Business document auditing method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to the technical field of artificial intelligence, discloses a business document auditing method and device, computer equipment, a readable storage medium and a program product, and can improve the business document auditing efficiency. The method comprises the following steps: determining an auditing emergency degree value of a business document according to document content and business information of the business document; if the auditing emergency degree value is greater than or equal to a preset threshold value, extracting target document content based on the document content by the document processing model, and generating a voice auditing prompt according to the target document content and the verbal skill template; calling a terminal of the document audit object, playing a voice audit prompt through the terminal when the call is successful, obtaining an audit result of the business document according to audit voice replied by the document audit object, and sending a confirmation request for the audit result to the terminal; and in response to a confirmation instruction returned by the terminal for the confirmation request, determining that the business document has been audited.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a business document review method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Technology

[0002] With the development of computer technology, users' demand for financial services is increasing, especially for efficient and convenient online services. Online business document verification is one such area that has attracted significant attention.

[0003] Some business documents can be intelligently identified and processed using technologies such as Robotic Process Automation (RPA) and Optical Character Recognition (OCR), thereby improving review efficiency. However, for documents requiring manual review, the document generation and review processes are independent. The reviewer can only review the document after the document creator has completed it, and the reviewer needs to log into the relevant system to locate the document before performing the review. Therefore, there is a significant time lag between the document creator's completion of document creation and the reviewer's actual review, indicating that business document review efficiency needs improvement. Summary of the Invention

[0004] Therefore, it is necessary to provide a business document review method, apparatus, computer equipment, computer-readable storage medium, and computer program product to address the aforementioned technical problems.

[0005] Firstly, this application provides a method for reviewing business documents, including:

[0006] Obtain the business documents to be reviewed, and determine the urgency value of the review of the business documents based on the document content and business information of the business documents;

[0007] If the urgency level of the audit is less than a preset threshold, the business document is submitted to the non-urgent document pool to wait. When other business documents with an urgency level greater than or equal to the preset threshold are processed, the trained document processing model will process the business document in the non-urgent document pool.

[0008] If the urgency level of the review is greater than or equal to the preset threshold, the document processing model extracts the target document content based on the document content, and generates a voice review prompt for the business document according to the target document content and the script template.

[0009] The terminal that calls the document review object plays the voice review prompt through the terminal when the call is successful. The review result of the business document is obtained according to the review voice reply from the document review object, and a confirmation request for the review result is sent to the terminal.

[0010] In response to the terminal returning a confirmation instruction for the confirmation request, it is determined that the business document has been reviewed.

[0011] In one embodiment, generating a voice verification prompt for the business document based on the target document content and the script template includes: the document processing model determining a first business feature corresponding to the business document based on the document content; determining a target script template from the multiple candidate script templates based on the matching result of the first business feature and a second business feature of multiple candidate script templates; each candidate script template including a script that highlights the document content associated with the second business feature; and generating a voice verification prompt for the business document based on the target document content and the script in the target script template.

[0012] In one embodiment, the target document content includes document content under multiple fields; generating a voice verification prompt for the business document based on the target document content and the script in the target script template includes: inputting the document content under the multiple fields into a trained semantic analysis model to obtain a risk weight for the document content under each field output by the semantic analysis model; determining the playback priority of the document content under the multiple fields based on each risk weight; and determining the voice parameters of the document content under the multiple fields based on each risk weight; the playback priority increases with the increase of the risk weight, and the voice parameters of the document content with different risk weights are different; generating a voice verification prompt for the business document based on the playback priority, the voice parameters, and the script in the target script template.

[0013] In one embodiment, the step of extracting target document content based on the document content by the document processing model includes: obtaining the document content under a specified field from the business document by the document processing model; inputting the document content under the specified field into an ambiguity recognition model, and when the ambiguity recognition model indicates that the document content under the specified field is ambiguous during voice broadcast, obtaining context information of the document content under the specified field based on the document content of the business document and the business information; performing disambiguation processing on the document content under the specified field based on the context information, and obtaining the target document content based on the disambiguation processing result.

[0014] In one embodiment, after the step of playing the voice verification prompt through the terminal upon successful call, the method further includes: receiving the verification voice segments of the document verification object returned by the terminal, and determining the first segment identifier corresponding to the received verification voice segments; if communication with the terminal is interrupted, obtaining the segment identifier set returned by the terminal upon communication resumption; the segment identifier set includes the second segment identifiers of each verification voice segment cached locally by the terminal; determining a third segment identifier based on the first segment identifier and the segment identifier set and sending it to the terminal; the third segment identifier is a second segment identifier in the segment identifier set that does not match the first segment identifier; and obtaining the verification voice response from the document verification object based on the fusion result of the verification voice segments corresponding to each of the third segment identifiers returned by the terminal and the received verification voice segments.

[0015] In one embodiment, determining the third fragment identifier and sending it to the terminal includes: determining a plurality of third fragment identifiers and obtaining the sending priority of each of the plurality of third fragment identifiers; the sending priority decreases as the order of the third fragment identifiers decreases; determining the batch processing quantity based on the real-time monitored network throughput, and sending fragment acquisition requests carrying the third fragment identifiers to the terminal in batches according to the batch processing quantity and the sending priority; wherein the number of fragment acquisition requests sent in each batch does not exceed the batch processing quantity.

[0016] Secondly, this application also provides a business document verification device, comprising:

[0017] The document acquisition module is used to acquire business documents to be reviewed, and determine the urgency value of the review of the business documents based on the document content and business information of the business documents.

[0018] The caching module is used to submit the business document to a non-urgent document pool if the urgency value of the audit is less than a preset threshold, so that when other business documents with an urgency value greater than or equal to the preset threshold are processed, the trained document processing model will process the business document in the non-urgent document pool.

[0019] The prompt generation module is used to extract the target document content based on the document content by the document processing model if the urgency value of the review is greater than or equal to the preset threshold, and generate a voice review prompt for the business document based on the target document content and the script template.

[0020] The calling module is used to call the terminal of the document review object. When the call is successful, the voice review prompt is played through the terminal. The review result of the business document is obtained according to the review voice reply from the document review object, and a confirmation request for the review result is sent to the terminal.

[0021] The status switching module is used to respond to the terminal's confirmation request by returning a confirmation instruction to determine that the business document has been reviewed.

[0022] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the business document review method as described in any of the preceding claims.

[0023] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the business document review method as described in any of the preceding claims.

[0024] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the business document review method as described in any of the preceding claims.

[0025] The aforementioned business document review method, apparatus, computer equipment, computer-readable storage medium, and computer program product can acquire business documents to be reviewed, determine the review urgency value of the business documents based on their content and business information, and submit the business documents to a non-urgent document pool if the urgency value is less than a preset threshold. Once other business documents with urgency values ​​greater than or equal to the preset threshold have been processed, the trained document processing model will process the business documents in the non-urgent document pool. If the urgency value is greater than or equal to the preset threshold, the document processing model will extract the target document content based on the document content and generate a voice review prompt for the business document based on the target document content and a dialogue template. The terminal that calls the document review target will play the voice review prompt upon successful call, obtain the review result of the business document based on the review voice response from the document review target, send a confirmation request for the review result to the terminal, and then respond to the terminal by returning a confirmation instruction to confirm that the business document has been reviewed. In this embodiment, on the one hand, for business documents with an urgency value greater than or equal to a preset threshold, the content of the target document is extracted, and a voice review prompt is generated based on the target document content and a script template. This prompt is then broadcast instantly to the document review target via the terminal, enabling rapid and immediate review results for urgent business documents. This significantly shortens the business document review cycle, reduces the time lag between document creation and review, and achieves real-time processing. On the other hand, by classifying business documents according to their urgency value—that is, caching non-urgent business documents in a non-urgent document pool and processing urgent business documents only after the outbound review process is completed—the orderly scheduling of limited review resources can be achieved, ensuring that urgent business documents are processed first. Thus, this embodiment significantly improves the efficiency of business document review, as well as the efficiency of resource flow and business operations in the upstream and downstream of the business, thereby enhancing the user experience and the quality and experience of document review services.

[0026] Furthermore, by providing multiple candidate script templates in advance, and setting scripts in each candidate script template to highlight the document content associated with the second business feature, the efficiency of document reviewers in capturing key risk information in business documents can be improved.

[0027] Furthermore, by determining the broadcast priority of document content under multiple fields and the voice parameters of document content under multiple fields based on various risk weights, the prompting effect of key information in the target document content can be enhanced, thereby improving the accuracy and efficiency of document review.

[0028] Furthermore, by identifying whether there is ambiguity in the document content under a specified field during voice broadcast, and performing disambiguation processing when ambiguity exists, it is possible to reduce the possible misunderstandings of the document review object when playing the target document content, and avoid obtaining incorrect review results due to ambiguity.

[0029] Furthermore, by matching the first fragment identifier with the fragment identifier aggregator to determine the third fragment identifier and sending it to the terminal, data transmission efficiency can be improved while ensuring data integrity.

[0030] Furthermore, by sending fragment retrieval requests carrying the third fragment identifier to the terminal in batches according to the batch processing quantity and sending priority, the loss rate of retransmitted audit voice fragments can be effectively reduced. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is an application environment diagram of a business document review method in one embodiment;

[0033] Figure 2 This is a flowchart illustrating a business document review method in one embodiment;

[0034] Figure 3 A flowchart illustrating the steps for generating a voice verification tool in one embodiment;

[0035] Figure 4 A flowchart illustrating another step in generating a voice verification tool in one embodiment;

[0036] Figure 5 This is a schematic diagram illustrating the framework of a business document review method in one embodiment;

[0037] Figure 6 This is a flowchart illustrating another business document review method in one embodiment;

[0038] Figure 7 This is a structural block diagram of a business document verification device in one embodiment;

[0039] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. It should be noted that existing industry solutions such as software, components, and models may be mentioned in the embodiments of this application. These should be considered exemplary and are intended only to illustrate the feasibility of implementing the technical solutions of this application, but do not imply that the applicant has already used or necessarily used such solutions.

[0041] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant regulations. The acquisition, storage, use and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations.

[0042] The terms "first," "second," etc., used in this application may be used to describe various objects, but these objects are not limited by these terms. These terms are used only to distinguish the first object from the second object. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more.

[0043] With the development of computer technology and in the context of the digital age, users (such as corporate users and individual users) have an increasing demand for financial services, especially efficient and convenient online services. Online business document review is a crucial and closely watched aspect of this process. Some business documents can be intelligently identified and streamlined using technologies such as RPA robots and OCR, thereby improving review efficiency.

[0044] For documents requiring manual review, the document generation and review processes are independent. The reviewer (such as a checker or supervisor) can only review the document after the document creator (e.g., a salesperson or other document creator) has produced it. Since the document creator and reviewer are often not the same person, the document creator must wait for the reviewer to log into the relevant system and locate the document before performing the review. Therefore, there is a significant time lag between the document creator completing its creation and the reviewer actually reviewing it, making immediate review impossible.

[0045] Furthermore, if the review process for business documents is too lengthy after creation, it may delay business decisions. For example, when making urgent resource requests, the time lag between document creation and review could hinder the progress of critical business operations and impact the processing efficiency of downstream businesses.

[0046] It is evident that the efficiency of the relevant technologies in reviewing business documents still needs to be improved.

[0047] This application provides a business document review method, apparatus, computer equipment, computer-readable storage medium, and computer program product, which can effectively improve the review efficiency of business documents.

[0048] In one embodiment, the business document review method provided in this application can be applied to, for example... Figure 1 The application environment shown includes a server and terminals for document review. The terminals can communicate with the server via a network. The server may have a corresponding data storage system that stores various business documents to be reviewed. The data storage system may be integrated on server 104 or located in the cloud or on other network servers.

[0049] The terminals can be, but are not limited to, various personal computers, laptops, smartphones, tablets, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart TVs, smart in-vehicle devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted displays, etc. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, etc. Servers can be independent physical servers, server clusters or distributed systems composed of multiple physical servers, or cloud servers providing cloud computing services.

[0050] In one exemplary embodiment, such as Figure 2 As shown, a business document approval method is provided, which is applied to... Figure 1 The following steps, from step 202 to step 206, will be used as an example to illustrate the process.

[0051] Step S201: Obtain the business documents to be reviewed, and determine the urgency level of the review of the business documents based on the document content and business information.

[0052] Among them, the business documents to be reviewed can refer to documents containing relevant information about business processing and awaiting confirmation of their validity or compliance through the review process, such as expense reports, loan application forms, supplier settlement forms, or one or more other types.

[0053] In some possible application scenarios, business documents can be financial business documents, which are one of the links in the online resource exchange through the financial system. The process of creating online business documents is also called document creation, which refers to the process by which relevant business personnel enter business-related information and generate electronic documents on the online financial platform based on actual financial transactions, such as payments, receipt confirmations, and transfers.

[0054] The person creating business documents can be called a document creator or the recipient of the document, such as the person in charge or the business personnel. During the document creation process, the creator can fill in detailed information about various key aspects of the document. In one example, this might include, but is not limited to, the quantity of resources, the recipient's account and name, the purpose of the resources, and the transaction date. Furthermore, depending on the business type and pre-defined management requirements, relevant supporting materials, such as contracts and invoices, can be uploaded as evidence and proof of the transaction. Online document creation, as a front-end step in the business process, lays the foundation for subsequent review, approval, and final transaction execution. Standardized and accurate document creation ensures the authenticity, legality, and accuracy of business transactions, while simultaneously improving the efficiency and informatization level of resource management.

[0055] The content of a document can refer to the information directly recorded in a business document. In some exemplary scenarios, it can include one or more pieces of information originally recorded in the business document.

[0056] Business information can be various information associated with a business document. In an exemplary embodiment, business information may include background information associated with the business document, which may include, but is not limited to, one or more of the following: the business type corresponding to the business document, the document submission deadline, the business document review deadline, the priority of associated business, etc.

[0057] The urgency level value is a numerical value used to quantify the urgency of reviewing business documents. The higher the value, the higher the priority of the document's review.

[0058] In practice, the server can retrieve business documents awaiting review. In some exemplary embodiments, documents submitted by the document creator via a terminal can be directly identified as business documents awaiting review. Alternatively, business documents in the pending review state can be read from the data storage system.

[0059] For the obtained business documents, the urgency of the review can be analyzed based on the document content and related business information to obtain the review urgency value. Then, based on the comparison between the review urgency value and the preset threshold, it can be determined which review path the business document should be processed through.

[0060] In one embodiment, when determining the urgency value of the audit, the document content and associated business information of the business document can be input into a trained urgency recognition model. The model can determine one or more information such as document type and document importance based on the document content, and one or more information such as business priority, audit time limit, and importance of the business document to downstream business links based on the business information. Then, the audit urgency value of the business document can be determined based on the analysis results of both.

[0061] In step S202, if the urgency value is less than a preset threshold, the business document is submitted to the non-urgent document pool to wait. When other business documents with an urgency value greater than or equal to the preset threshold are processed, the trained document processing model will process the business document in the non-urgent document pool.

[0062] The preset threshold is a critical value used to determine the urgency of business documents and can be set according to the business scenario. The non-urgent document pool is a virtual queue or storage module used to store non-urgent business documents. In one example, documents can be managed using a first-in-last-out rule or a secondary sorting rule based on the urgency value. After the urgent documents are processed, the document processing model is called to process them.

[0063] In practical applications, if the urgency level of a business document is determined to be less than a preset threshold, the business document can be submitted to the non-urgent document pool for caching and waiting for processing. In other words, business documents with an urgency level less than the preset threshold can be identified as non-urgent business documents, and such documents can be delayed to avoid non-urgent documents occupying the review resources of urgent documents.

[0064] After submitting non-urgent business documents to the non-urgent document pool, other business documents with an urgency value greater than or equal to a preset threshold can be checked at preset time intervals to see if they have been processed. For example, business documents with an urgency value greater than or equal to a preset threshold can be stored in the emergency document pool, and the status of each business document in the emergency document pool can be checked at preset time intervals to see if they have all been reviewed.

[0065] If not, the trained document processing model will continue to prioritize processing other business documents with an urgency value greater than or equal to a preset threshold. If so, the document processing model can be invoked to begin processing business documents in the non-urgent document pool. For example, the non-urgent document pool can be sorted by submission time, and the document processing model can be automatically invoked based on the sorting results to batch process the documents in the non-urgent document pool. The document processing model's processing methods for non-urgent documents can include automatically generating preliminary review opinions and pushing them to the terminal of the document review recipient for confirmation, or distributing business documents to the terminal of the document review recipient and associating them with prompt information to promptly remind the document review recipient that they can review the distributed documents later.

[0066] Step S203: If the urgency level of the audit is greater than or equal to the preset threshold, the document processing model extracts the target document content based on the document content, and generates a voice audit prompt for the business document based on the target document content and the script template.

[0067] The target document content can be key information directly related to the business document review result, extracted by the document processing model based on the document content. In some examples, the target document content can be extracted based on one or more factors such as business configuration, experience information, and semantic understanding results. The script template can be understood as a pre-defined standardized text framework used to generate voice review prompts. The script template can contain fixed introductory phrases and replaceable variable fields, such as "Hello, there is a document submitted by [Applicant Name] for [Business Type], the key information of which is [Target Document Content]. Please review it." The voice review prompt is a prompt message displayed to the document review recipient via voice broadcast. The voice review prompt can contain information related to the document review result decision, such as the target document content.

[0068] In some related technologies, after a document clerk completes the document creation process, the document reviewer needs to actively discover the pending document before they can perform the review operation. Furthermore, the reviewer needs to log in to the relevant system to execute the review. This process is inconvenient. On the one hand, the reviewer may not notice the pending document for some time; on the other hand, they need to perform a series of login operations on a device with system login capabilities before they can begin the review. For documents that urgently need review, this method makes it difficult to trigger the review quickly and promptly, easily leading to a poor user experience.

[0069] In this regard, if the urgency value of a business document is determined to be greater than or equal to a preset threshold, it can be determined that the business document needs to be processed promptly. This embodiment can trigger an immediate review process for the business document, allowing the document review target to undergo review as quickly as possible. In some embodiments, the document clerk can choose the document review method. The server can determine whether to use a traditional manual review method or an artificial intelligence (AI) outbound call review method based on the selected document review method and the urgency value. For example, when the document clerk selects to use AI outbound call review and the urgency value is greater than or equal to a preset threshold, the server can make an immediate outbound call to the document review target through steps S203 to S205 to obtain the immediate review result.

[0070] Specifically, for business documents with an urgency level greater than or equal to a preset threshold, the server can call the document processing model. The document processing model extracts the target document content based on the document content and generates a voice review prompt for the business document based on the target document content and a pre-selected script template. This voice review prompt is a personalized voice prompt generated for various business documents, which can inform the document review recipient of key information in the business document, so that the document review recipient can quickly determine the preliminary review result of the business document based on the voice review prompt.

[0071] In this embodiment, by generating audit prompt voice from the document processing model, artificial intelligence technology can be used to ensure that the information conveyed to the document audit recipient matches the document content, ensuring the accuracy and consistency of information during the information transmission process and reducing the risk of human error.

[0072] Step S204: Call the terminal of the document review object, play a voice review prompt on the terminal when the call is successful, obtain the review result of the business document according to the review voice reply of the document review object, and send a confirmation request for the review result to the terminal.

[0073] The document reviewer, also known as the examiner, refers to the personnel who can review business documents; they are the staff responsible for reviewing business documents. In some examples, the document reviewer's terminal can be a mobile phone, landline, or dedicated review device, which can support voice calls, voice playback, and instruction feedback functions.

[0074] In this step, after receiving the voice review prompt, the terminal of the document review target can be called. In some exemplary embodiments, the document preparer can select the calling method, or the matching calling method can be determined based on the review urgency value. For example, the calling method for the document review target's terminal can include, but is not limited to, any one or more of the following AI outbound calling methods: instant outbound calling, scheduled outbound calling, and recurring outbound calling. Instant outbound calling is made immediately after the business document is generated or after the business document is determined to be an urgent document awaiting review (i.e., the review urgency value is greater than or equal to a preset threshold) to the terminal of the document review target. Scheduled outbound calling is made at a time specified by the document preparer after the business document is generated or after the business document is determined to be an urgent document awaiting review. Recurring outbound calling refers to continuing to make outbound calls after the first unsuccessful call until the terminal of the document review target is successfully reached.

[0075] Furthermore, upon successful call, the terminal can play a voice review prompt to the document review recipient and obtain the review voice response from the recipient. This review voice can be understood as the voice response from the document review recipient after listening to the prompt; it may include the review result of the document review recipient on the business document, or other instructions. After obtaining the review voice, the server can retrieve the review result of the business document based on the recipient's response and send a confirmation request to the terminal, allowing the document review recipient to reconfirm their review result.

[0076] Step S205: In response to the terminal's confirmation request, a confirmation instruction is returned to confirm that the business document has been reviewed.

[0077] In some examples, the terminal can display confirmation requests in the form of voice or text. For example, it can ask the document reviewer whether to confirm the review result of the business document as passed or failed through voice broadcast. It can also send confirmation requests through SMS or other text forms. In this way, the accuracy, reliability and smoothness of the review results obtained by the server from the review voice analysis can be further improved, and subsequent business can be kept smooth. It can also avoid the impact on subsequent processing due to errors in the review voice analysis and the obtaining of incorrect review results.

[0078] Furthermore, the document reviewer can respond to the confirmation request in an appropriate manner. If the server receives a confirmation instruction from the terminal in response to the confirmation request, that is, confirms the review result obtained by the server, the server can determine that the business document has been reviewed by the document reviewer based on the confirmation instruction returned by the terminal, and can change the review status of the business document from pending review to reviewed.

[0079] In some embodiments, if the review result is "approved," the document status can be automatically updated to "approved," and subsequent business processes can proceed. If the review result is "disapproved," the document status can be automatically updated to "approved," and the document preparer can be notified to make corresponding modifications or additions.

[0080] In the above-mentioned business document review method, business documents to be reviewed can be obtained. Based on the document content and business information, the urgency value of the review of the business document can be determined. If the urgency value is less than a preset threshold, the business document is submitted to a non-urgent document pool to wait. When other business documents with an urgency value greater than or equal to the preset threshold are processed, the trained document processing model processes the business documents in the non-urgent document pool. If the urgency value is greater than or equal to the preset threshold, the document processing model extracts the target document content based on the document content and generates a voice review prompt for the business document based on the target document content and a dialogue template. The terminal that calls the document review target plays the voice review prompt through the terminal when the call is successful. The review result of the business document is obtained based on the review voice reply from the document review target, and a confirmation request for the review result is sent to the terminal. Then, in response to the confirmation request from the terminal, a confirmation instruction is returned to confirm that the business document has been reviewed. In this embodiment, on the one hand, for business documents with an urgency value greater than or equal to a preset threshold, the content of the target document is extracted, and a voice review prompt is generated based on the target document content and a script template. This prompt is then broadcast instantly to the document review target via the terminal, enabling rapid and immediate review results for urgent business documents. This significantly shortens the business document review cycle, reduces the time lag between document creation and review, and achieves real-time processing. On the other hand, by classifying business documents according to their urgency value—that is, caching non-urgent business documents in a non-urgent document pool and processing urgent business documents only after the outbound review process is completed—the orderly scheduling of limited review resources can be achieved, ensuring that urgent business documents are processed first. Thus, this embodiment significantly improves the efficiency of business document review, as well as the efficiency of resource flow and business operations in the upstream and downstream of the business, thereby enhancing the user experience and the quality and experience of document review services.

[0081] Furthermore, in related technologies, a large backlog of documents awaiting review can lead to significant business delays. This not only increases the pressure on subsequent processing but may also cause delays in important business operations. For example, in some scenarios, during financial settlement, a large number of unreviewed online banking documents may affect the preparation of financial documents and the processing of fund settlements. To address this, the document review method provided in this embodiment can promptly and specifically remind document reviewers to review urgent documents based on their urgency, helping to reduce the workload of document reviewers and allowing them to focus more on complex review tasks and decision-making.

[0082] In one embodiment, such as Figure 3 As shown, in step S203, based on the content of the target document and the script template, a voice verification prompt for the business document is generated, which may include the following steps:

[0083] Step S301: The document processing model determines the first business feature corresponding to the business document based on the document content.

[0084] When generating voice review prompts, one possible implementation is to provide a standardized script template. Based on this template and the extracted target document content, the system reminds the document reviewer to review business documents in various scenarios. However, in this approach, the pre-set script template often has poor compatibility with the target document content, resulting in voice review prompts that lack contextualization and readability, making it difficult to effectively highlight the content that the document reviewer needs to focus on. For example, standardized script templates often use fixed phrases, such as "Please review [Field 1], [Field 2], [Field 3]".

[0085] In this embodiment, the document processing model can extract business features based on the document content of the business document to determine the business features corresponding to the business document. For ease of distinction, the business features corresponding to the business document are referred to as the first business features. These business features can be a set of features characterizing the main attributes of the document. In some examples, the business features can reflect one or more of the following: document type, the quantity level of the resources involved in the business document, and the risk level / abnormality level associated with the business document.

[0086] In some embodiments, business documents awaiting review can be input into a document processing model. The model can perform structured parsing of the document content, extract key fields such as amount, type, and related information, and then, based on preset feature mapping rules, transform the parsed fields into first business features that reflect the main business attributes of the document. For example, based on the feature extraction results of a certain business document A, one or more of the following first business features can be obtained: "regular amount transfer", "high amount transfer", "unknown account transfer", and "abnormal time transfer".

[0087] Step S302: Based on the matching results of the first business feature and the second business feature of multiple candidate script templates, determine the target script template from the multiple candidate script templates; each candidate script template includes a script that highlights the document content associated with the second business feature.

[0088] In addition, multiple candidate script templates can be pre-stored in the data storage system. Each candidate script template can be pre-associated with corresponding business features. For ease of differentiation, the business features associated with the candidate script templates are referred to as the second business features. For each candidate script template, the template can include script information that highlights the document content associated with the second business feature.

[0089] For example, a candidate script template associated with the second business characteristic "high-amount transfer" could include the script information "Please note that the current business document is a high-amount transfer. Please check whether the transaction amount [amount field] matches the approved limit." Similarly, a candidate script template associated with the second business characteristic "abnormal time transfer" could include the script content "The current transaction time is [time field] (abnormal time period). Please verify the initiator's transaction purpose [purpose field]."

[0090] Furthermore, after obtaining the first business feature of the business document, it is possible to match the first business feature with the second business features of multiple candidate script templates, and determine the target script template from the multiple candidate script templates based on the matching results. For example, the number of feature matches between the first business feature and the second business feature of each candidate script template can be determined, and the candidate script template with the highest number of feature matches can be used as the target script template.

[0091] Step S303: Generate voice verification prompts for business documents based on the content of the target document and the script in the target script template.

[0092] Once the target script template is obtained, a voice verification prompt can be generated based on the content of the target document and the script in the target script template.

[0093] In this embodiment, by providing multiple candidate script templates in advance, and setting scripts in each candidate script template to highlight the document content associated with the second business feature, and then determining the target script template based on the matching results of the first business feature and the second business feature of the multiple candidate script templates, the generated voice audit prompts can be matched with the business scenario of the business document, thereby improving the accuracy and logic of the voice audit prompts and increasing the efficiency of the document audit object in capturing key risk information in the business document.

[0094] In one embodiment, the target document content includes document content under multiple fields. The document content under multiple fields can be understood as different dimensions of text data in the business document used to represent key business information. For example, such as "for the development of XX project" under the "Purpose of Loan" field, "repayment source" under the "Repayment from Project" field, and "no collateral or guarantee" under the "Guarantee Method" field.

[0095] like Figure 4 As shown, in step S03, based on the content of the target document and the script in the target script template, a voice verification prompt for the business document is generated, which may include the following steps:

[0096] Step S401: Input the document content under multiple fields into the trained semantic analysis model to obtain the risk weight of the document content under each field output by the semantic analysis model.

[0097] The semantic analysis model can be a natural language processing model trained on a large number of sample business documents (including normal business documents and risk business documents). This model can output a quantitative value of the risk level corresponding to the text, i.e., risk weight, by learning the mapping relationship between text semantics and risk labels.

[0098] In this step, after obtaining the document content under multiple fields, the document content under multiple fields can be input into the semantic analysis model, and the risk weight of the document content under each field can be determined by the model.

[0099] Step S402: Determine the broadcast priority of document content under multiple fields according to each risk weight, and determine the voice parameters of document content under multiple fields according to each risk weight; the broadcast priority increases with the increase of the risk weight, and the voice parameters of document content with different risk weights are different.

[0100] In practical applications, as the audio playback progresses, the listener's attention to the audio content may decrease. To address this, this embodiment addresses the following: First, the playback priority of document content across multiple fields can be determined based on various risk weights, prioritizing fields with high risk weights and delaying those with low risk weights. Second, the audio parameters of the document content across multiple fields can be determined based on various risk weights. This means the audio parameters can be dynamically adjusted according to each risk weight. These audio parameters can include one or more of the following: speech rate, volume, and tone. For example, high-risk fields can be set with a 15% slower speech rate, a 10% higher volume, and a 5Hz higher tone, while low-risk fields use standard audio parameters, enhancing the prompting effect of key information through auditory differences.

[0101] Step S403: Generate voice review prompts for business documents based on the broadcast priority, voice parameters, and the script in the target script template.

[0102] Furthermore, voice review prompts can be generated based on broadcast priority, voice parameters, and the script in the target script template.

[0103] In this embodiment, by determining the broadcast priority of document content under multiple fields and the voice parameters of document content under multiple fields according to each risk weight, the prompting effect of key information in the target document content can be enhanced, auditory fatigue of auditors during the voice audit process can be reduced, and the audit accuracy and efficiency of the document audit object can be improved.

[0104] In one embodiment, step S203, in which the document processing model extracts the target document content based on the document content, may include the following steps:

[0105] Step S2031: The document processing model retrieves the document content under the specified field from the business document.

[0106] In practice, the document processing model can first obtain the document content under the specified field from the business document. For example, the document processing model can perform optical character recognition on the business document, determine the document type based on the optical character recognition, and use the fields involved in the document type as the specified fields. The document content under the specified fields can then be determined based on the optical character recognition.

[0107] Step S2032: Input the document content under the specified field into the ambiguity recognition model, and if the ambiguity recognition model indicates that the document content under the specified field is ambiguous during voice broadcast, obtain the context information of the document content under the specified field based on the document content and business information of the business document.

[0108] The ambiguity recognition model can be a binary classification model, trained using field content samples labeled "ambiguous" and "unambiguous." It can determine whether the field content has multiple pronunciations, meanings, or is easily confused during voice playback. Ambiguity during voice playback means that listeners may have multiple interpretations of the field content when it is played aloud.

[0109] After obtaining the document content under a specified field, this content can be input into a trained ambiguity recognition model. The model will then identify whether the document content under the specified field is ambiguous during voice playback. For example, the ambiguity recognition model can output the probability of ambiguity in the document content under the specified field during voice playback. If the probability of ambiguity is greater than a preset threshold, then it can be determined that the document content under the specified field is ambiguous.

[0110] If no ambiguity is identified, the document content under the specified field can be directly used as the target document content. If ambiguity is identified, the contextual information of the document content under the specified field can be obtained first based on the document content and business information of the business document. The contextual information of the document content refers to other information related to the document content under the specified field that can help understand the document content under the specified field. It can come from other field content of the business document or from the business information. The contextual information can eliminate the ambiguity of the document content under the specified field during voice broadcast.

[0111] Step S2033: Perform disambiguation processing on the document content under the specified field according to the context information, and obtain the target document content based on the disambiguation processing result.

[0112] In practice, after obtaining the context information, ambiguity in document content under a specified field can be eliminated based on the collected context information using methods such as multi-dimensional information verification, rule matching, and historical data comparison. In one embodiment, the document content of a specified field can be matched with the context information to filter out candidate meanings that conform to the rules; then, the candidate meanings can be verified by comparing them with the content of other fields in the same document to eliminate contradictions. In addition, the accurate meaning of the specified field content can be determined by combining historical data of similar documents, thereby completing the disambiguation process.

[0113] In this embodiment, by identifying whether there is ambiguity in the document content under a specified field during voice broadcast, and performing disambiguation processing in the case of ambiguity to obtain the target document content, the misunderstanding that the document review object may have when playing the target document content can be reduced, and incorrect review results can be avoided due to ambiguity, which helps to improve the accuracy and efficiency of real-time review of business documents.

[0114] In one embodiment, after step S204, which involves playing a voice verification prompt on the terminal upon successful call, the following steps may also be included:

[0115] The system receives the audit voice segments of the document audit object returned by the terminal and determines the first segment identifier corresponding to the received audit voice segments. If communication with the terminal is interrupted, upon resumption of communication, it obtains the segment identifier set returned by the terminal. The segment identifier set includes the second segment identifiers of each audit voice segment cached locally on the terminal. Based on the first segment identifier and the segment identifier set, it determines the third segment identifier and sends it to the terminal. The third segment identifier is the second segment identifier in the segment identifier set that does not match the first segment identifier. Based on the fusion result of the audit voice segments corresponding to each third segment identifier returned by the terminal and the received audit voice segments, it obtains the audit voice response from the document audit object.

[0116] In practice, the server relies on the voice feedback from the document reviewer to obtain the review result. If the voice data from the document reviewer fails to be saved successfully, and the network is interrupted, the collected voice segments will not be persistently stored, potentially leading to data recovery issues and ultimately causing the review process to stall. The status of business documents cannot be updated, requiring the previous steps to be repeated, which affects the review efficiency of urgent business documents.

[0117] In this embodiment, the terminal can split the collected voice data of the document review object into multiple review voice segments at a fixed duration and cache them locally on the terminal. In addition, the terminal can generate a unique segment identifier for each review voice segment cached locally on the terminal. For easy differentiation, the segment identifier corresponding to the review voice segment cached locally on the terminal is called the second segment identifier.

[0118] In practice, the server acts as a receiver, establishing a communication connection with the terminal via a wireless or wired network to receive the audit voice segments returned by the terminal. For each audit voice segment received, the server can obtain the segment identifier from its segment information. It can be understood that the segment identifier of the audit voice segment can be the same as the segment identifier of the same audit voice segment cached locally on the terminal. For ease of distinction, the segment identifier corresponding to the audit voice segment received by the server is referred to as the first segment identifier.

[0119] During communication between the server and the terminal, the server may receive one or more audit voice segments. When communication between the server and the terminal is interrupted, the transmission of audit voice segments may also be interrupted accordingly. After communication between the server and the terminal is restored, the server can obtain the segment identifier set returned by the terminal. This set may include the second segment identifier corresponding to each of all audit voice segments cached locally on the terminal.

[0120] Then, the server can match the first segment identifier corresponding to the audit voice segment it has obtained with each second segment identifier in the segment identifier set to determine the third segment identifier that the server has not recorded. Then, the server can send the third segment identifier to the terminal to request the audit voice segment corresponding to each third segment identifier.

[0121] After obtaining the audit voice segment corresponding to the third segment identifier returned by the terminal, the audit voice segment can be combined with the audit voice segments received before the communication was interrupted to form an audit voice.

[0122] In this embodiment, by matching the first fragment identifier with the fragment identifier aggregator, the third fragment identifier is determined and sent to the terminal. On the one hand, the integrity of the fragment identifier can be checked to ensure that the server can receive the complete audit voice fragment. On the other hand, the fragment retransmission mechanism can be used to selectively retransmit the unreceived fragments after communication is restored, avoiding the need to retransmit all fragments. This can improve data transmission efficiency while ensuring data integrity.

[0123] In one embodiment, determining the third fragment identifier and sending it to the terminal may include the following steps:

[0124] Multiple third fragment identifiers are determined, and the sending priority of each third fragment identifier is obtained; the sending priority decreases as the order of the third fragment identifiers decreases; the batch size is determined based on the real-time monitored network throughput, and fragment acquisition requests carrying third fragment identifiers are sent to the terminal in batches according to the batch size and sending priority; wherein the number of fragment acquisition requests sent in each batch does not exceed the batch size.

[0125] In practical applications, multiple third fragment identifiers can be determined by comparing and matching the first fragment identifier with the set of fragment identifiers. Although the server can send multiple fragment retransmission requests carrying third fragment identifiers to the terminal at one time, this process may cause the terminal to send a large amount of voice fragment data to the server at the same time, causing network bandwidth congestion and leading to new fragment loss.

[0126] In this embodiment, on the one hand, the transmission priority of each of the multiple third segment identifiers can be obtained first. Specifically, the basic priority can be determined based on the order of the third segment identifiers. Specifically, since the audit voice segments are generated in chronological order, segments with consecutive sequence numbers are more important for subsequent voice splicing. Therefore, in some embodiments, the priority of multiple segments with consecutive sequence numbers and missing segments (such as multiple segments numbered 10 to 15) can be set to high, and the priority of discrete missing segments (such as segments 5, 20, and 35) can be set to medium.

[0127] On the other hand, the batch processing quantity can be determined based on real-time monitored network throughput. Network throughput refers to the amount of data that can be transmitted between the server and the terminal per unit time, reflecting the current network's transmission capacity. The batch processing quantity refers to the maximum number of fragment retrieval requests that the server sends to the terminal each time. In one embodiment, the server can collect real-time network throughput data between itself and the terminal through a network monitoring module, and then determine the batch processing quantity for the current scenario based on a preset correspondence between network throughput and batch processing quantity.

[0128] Furthermore, all fragment acquisition requests carrying the third fragment identifier can be sorted according to the sending priority from high to low. Then, based on the determined batch processing quantity, the sorted fragment acquisition requests can be divided into multiple batches, with the number of requests in each batch not exceeding the batch processing quantity. Finally, the fragment acquisition requests are sent to the terminal in the order of the batches.

[0129] In this embodiment, by sending fragment retrieval requests carrying the third fragment identifier to the terminal in batches according to the batch processing quantity and sending priority, congestion during fragment retransmission can be avoided. Through the combination of priority scheduling and batch processing requests, the network bandwidth occupancy during retransmission can be controlled within the allowable range, effectively reducing the loss rate of retransmitted audit voice fragments.

[0130] To enable those skilled in the art to better understand the above embodiments, the present application is illustrated below with specific examples. However, it should be understood that the embodiments of the present application are not limited thereto.

[0131] like Figure 5 The diagram illustrates the processing framework of the business document review method in this application. Under this framework, document preparers can fill in document information and submit business documents for review in the online system. After the document preparer submits the document, the system will automatically detect the new business document for review and then extract key information from the document, such as the payment amount, purpose, and trading party.

[0132] Furthermore, the AI-powered outbound calling function can be activated, automatically dialing reviewers based on preset script templates. Once the reviewer answers, the AI ​​voice assistant clearly conveys key information about the document and solicits their review comments. Reviewers can provide their review results via voice commands or button presses, such as "passed," "disapproved," or "further verification required." The system accurately records the reviewer's feedback and transmits the results back to the enterprise's business system in real time. Finally, the document status can be updated. For example, if the review is approved, the enterprise's business system automatically updates the document status to "approved," facilitating subsequent business processes; if the review is disapproved, the system automatically updates the document status to "approved," and notifies the document creator to make corresponding modifications or additions.

[0133] like Figure 6 As shown, the specific steps may include the following:

[0134] Step S601: Obtain the business documents to be reviewed, and determine the urgency level of the business documents based on their content and business information.

[0135] In step S602, if the urgency value is less than the preset threshold, the business document is submitted to the non-urgent document pool to wait. When other business documents with an urgency value greater than or equal to the preset threshold are processed, the trained document processing model will process the business documents in the non-urgent document pool.

[0136] Step S603: If the urgency level value of the audit is greater than or equal to the preset threshold, the document content under the specified field in the business document is input into the ambiguity recognition model. If the ambiguity recognition model indicates that the document content under the specified field is ambiguous during voice broadcast, the context information of the document content under the specified field is obtained based on the document content and business information of the business document.

[0137] Step S604: Perform disambiguation processing on the document content under the specified field according to the context information, and obtain the target document content based on the disambiguation processing result.

[0138] Step S605: The document processing model determines the first business feature corresponding to the business document based on the document content of the business document.

[0139] Step S606: Based on the matching results of the first business feature and the second business feature of multiple candidate script templates, determine the target script template from the multiple candidate script templates.

[0140] Step S607: Input the document content under multiple fields in the target document content into the trained semantic analysis model to obtain the risk weight of the document content under each field output by the semantic analysis model.

[0141] Step S608: Determine the broadcast priority of document content under multiple fields according to each risk weight, and determine the voice parameters of document content under the multiple fields according to each risk weight.

[0142] Step S609: Generate voice review prompts for business documents based on broadcast priority, voice parameters, and the script in the target script template.

[0143] Step S610: Call the terminal of the document review object, play the voice review prompt through the terminal when the call is successful, obtain the review result of the business document according to the review voice reply of the document review object, and send a confirmation request for the review result to the terminal.

[0144] Step S611: In response to the terminal's confirmation request, a confirmation instruction is returned to confirm that the business document has been reviewed.

[0145] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0146] Based on the same inventive concept, this application also provides a business document review device for implementing the business document review method described above. The solution provided by this device is similar to the implementation scheme described in the above method; therefore, the specific limitations of one or more business document review device embodiments provided below can be found in the limitations of the business document review method described above, and will not be repeated here.

[0147] In one exemplary embodiment, such as Figure 7 As shown, a business document approval device is provided, including:

[0148] The document acquisition module 701 is used to acquire business documents to be reviewed, and determine the urgency value of the review of the business documents based on the document content and business information of the business documents.

[0149] The caching module 702 is used to submit the business document to a non-urgent document pool to wait if the urgency value of the audit is less than a preset threshold, so that when other business documents with an urgency value greater than or equal to the preset threshold are processed, the trained document processing model will process the business document in the non-urgent document pool.

[0150] The prompt generation module 703 is used to extract the target document content based on the document content by the document processing model if the urgency value of the review is greater than or equal to the preset threshold, and generate a voice review prompt for the business document according to the target document content and the script template.

[0151] Call module 704 is used to call the terminal of the document review object. When the call is successful, the voice review prompt is played through the terminal. The review result of the business document is obtained according to the review voice reply from the document review object, and a confirmation request for the review result is sent to the terminal.

[0152] The status switching module 705 is used to respond to the terminal returning a confirmation instruction in response to the confirmation request, and to determine that the business document has been reviewed.

[0153] In one embodiment, the prompt generation module 703 is used to:

[0154] The document processing model determines the first business feature corresponding to the business document based on the document content of the business document;

[0155] Based on the matching results between the first business feature and the second business feature of the multiple candidate script templates, a target script template is determined from the multiple candidate script templates; each candidate script template includes a script that highlights the document content associated with the second business feature.

[0156] Based on the content of the target document and the script in the target script template, generate a voice audit prompt for the business document.

[0157] In one embodiment, the target document content includes document content under multiple fields;

[0158] The prompt generation module 703 is used for:

[0159] The document content under the multiple fields is input into the trained semantic analysis model to obtain the risk weight of the document content under each field output by the semantic analysis model.

[0160] The broadcast priority of the document content under the plurality of fields is determined according to each of the risk weights, and the voice parameters of the document content under the plurality of fields are determined according to each of the risk weights; the broadcast priority increases with the increase of the risk weights, and the voice parameters of the document content are different for different risk weights.

[0161] Based on the broadcast priority, the voice parameters, and the script in the target script template, a voice review prompt is generated for the business document.

[0162] In one embodiment, the prompt generation module 703 is used to:

[0163] The document processing model retrieves the document content under a specified field from the business document;

[0164] The document content under the specified field is input into the ambiguity recognition model. When the ambiguity recognition model indicates that the document content under the specified field is ambiguous during voice broadcast, the context information of the document content under the specified field is obtained based on the document content of the business document and the business information.

[0165] Based on the context information, disambiguation processing is performed on the document content under the specified field, and the target document content is obtained based on the disambiguation processing result.

[0166] In one embodiment, the call module 704 is further configured to:

[0167] Receive the audit voice segment of the document audit object returned by the terminal, and determine the first segment identifier corresponding to the received audit voice segment;

[0168] If communication with the terminal is interrupted, when the communication is restored, the fragment identifier set returned by the terminal is obtained; the fragment identifier set includes the second fragment identifier of each audit voice fragment that has been cached locally on the terminal.

[0169] Based on the first fragment identifier and the fragment identifier set, a third fragment identifier is determined and sent to the terminal; the third fragment identifier is a second fragment identifier in the fragment identifier set that does not match the first fragment identifier.

[0170] Based on the fusion result of the audit voice segments corresponding to each of the third segment identifiers returned by the terminal and the received audit voice segments, the audit voice response from the document audit object is obtained.

[0171] In one embodiment, the call module 704 is further configured to:

[0172] Multiple third fragment identifiers are determined, and the transmission priority of each of the multiple third fragment identifiers is obtained; the transmission priority decreases as the order of the third fragment identifiers decreases;

[0173] The batch size is determined based on the real-time network throughput, and fragment acquisition requests carrying the third fragment identifier are sent to the terminal in batches according to the batch size and the sending priority; wherein the number of fragment acquisition requests sent in each batch does not exceed the batch size.

[0174] Each module in the aforementioned business document review device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0175] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores business document data. The I / O interfaces allow the processor to exchange information with external devices. The communication interface allows communication with external terminals via a network connection. When executed by the processor, the computer program implements a business document review method.

[0176] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0177] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0178] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0179] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0180] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0181] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0182] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for reviewing business documents, characterized in that, The method includes: Obtain the business documents to be reviewed, and determine the urgency value of the review of the business documents based on the document content and business information of the business documents; If the urgency level of the audit is less than a preset threshold, the business document is submitted to the non-urgent document pool to wait. When other business documents with an urgency level greater than or equal to the preset threshold are processed, the trained document processing model will process the business document in the non-urgent document pool. If the urgency level of the review is greater than or equal to the preset threshold, the document processing model extracts the target document content based on the document content, and generates a voice review prompt for the business document according to the target document content and the script template. The terminal that calls the document review object plays the voice review prompt through the terminal when the call is successful. The review result of the business document is obtained according to the review voice reply from the document review object, and a confirmation request for the review result is sent to the terminal. In response to the terminal returning a confirmation instruction for the confirmation request, it is determined that the business document has been reviewed.

2. The method according to claim 1, characterized in that, The step of generating voice verification prompts for the business document based on the target document content and the script template includes: The document processing model determines the first business feature corresponding to the business document based on the document content of the business document; Based on the matching results between the first business feature and the second business feature of the multiple candidate script templates, a target script template is determined from the multiple candidate script templates; each candidate script template includes a script that highlights the document content associated with the second business feature. Based on the content of the target document and the script in the target script template, generate a voice audit prompt for the business document.

3. The method according to claim 2, characterized in that, The target document content includes document content under multiple fields; The step of generating a voice verification prompt for the business document based on the content of the target document and the script in the target script template includes: The document content under the multiple fields is input into the trained semantic analysis model to obtain the risk weight of the document content under each field output by the semantic analysis model. The broadcast priority of the document content under the plurality of fields is determined according to each of the risk weights, and the voice parameters of the document content under the plurality of fields are determined according to each of the risk weights; the broadcast priority increases with the increase of the risk weights, and the voice parameters of the document content are different for different risk weights. Based on the broadcast priority, the voice parameters, and the script in the target script template, a voice review prompt is generated for the business document.

4. The method according to claim 1, characterized in that, The step of extracting the target document content from the document content using the document processing model includes: The document processing model retrieves the document content under a specified field from the business document; The document content under the specified field is input into the ambiguity recognition model. When the ambiguity recognition model indicates that the document content under the specified field is ambiguous during voice broadcast, the context information of the document content under the specified field is obtained based on the document content of the business document and the business information. Based on the context information, disambiguation processing is performed on the document content under the specified field, and the target document content is obtained based on the disambiguation processing result.

5. The method according to claim 1, characterized in that, After the step of playing the voice verification prompt through the terminal upon successful call, the method further includes: Receive the audit voice segment of the document audit object returned by the terminal, and determine the first segment identifier corresponding to the received audit voice segment; If communication with the terminal is interrupted, when the communication is restored, the fragment identifier set returned by the terminal is obtained; the fragment identifier set includes the second fragment identifier of each audit voice fragment that has been cached locally on the terminal. Based on the first fragment identifier and the fragment identifier set, a third fragment identifier is determined and sent to the terminal; the third fragment identifier is a second fragment identifier in the fragment identifier set that does not match the first fragment identifier. Based on the fusion result of the audit voice segments corresponding to each of the third segment identifiers returned by the terminal and the received audit voice segments, the audit voice response from the document audit object is obtained.

6. The method according to claim 5, characterized in that, The step of determining the third fragment identifier and sending it to the terminal includes: Multiple third fragment identifiers are determined, and the transmission priority of each of the multiple third fragment identifiers is obtained; the transmission priority decreases as the order of the third fragment identifiers decreases; The batch size is determined based on the real-time network throughput, and fragment acquisition requests carrying the third fragment identifier are sent to the terminal in batches according to the batch size and the sending priority; wherein the number of fragment acquisition requests sent in each batch does not exceed the batch size.

7. A business document verification device, characterized in that, The device includes: The document acquisition module is used to acquire business documents to be reviewed, and determine the urgency value of the review of the business documents based on the document content and business information of the business documents. The caching module is used to submit the business document to a non-urgent document pool if the urgency value of the audit is less than a preset threshold, so that when other business documents with an urgency value greater than or equal to the preset threshold are processed, the trained document processing model will process the business document in the non-urgent document pool. The prompt generation module is used to extract the target document content based on the document content by the document processing model if the urgency value of the review is greater than or equal to the preset threshold, and generate a voice review prompt for the business document based on the target document content and the script template. The calling module is used to call the terminal of the document review object. When the call is successful, the voice review prompt is played through the terminal. The review result of the business document is obtained according to the review voice reply from the document review object, and a confirmation request for the review result is sent to the terminal. The status switching module is used to respond to the terminal's confirmation request by returning a confirmation instruction to determine that the business document has been reviewed.

8. The apparatus according to claim 7, characterized in that, The prompt generation module is used for: The document processing model determines the first business feature corresponding to the business document based on the document content of the business document; Based on the matching results between the first business feature and the second business feature of the multiple candidate script templates, a target script template is determined from the multiple candidate script templates; each candidate script template includes a script that highlights the document content associated with the second business feature. Based on the content of the target document and the script in the target script template, generate a voice audit prompt for the business document.

9. The apparatus according to claim 8, characterized in that, The target document content includes document content under multiple fields; The prompt generation module is used for: The document content under the multiple fields is input into the trained semantic analysis model to obtain the risk weight of the document content under each field output by the semantic analysis model. The broadcast priority of the document content under the plurality of fields is determined according to each of the risk weights, and the voice parameters of the document content under the plurality of fields are determined according to each of the risk weights; the broadcast priority increases with the increase of the risk weights, and the voice parameters of the document content are different for different risk weights. Based on the broadcast priority, the voice parameters, and the script in the target script template, a voice review prompt is generated for the business document.

10. The apparatus according to claim 7, characterized in that, The prompt generation module is used for: The document processing model retrieves the document content under a specified field from the business document; The document content under the specified field is input into the ambiguity recognition model. When the ambiguity recognition model indicates that the document content under the specified field is ambiguous during voice broadcast, the context information of the document content under the specified field is obtained based on the document content of the business document and the business information. Based on the context information, disambiguation processing is performed on the document content under the specified field, and the target document content is obtained based on the disambiguation processing result.

11. The apparatus according to claim 7, characterized in that, The call module is also used for: Receive the audit voice segment of the document audit object returned by the terminal, and determine the first segment identifier corresponding to the received audit voice segment; If communication with the terminal is interrupted, when the communication is restored, the fragment identifier set returned by the terminal is obtained; the fragment identifier set includes the second fragment identifier of each audit voice fragment that has been cached locally on the terminal. Based on the first fragment identifier and the fragment identifier set, a third fragment identifier is determined and sent to the terminal; The third fragment identifier is a second fragment identifier in the fragment identifier set that does not match the first fragment identifier; Based on the fusion result of the audit voice segments corresponding to each of the third segment identifiers returned by the terminal and the received audit voice segments, the audit voice response from the document audit object is obtained.

12. The apparatus according to claim 11, characterized in that, The call module is also used for: Multiple third fragment identifiers are determined, and the transmission priority of each of the multiple third fragment identifiers is obtained; the transmission priority decreases as the order of the third fragment identifiers decreases; The batch size is determined based on the real-time network throughput, and fragment acquisition requests carrying the third fragment identifier are sent to the terminal in batches according to the batch size and the sending priority; wherein the number of fragment acquisition requests sent in each batch does not exceed the batch size.

13. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images