For attachments that need to be extracted, the workflow calls the corresponding parameter extraction model. Preferably, the parameter extraction model also uses a large language model.

[0057] The workflow inputs the complete parsed text of the attachment into another set of prompt input models focused on information extraction. For example, the prompt can be: You are an information extractor. Please find the specific values or content of "credit limit", "platform fee rate", and "validity period" from the following protocol text, and output them strictly in the following JSON format, ensuring the accuracy of the key names: {"credit_line": "found credit limit value", "platform_fee_rate": "found platform fee rate value", "term": "found validity period value"}. The protocol text is as follows:

Fill in the parsed text of the attachment here

[0058] The model understands the document semantics, locates and extracts predefined business parameters, and outputs them strictly in the specified format (such as JSON), generating structured key-value pair data.

[0059] (4) Result integration and enhanced analysis: The loop node of the workflow ends after processing all attachments. The system aggregates the structured data (JSON) extracted from each attachment.

[0060] In some embodiments, the workflow can include an integrated analysis large model node. This node receives two major inputs: 1) the parameter summary extracted from all attachments; 2) the user-filled fields from the approval system form. The model performs comparison and analysis tasks, such as checking whether the "enterprise name in the attachment is consistent with the one filled in the form". Its prompt indicates that the model performs field comparison and outputs an enhanced version of the structured extraction result containing the original extraction data and consistency analysis conclusion. For example, in a specific approval instance, the user may fill in the "enterprise name" as "ABC Technology Co., Ltd." in the form, while the AI extracts the corresponding name as "ABC Technology Co., Ltd." from the uploaded "Three-Party Agreement.pdf". At this time, the consistency comparison and analysis model will identify this subtle difference and generate a prompt message similar to "enterprise name inconsistency: form filled in 'ABC Technology Co., Ltd.', agreement in 'ABC Technology Co., Ltd.' " in the returned enhanced result. This precise difference positioning can directly guide the initiator to focus on and verify possible data entry errors or file version problems.

[0061] After the entire workflow is executed, the final integrated and analyzed structured extraction result is assembled into a complete JSON response data and returned to the approval system.

[0062] In step S103, the structured extraction result returned by the artificial intelligence processing platform is received, and the extraction result is returned to the approval process system to be displayed to the user.

[0063] This step is the information feedback and presentation stage of the intelligent pre-examination process. The core purpose is to accurately and timely return the structured extraction result generated by the AI processing platform after analysis to the context of the original approval process, and visually display it to the process initiator (user). This step completes the key bridge between machine intelligence processing and human cognition and decision-making, and is a necessary link to realize human-machine cooperation and closed loop.

[0064] In some embodiments, step S103 includes two parts: reception and analysis of result data and rendering and visual display of result data, involving the cooperative work of event triggering and interface module, result display and interaction module, and front-end interface rendering technology. Specifically: While the AI platform initiates API calls in step S102, the event triggering and interface module of the approval system will monitor the callback interface or wait for API responses in a synchronous manner. When the AI processing platform completes workflow execution, it will return the encapsulated structured extraction result (usually a JSON object) to the designated endpoint of the approval system through its API access layer.

[0065] The event triggering and interface module successfully receives the response data. The module first performs basic verification, such as HTTP status code and data format legality, to ensure that the received data is complete and has not been tampered with.

[0066] The module parses the received JSON data. Since the data structure returned by the AI platform is for general processing, it may need to be converted or mapped again according to the specific requirements of the current approval form template. For example, the general field name "company_name" is mapped to the predefined label name "Enterprise Name" of the current form interface. This mapping relationship is pre-configured during system integration.

[0067] The parsed and converted data is passed to the specially designed result display and interaction module. This module is responsible for handling the final presentation logic of the data.

[0068] The result display and interaction module will dynamically insert or update a specific visualization area in the approval form page where the user is currently located, according to the context of the current approval process instance and the layout definition of the form. This area is usually located below the original form field, serving as an independent AI intelligent pre-examination result panel.

[0069] The extraction result will be rendered in a form that is easy for the user to understand and check, including but not limited to: Structured Table: The most common form. The extracted key-value pairs are presented in a table format, with each row clearly showing the parameter name (e.g., credit limit) and the AI-extracted value (e.g., 100 million RMB).

[0070] Highlight Contrast Card: Based on the consistency comparison analysis function in the previous embodiment, the display module will perform special rendering. For example, the user's filled-in value and the AI-extracted value are displayed side by side, and different colors or icons are used to visually identify consistency or differences.

[0071] Summary and Tips: The module may generate a short text summary or highlight key differences, guiding the user to focus on them.

[0072] In this visualization area or its adjacent position, the result display and interaction module will render decision buttons for user operation, i.e., the trigger interfaces of confirmation instructions (such as confirmation submission) and modification instructions (such as rollback modification). These buttons are deeply bound with the subsequent process control engine.

[0073] It should be noted that, at the front-end technology level, it is usually implemented through JavaScript (or similar front-end frameworks). After receiving the result data transmitted from the back-end, the front-end script dynamically manipulates the DOM (Document Object Model), creates new Containers, and use of tables ) and apply CSS styles for beautification, and finally seamlessly embed this newly added AI pre-audit result area into the existing form page, realizing the smooth experience of users viewing the results without refreshing.

[0074] In step S104, a confirmation instruction or a modification instruction made by the user based on the extraction result is received.

[0075] The core purpose of this step is to accurately capture the user's (process initiator) explicit follow-up operation intention after reviewing and understanding the AI pre-audit result, and to convert this intention into a system-recognizable standardized instruction. This step embodies the ultimate control of the user in the intelligent auxiliary process, and its output will directly determine the final outcome of the entire approval process.

[0076] In some embodiments, step S104 includes three parts: listening and event binding of the interactive interface, capturing and instructing of user operations, and distribution of decision instructions to the process engine. The implementation of this step relies on the control of the result display and interaction module over the front-end user interface, as well as the close event communication mechanism between the module and the process control engine. Specifically: In step S103, when the result display and interaction module renders the AI extraction result to the front-end page, it will bind specific event listeners to the core decision buttons such as "Confirm Submission" and "Back Modification" (or similar words). These listeners can usually be implemented through front-end technologies such as JavaScript, continuously monitoring user operations on these specific UI elements on the page.

[0077] When the user finishes checking the AI pre-audit result, a decision will be made. If the user believes the result is correct, they will click the "Confirm Submission" button; if they find the result incorrect or wish to adjust, they will click the "Modify Attachment" or "Back Modification" button. At the moment the user clicks the button, the front-end event listener is triggered, capturing this user interaction event.

[0078] The result display and interaction module receives this front-end event and immediately processes it. The core of the processing is to map and encapsulate this specific UI operation into a standardized instruction with clear business semantics that can be processed internally by the system. For example: Clicking "Confirm Submission" → generating "confirm_submission" instruction (confirmation instruction).

[0079] Clicking "Modify Attachment" or "Back Modification" → generating "modify_attachment" instruction (modification instruction).

[0080] The above instruction object usually contains instruction type, triggered process instance ID, current user ID, timestamp, and other context information.

[0081] After the instructions are generated, the result display and interaction module is not responsible for handling the complex process logic behind the instructions. Its core responsibility is to act as an intermediary between the user and the process control engine. The module, through a predefined internal communication mechanism (such as calling the engine's service interface, publishing a system event message, etc.), encapsulates the confirmation instructions or modification instructions, along with the necessary context information, and accurately and timely distributes (transfers) them to the system's process control engine. This distribution action marks that the user's decision has been officially input into the system, and the control is transferred from the display interaction layer to the process logic layer.

[0082] In step S105, in response to the confirmation instruction, the control approval process is transferred to the next approval node; or in response to the modification instruction, the user is allowed to modify the form or the attachment, and the pre-approval method is re-triggered based on the modified form or attachment.

[0083] This step is the final execution and control phase of the intelligent pre-approval process. Its core purpose is to receive and execute the instructions generated by user decisions, thereby intelligently controlling the state transition and direction of the entire approval process. It realizes the automatic management of two key paths: one is normal advancement, and the other is intelligent rollback and cyclic self-checking. This step is the final landing of the human-machine collaboration concept, ensuring that the intelligent pre-approval conclusion can substantially affect and optimize the operation of the approval process.

[0084] In some embodiments, step S105 includes instruction receiving and path branching decision, response to confirmation instruction, forward flow or response to modification instruction, reverse rollback and cyclic triggering. The core performer of this step is the process control engine, which needs to interact deeply with the form module, event triggering and interface module, and the front-end interface. Specifically: The process control engine receives the standard instruction (confirm_submission or modify_attachment) distributed from the result display and interaction module. According to the type of the received instruction, the engine immediately makes a branch decision to determine which pre-set control logic to execute.

[0085] When the instruction is confirm_submission, the engine determines that the user has approved the AI pre-approval result and the form content. The engine updates the database state of the current approval process instance, for example, marking the process from pre-approval or initiator confirmation state as officially pending next node approval.

[0086] The engine calculates and jumps to the next legal approval node according to the pre-defined approval process diagram (including approvers, countersigning rules, etc.). The control front-end interface jumps or refreshes. For the initiator, the submission task is completed; for the next approver, the to-do list will have this to-be-reviewed document.

[0087] When the instruction is modify_attachment, the engine determines that the user needs to revise the data. The engine performs a reverse state control operation. It reverts the state of the current process instance from the pre-approval stage back to the initiation node or the drafting node. This is not a deletion of the process, but a reactivation of the process into an editable state.

[0088] The engine controls the front-end interface to switch the form page from the read-only confirmation view back to the editable modification view. The form data filled in by the user and the list of uploaded attachments become editable, deletable, and up-loadable again.

[0089] In the reactivated form interface, the user can perform modification operations, including modifying form fields and modifying attachments, such as correcting previously filled-in business fields, deleting erroneous attachments, or uploading new and correct attachment files.

[0090] In some embodiments, the pre-approval is re-triggered based on the modified content, including: When the user finishes the modification and clicks the submit button again, the event trigger and interface module will listen to the submit operation again, which is exactly the same as the initial trigger logic in step S101. At this time, the module obtains the latest form data and attachment set modified by the user.

[0091] The engine controls the process to start again from the initiation node, indicating that it will pass through the event node bound with the AI workflow call event again. The AI processing platform will completely re-execute steps S102 to S104 based on the new and modified data, i.e., perform document type judgment and key parameter extraction again, and display the new pre-approval result to the user.

[0092] The user will face a new AI pre-approval result based on the revised data and make a decision to confirm or modify again, which may enter the next control cycle.

[0093] Corresponding to the above method, the application also provides an artificial intelligence-based approval process attachment pre-approval system, which is used to implement the artificial intelligence-based approval process attachment pre-approval method. The system comprises: A form module is used to provide an approval template, receive form data input by a user, and upload attachment files.

[0094] An event trigger and interface module is coupled with the form module and is used to automatically encapsulate business fields and attachment files in the form after listening to a process submission event, and call an interface of an artificial intelligence processing platform to send data.

[0095] A result display and interaction module is used to receive and display structured extraction results returned by the artificial intelligence processing platform, and provide an interactive interface for confirmation and modification.

[0096] A flow control engine is configured to control the flow direction of the approval process according to the instructions made by the user through the interactive interface.

[0097] An artificial intelligence processing platform is communicatively connected to the event trigger and interface module, and is configured to execute a corresponding preconfigured workflow according to the received business field to complete the attachment document type judgment and key parameter extraction.

[0098] Corresponding to the above method, the application also provides an electronic device, which comprises a computer device, the computer device comprising a processor and a memory, the memory storing computer instructions, and the processor being configured to execute the computer instructions stored in the memory, so that the electronic device implements the steps of the method as described above.

[0099] The embodiment of the application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the above method. The computer readable storage medium can be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the art.

[0100] Those of ordinary skill in the art should understand that the example components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether the implementation is in hardware or software depends on the specific application and design constraints imposed on the overall system. Skilled artisans can employ different elements in different combinations to implement the described functions, without departing from the scope of the application. When implemented in hardware, the elements of the application can be implemented in, for example, electronic circuitry, an application-specific integrated circuit (ASIC), suitable firmware, a plug-in, a functional card, or the like. When implemented in software, the elements of the application are the program or code segments to perform the necessary tasks. The program or code segments can be stored in a machine-readable medium, or transmitted by a carrier wave as data signals over a transmission medium or communication link.

[0101] It should be understood that the application is not limited to the particular configurations and processes described above and shown in the drawings. For the sake of brevity, conventional methods and systems will not be described in detail. In the above embodiments, several specific steps are described and shown as examples. However, the method processes of the application are not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order of the steps, after understanding the spirit of the application.

[0102] In the present application, features described and / or illustrated with respect to one embodiment can be used in the same way or in a similar way in one or more other embodiments, and / or in combination with or in place of features of other embodiments.

[0103] The above description is merely illustrative of the application, and is not intended to limit the application. The embodiments of the application can be variously changed and modified without departing from the spirit and scope of the application. Therefore, the scope of the application should be defined by the appended claims rather than by the detailed description, and all modifications or changes that come within the scope of the application are intended to be protected by the claims. ), list (

Claims

1. A method for pre-reviewing attachments in an approval process based on artificial intelligence, characterized in that, The method includes: In response to a user's submission operation in the approval process, the system obtains at least one business field from the form corresponding to the approval process and at least one attachment uploaded by the user, and sends the business field and the attachment to the artificial intelligence processing platform. Based on the received business fields, the corresponding pre-configured workflow is determined and executed in the artificial intelligence processing platform; wherein, executing the pre-configured workflow includes: traversing the attachments, determining the document type of each attachment based on the business fields and attachment content, and performing structured extraction of key parameters for document types that meet preset conditions; Receive the structured extraction results returned by the artificial intelligence processing platform, and return the extraction results to the approval process system to display to the user; Receive confirmation or modification instructions from the user based on the extraction results; In response to the confirmation instruction, the approval process is controlled to proceed to the next approval node; or in response to the modification instruction, the user is allowed to modify the form or the attachment, and the pre-approval method is retried based on the modified form or attachment.

2. The artificial intelligence-based pre-review method for approval process attachments according to claim 1, characterized in that, The document type of each attachment is determined based on the aforementioned business fields and attachment content, including: Use the aforementioned business fields as routing conditions to select a pre-configured document judgment model; The attachment content is input into the document judgment model to obtain the attachment type identifier; wherein, the document judgment model is a large language model, and document type judgment is achieved by inputting prompt words containing the business field into the large language model and constraining its output format.

3. The artificial intelligence-based pre-review method for approval process attachments according to claim 2, characterized in that, For document types that meet preset conditions, perform structured extraction of key parameters, including: Using the type identifier as a selection criterion, the parameter extraction model corresponding to the document type is invoked. The attachment content is input into the parameter extraction model to extract predefined fields and output them in a structured data format. The parameter extraction model is a large language model. Structured parameter extraction is achieved by inputting prompt words containing document content into the large language model and constraining its output format.

4. The artificial intelligence-based pre-review method for approval process attachments according to claim 1, characterized in that, The pre-configured workflow implements branching logic through selector nodes. The selector nodes route the processing flow to one of multiple different business branches based on the value of the business field.

5. The artificial intelligence-based pre-review method for approval process attachments according to claim 1, characterized in that, After generating the structured extraction result, the process also includes: The parameters contained in the structured extraction results are compared and analyzed for consistency with the field values ​​filled in by the user in the form; The comparison analysis results are returned as part of the extracted results and displayed to the user.

6. The artificial intelligence-based pre-review method for approval process attachments according to claim 1, characterized in that, Receiving a modification instruction from the user based on the extraction results, allowing the user to modify the form or the attachment, and re-triggering the pre-screening method, including: In response to the modification instruction, the approval process is controlled to revert to the initiating node; Receive the user's modification and resubmission operations on the form or attachment at the initiating node; In response to the resubmission operation, the pre-review method is re-executed.

7. The artificial intelligence-based pre-review method for approval process attachments according to claim 1, characterized in that, The steps implemented in the artificial intelligence processing platform by the method are defined by a visual workflow consisting of multiple nodes, including at least a logical node for branch selection, a loop node for processing multiple attachments, and a model node for calling artificial intelligence model services.

8. A pre-approval system for approval process attachments based on artificial intelligence, characterized in that, The system is used to implement the method according to any one of claims 1 to 7, the system comprising: The form module is used to provide approval templates, receive form data entered by users, and upload attachment files; The event triggering and interface module, coupled with the form module, is used to automatically encapsulate the business fields in the form and the attachment file after listening to the process submission event, and call the interface of the artificial intelligence processing platform to send data. The results display and interaction module is used to receive and display the structured extraction results returned by the artificial intelligence processing platform, and provide an interactive interface for confirmation and modification. The process control engine is used to control the flow direction of the approval process based on the instructions made by the user through the interactive interface; The artificial intelligence processing platform is communicatively connected to the event triggering and interface module, and is used to execute the corresponding pre-configured workflow according to the received business fields to complete the document type judgment and key parameter extraction of the attachments, and return the results.

9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1 to 7.

Images