An intelligent review device and method for a business system

Abstract

This invention provides an intelligent review device and method for a business system. The intelligent review device includes an intelligent recognition module, an interface communication module, and a review module, with the review module located within the business system. The intelligent recognition module identifies key information in the document to be recognized and outputs the recognition result; the data type of the recognition result is structured data. The interface communication module encapsulates the recognition result into a data packet of a preset format and transmits the data packet to the review module in the business system through a preset interface. The review module initiates a review of the business system upon receiving the data packet of the preset format, verifying whether the key information entered into the document to be recognized pre-stored in the business system is consistent with the data packet of the preset format. This invention achieves automatic review of the accuracy of information entered into the business system, improving the review efficiency of the business system.

Description

Technical Field

[0001] This invention relates to the field of intelligent review technology, and in particular to an intelligent review device and method for a business system. Background Technology

[0002] In core business processes such as procurement, legal affairs, and contract management, compliance review of formatted documents, primarily PDF scans, such as purchase orders or contracts (e.g., verification of complete signatures, key terms, and date accuracy) is a necessary but demanding task. The review technology used by relevant business systems typically relies on traditional manual processes and semi-automated tools. Reviewers manually visually inspect documents or convert images to text using recognition software, then verify the data through keyword searches. This requires manual switching and comparison between the model recognition interface and the business system interface, resulting in low review efficiency, a high risk of errors, and reduced reliability of the business system's audit process. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide an intelligent review device and method for a business system. After the intelligent recognition module identifies key information of the document to be identified, the review module inside the business system can be triggered to automatically review the pre-stored key information input data of the document to be identified, thereby improving the review efficiency and accuracy of the business system. At the same time, the intelligent recognition module capable of identifying key information is independently set outside the business system. The intelligent recognition module and the business system can be independently iterated and upgraded, with good maintainability and scalability, thereby improving the reliability of the business system's review.

[0004] To achieve the above objectives, the technical solutions adopted in the embodiments of the present invention are as follows: In a first aspect, embodiments of the present invention provide an intelligent review device for a business system, comprising: an intelligent identification module, an interface communication module, and a review module, wherein the review module is disposed within the business system, and the interface communication module is communicatively connected to the review module within the business system through a preset interface; The intelligent recognition module is used to identify key information in the document to be recognized and output the recognition result; wherein the data type of the recognition result is structured data; The interface communication module is used to encapsulate the recognition result into a data packet of a preset format, and transmit the data packet of the preset format to the review module in the business system through the preset interface; The review module is used to initiate a business system review when the data packet of the preset format is received, and to review whether the key information input data corresponding to the document to be identified pre-stored in the business system is consistent with the data packet of the preset format.

[0005] Furthermore, the present invention provides a first possible implementation of the first aspect, wherein the intelligent review device of the business system further includes: a preprocessing module; The preprocessing module is used to convert the document to be recognized into a standard document image in a preset format, and input the standard document image into the intelligent recognition module to identify key information of the document to be recognized.

[0006] Furthermore, this embodiment of the invention provides a second possible implementation of the first aspect, wherein the intelligent recognition module includes an image recognition unit and a keyword recognition unit; the key information includes a preset pattern and keywords; The image recognition unit is used to recognize the preset pattern in the document to be recognized and output the recognition result of the preset pattern; The keyword recognition unit is used to locate and recognize keywords in the document to be recognized, and output the keyword recognition results.

[0007] Furthermore, this embodiment of the invention provides a third possible implementation of the first aspect, wherein the keyword recognition unit includes a key field extraction model and a keyword judgment model; the keywords include predefined key fields and preset keywords; The image recognition unit is used to identify whether the document to be recognized contains a preset pattern, and outputs the structured determination result of the preset pattern; The key field extraction model is used to perform character recognition on the document to be identified to obtain the full text of the document to be identified, locate and extract predefined key fields from the full text, and output the structured data of the predefined key fields. The keyword judgment model is used to perform full-text scanning of the full text of the document to be identified based on preset keywords, determine whether the full text contains the preset keywords, and output the structured judgment result of the preset keywords.

[0008] Furthermore, this embodiment of the invention provides a fourth possible implementation of the first aspect, wherein the document to be identified includes a contract document or a purchase order document; The preset pattern includes a valid signature pattern, the predefined key fields include the contract start date and the contract end date, and the preset keywords include the set clause name.

[0009] Furthermore, this embodiment of the invention provides a fifth possible implementation of the first aspect, wherein the review module is further configured to parse the structured determination result of the preset pattern, the structured data of the key field, and the structured determination result of the preset keyword from the data packet of the preset format, and query the key information input data corresponding to the document to be identified from the database of the business system, and perform consistency verification on the parsing result of the data packet of the preset format and the key information input data.

[0010] Furthermore, this embodiment of the invention provides a sixth possible implementation of the first aspect, wherein the review module is further configured to generate an intelligent review conclusion corresponding to the document to be identified after reviewing whether the key information input data corresponding to the document to be identified pre-stored in the business system is consistent with the parsing result of the data packet in the preset format, and store the intelligent review conclusion in the database of the business system.

[0011] Furthermore, this embodiment of the invention provides a seventh possible implementation of the first aspect, wherein the intelligent review device of the business system further includes: a display module, the display module being disposed within the business system; The display module is used to present the intelligent review conclusions in the form of a comparison table or a structured list.

[0012] Furthermore, this embodiment of the invention provides an eighth possible implementation of the first aspect, wherein the intelligent review device of the business system further includes: a message queue and a cluster scheduler, and the intelligent identification module includes multiple parallel AI models; The message queue is used to receive multiple document recognition tasks; The cluster scheduler is used to assign multiple document recognition tasks to multiple parallel AI models for key information recognition.

[0013] Secondly, embodiments of the present invention also provide an intelligent review method for a business system, applied to the intelligent review apparatus for the business system described in any one of the first aspects, the intelligent review method for the business system comprising: The intelligent recognition module identifies key information in the document to be recognized and outputs the recognition results; wherein the data type of the recognition results is structured data. The identification result is encapsulated into a data packet of a preset format based on the interface communication module, and the data packet of the preset format is transmitted to the review module through the preset interface; Upon receiving the data packet in the preset format, the business system initiates a review, and the review module examines whether the key information input data corresponding to the document to be identified, which is pre-stored in the business system, is consistent with the data packet in the preset format.

[0014] Furthermore, the document to be identified includes contract documents or purchase order documents; The preset pattern includes a valid signature pattern, the predefined key fields include the contract start date and the contract end date, and the preset keywords include the set clause name.

[0015] This invention provides an intelligent review device and method for a business system. The intelligent review device includes an intelligent identification module, an interface communication module, and a review module. The review module is located inside the business system, while the intelligent identification module and the interface communication module are located outside the business system. The interface communication module communicates with the review module inside the business system through a preset interface. The intelligent identification module is used to identify key information in the document to be identified and output the identification result. The data type of the identification result is structured data. The interface communication module is used to encapsulate the identification result into a data packet of a preset format and transmit the data packet of the preset format to the review module through the preset interface. The review module is used to initiate a business system review upon receiving the data packet of the preset format and to review whether the key information input data corresponding to the document to be identified pre-stored in the business system is consistent with the data packet of the preset format. This invention utilizes an independent intelligent recognition module outside the business system to identify key information in the document to be identified. The data object encapsulated based on the recognition results is then transmitted to the review module via an interface provided by the business system. This triggers the review module within the business system to automatically review the pre-stored key information data of the document to be identified, thus achieving automatic review of the accuracy of information entered by the business system. This improves the review efficiency and accuracy of the business system. Furthermore, the intelligent recognition module and the business system can be independently upgraded and iterated, exhibiting good maintainability and scalability, thereby enhancing the reliability of the business system's review process.

[0016] Other features and advantages of the embodiments of the present invention will be set forth in the following description, or some features and advantages may be inferred from the description or determined without doubt, or may be learned by practicing the techniques described above in the embodiments of the present invention.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This invention provides a schematic diagram of the structure of an intelligent review device for a business system according to an embodiment of the present invention. Figure 2 A flowchart of an intelligent review method for a business system provided by an embodiment of the present invention is shown; Figure 3 This invention illustrates an automatic review process for a business system provided by an embodiment of the present invention; Figure 4 A schematic diagram of the intelligent review device structure of another business system provided by an embodiment of the present invention is shown. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0021] In the daily operations of the manufacturing industry, a large number of business orders rely on purchase orders in the form of PDF scans. Compliance reviews of purchase orders or contract documents primarily depend on manual visual inspection of key legal elements (such as the completeness of both parties' signatures), and manual entry of the order's effective date into the core business system. This approach suffers from inefficiency, poor data quality, and weak risk control. With the development of artificial intelligence technology, techniques have emerged that use optical character recognition (OCR) or deep learning models to automatically analyze documents.

[0022] One approach uses deep learning models to intelligently analyze and extract information from preprocessed document images, outputting structured review reports. However, this output is disconnected from the business system, failing to create an automated workflow. When users (such as contract reviewers) need to perform actual operations based on this report within the core business system, they must manually switch between two separate systems: first, reading the report in the AI ​​system interface, then manually logging into the business system, locating the corresponding order, visually comparing key information (such as dates) in the AI ​​report with records in the business system, and finally deciding whether and how to modify it. This verification process is interrupted and non-automated, essentially failing to integrate AI capabilities into the business process. Furthermore, there is a lack of automated verification mechanisms for existing data in the business system, and the human-computer interface is detached from the actual business operation environment. The review report display interface is independent, separate from the core business operation interface used by employees in their daily work. This context-insensitive display increases the user's cognitive load and operational steps, failing to provide "what you see is what you get" comparison and decision support within the business operation context (i.e., within the business system interface), resulting in limited user experience and efficiency improvements.

[0023] Another approach relies on large AI models to identify key contract information. However, this technology focuses solely on extracting key information from the document. While the model output is structured data, it doesn't compare and verify it against existing data in the business system. If manually entered contract dates or other information already exist in the business system, this technology cannot automatically detect discrepancies or drive the business system to correct the data, resulting in the AI ​​extraction and business data being two separate entities. Furthermore, using a single large AI model to process text and region types may limit the accuracy of recognition for contract documents containing complex visual elements (such as complete signatures).

[0024] In summary, most existing technical solutions are limited to standalone document recognition or information extraction tools, resulting in a disconnect between their output and the core backend business systems. Specifically, key purchase order data (such as contract dates) stored in the business systems is manually entered, and the systems cannot utilize AI to extract "real data" from the original documents for automated verification and error correction, leading to inconsistent data quality. Furthermore, AI review conclusions cannot be deeply integrated into the business operation interface, failing to form a complete closed loop from "intelligent perception" to "business processing." Users still need to switch between multiple systems for comparison and operation, resulting in process interruptions and low automation. Therefore, how to construct a technical solution that deeply integrates artificial intelligence with business processes to achieve a closed-loop processing from intelligent document review to business data verification has become an urgent technical problem to be solved.

[0025] To address the aforementioned issues, this invention provides an intelligent review device and method for a business system. The following provides a detailed description of the embodiments of this invention.

[0026] This embodiment provides an intelligent review device for a business system, see below. Figure 1 The diagram shows the structure of the intelligent review device of the business system. The intelligent review device of the business system includes: intelligent identification module 10, interface communication module 20 and review module 30. like Figure 1 As shown, the review module 30 is located inside the business system, while the intelligent identification module 10 and the interface communication module 20 can be located outside the business system. The interface communication module 20 communicates with the review module 30 inside the business system through a preset interface. The intelligent identification module 10 and the review module 30 are connected via an interface communication module 20. The intelligent recognition module 10 is used to identify key information in the document to be recognized and output the recognition results; wherein, the data type of the recognition results is structured data; The aforementioned documents to be identified can be purchase orders, contract documents, or other formatted documents that need to be verified against statistical information in the business system or undergo compliance audits. The aforementioned business system can be an information statistics system that records key information from purchase orders, contract documents, or other important documents or invoices.

[0027] The document to be identified is input into the intelligent recognition module 10 in the form of a document (such as a PDF or Word document) or a scanned image. The intelligent recognition module 10 can be an AI model or a trained deep learning model, so that the intelligent recognition module 10 can identify key information in the document to be identified. This key information can be statistical information of the document to be identified that is synchronously recorded in the business system, such as key terms or keywords, dates, signatures, amounts, and categories, etc., which are information that needs to be verified or reviewed.

[0028] To facilitate the review module 30 in the business system to verify and review the data based on the recognition module 10, the data type of the above recognition results (such as data formats such as JSON or Excel) can be a structured judgment result or recognition result. For example, it can be a Boolean value indicating whether the document to be recognized contains a valid signature or specific clause, or it can be structured data such as date, keywords or key clauses.

[0029] The interface communication module 20 is used to encapsulate the recognition results into a data packet of a preset format and transmit the data packet of the preset format to the review module in the business system through the interface provided by the business system. The intelligent recognition module 10 transmits the recognition result to the review module 30 within the business system via the interface communication module 20. The interface communication module 20 transmits data with the business system through a preset interface, thereby transmitting the recognition result of the intelligent recognition module 10 to the review module 30 within the business system. The preset interface can be the interface of the business system, or a matching interface capable of data communication with the interface of the business system.

[0030] The interface communication module 20 serves as the connection layer between the intelligent recognition module 10 and the business system. It encapsulates the structured data recognition results generated by the intelligent recognition module 10 into a standard data packet (i.e., a data packet in a preset format, such as a standard JSON data packet). The encapsulated data packet in the preset format is actively pushed to the backend of the business system through a preset interface (using an interface protocol (API protocol) corresponding to the interface of the business system) and bound to the order corresponding to the document to be recognized, so that the review module 30 in the business system can perform information review and verification based on the recognition results.

[0031] The review module 30 is used to initiate a business system review when a data packet of a preset format is received, and to review whether the key information input data corresponding to the document to be identified stored in the business system is consistent with the data packet of the preset format.

[0032] After receiving the recognition result from the intelligent recognition module 10 transmitted by the interface communication module 20, the review module 30 in the business system automatically triggers the verification logic, automatically compares whether each recognition result in the preset format data packet is consistent with the pre-stored manual record value in the business system, and determines whether the document to be recognized has the key information that should be present (such as valid signatures or key clauses), and generates the final review conclusion and persists it.

[0033] The intelligent review device for the business system provided in this embodiment identifies key information of the document to be identified through an independent intelligent recognition module outside the business system. The data object encapsulated based on the recognition result is transmitted to the review module inside the business system through the interface provided by the business system. This can trigger the review module in the business system to automatically review the pre-stored key information data of the document to be identified, thereby improving the review efficiency and accuracy of the business system. At the same time, the intelligent recognition module and the business system can be independently upgraded and iterated, which has good maintainability and scalability, and improves the reliability of the business system's review.

[0034] In one implementation, such as Figure 1 As shown, the intelligent review device for the business system provided in this embodiment further includes: a preprocessing module 40; The preprocessing module 40 is used to convert the document to be recognized into a standard document image in a preset format, and input the standard document image into the intelligent recognition module 10 to recognize key information of the document to be recognized.

[0035] The preprocessing module 40 receives a scanned copy of the document to be recognized (such as a PDF document or a scanned image), and uses document processing tools (such as a tool library that supports PDF document generation, editing, text and image extraction, such as the Apache PDFBox 3.0+ tool library) or image processing tools to convert the scanned copy of the document to be recognized into a digital image in a standard image format. It then uses image processing tools (such as the Java BufferedImage API and the Thumbnailator image processing library) to perform preprocessing operations such as size normalization and resolution processing on the digital image to convert the document to be recognized into an image of uniform size and resolution, denoted as a standard document image. This standard document image is then input into the intelligent recognition module 10 as the image of the document to be recognized for key information recognition.

[0036] In one embodiment, the key information provided in this embodiment includes preset patterns and keywords; the intelligent recognition module includes an image recognition unit and a keyword recognition unit; The image recognition unit is used to recognize preset patterns in the document to be recognized and output the recognition results; The aforementioned image recognition unit can be a trained image recognition model, which can identify whether the document to be recognized contains a preset pattern, and can also identify the pattern type of the preset pattern in the document to be recognized. The output recognition result can include any one or more of the following: whether the document to be recognized contains a preset pattern, the pattern type of the preset pattern in the document to be recognized, and the recognized preset pattern image.

[0037] The keyword recognition unit is used to locate and recognize keywords in the document to be recognized, and output the keyword recognition results.

[0038] The keyword recognition unit can locate, recognize, and extract keywords in the document to be recognized using character recognition algorithms, keyword matching algorithms, and cross-modal recognition methods between images and text. The output keyword recognition results can include the extracted keyword content, the keyword location results, and whether the document to be recognized contains the keyword.

[0039] In one embodiment, the keyword recognition unit provided in this embodiment includes a key field extraction model and a keyword judgment model; the keywords include predefined key fields and preset keywords; The image recognition unit is used to identify whether the document to be recognized contains a preset pattern and outputs the structured judgment result of the preset pattern; The preprocessed standard document image is invoked via a pre-packaged HTTP or gRPC protocol API to call an image recognition unit deployed on a local server. This image recognition unit is an image-language recognition model, such as a pre-trained Vision-Language (VL) model or other cross-modal recognition model between images and text. The image recognition unit is configured to perform end-to-end image understanding tasks. After the standard document image is input into the image recognition unit, it directly outputs a structured judgment result, which is a Boolean value indicating whether the document to be recognized contains a preset pattern. The input instructions for the image recognition unit are constructed through a multimodal Prompts dynamic generation module to improve the accuracy of the image recognition unit.

[0040] The aforementioned pre-trained visual-language model is a pre-defined prompt-guided visual-language model. This means that before calling the VL model for signature completeness recognition, it uses pre-written, fixed prompts specific to the business scenario to guide the model (e.g., using "valid signature exists" and "non-valid signature does not exist" as fixed prompts), rather than using general instructions or no instructions at all. General pre-trained visual-language models may suffer from insufficient accuracy or poor stability when facing abstract tasks like "determining whether a signature is complete." By using pre-defined, precisely written prompts tailored to the business scenario, the abstract task is transformed into concrete, executable instructions, significantly improving the model's inference accuracy and output stability for this specific task. This avoids misjudgments or omissions caused by ambiguous instructions, thus ensuring the reliability and consistency of AI perception results.

[0041] Since the recognition of pre-defined patterns such as signatures mainly relies on the visual features of images, the Vision-Language (VL) model has a natural advantage in pattern recognition based on visual features, thus improving the accuracy of pattern recognition.

[0042] The key field extraction model is mainly used to extract and output predefined key fields (usually standardized key-value pairs) from the document to be identified. The output of the key field extraction model is structured data, which includes the specific content of the predefined key fields. On the other hand, the keyword judgment model only determines whether the document to be identified contains preset keywords. The output of the keyword judgment model is Boolean data to indicate whether the preset keywords exist in the document to be identified.

[0043] The key field extraction model is used to perform character recognition on the document to be recognized to obtain the full text of the document, locate and extract predefined key fields from the full text, and output the structured data of the key fields. The key field extraction model performs optical character recognition (OCR) on standard document images to obtain the full text of the document to be recognized. It employs rule-based and keyword-based field extraction algorithms (such as rule / template methods (based on regular expressions, keyword matching, fixed positions or predefined templates, such as contract start and end dates recorded in a specific location in a contract, or tax number in the upper left corner of an invoice)), machine learning-based methods (using labeled data to train classification or sequence labeling models, and identifying fields through features such as part-of-speech, context, and position), and deep learning and large model methods) to extract key fields from the document to be recognized. It accurately locates and extracts predefined key fields (predefined key fields are key business fields, such as contract start date, contract end date, or purchase date) from the text, and outputs them as standardized key-value pair structured data objects, thus obtaining the structured data of the key fields.

[0044] The keyword judgment model is used to perform full-text scanning of the full text of the document to be recognized based on preset keywords, determine whether the full text contains preset keywords, and output the structured judgment result of the preset keywords.

[0045] The keyword identification model acquires the full text of the document to be identified, scans the full text according to a preset rule base, and outputs preset keyword tags. These preset keywords may include related company clauses, risk clauses, or other defined clause names. When the preset keyword is a risk clause, the keyword identification model outputs a risk clause tag when it determines that the full text of the document to be identified contains a risk clause.

[0046] The keyword judgment model uses a predefined keyword matching algorithm to perform fast scanning, automatically determine whether a document contains preset keywords, and outputs a structured judgment result and preset keyword tags, such as a boolean value corresponding to each preset keyword (this boolean value can indicate whether the preset keyword exists in the document to be identified).

[0047] Keyword identification models can quickly scan and determine whether a document to be identified contains preset keywords through keyword matching, and then quickly determine whether a specified clause name exists in the document to be identified. Key field extraction models, on the other hand, require text information extraction.

[0048] The extraction of text information such as dates and terms in the key field extraction model requires OCR and semantic understanding. A single technology cannot simultaneously handle two different types of extraction tasks with high precision. The key field extraction model and the keyword judgment model execute OCR, keyword field extraction algorithm and keyword matching in parallel without blocking each other. This can make full use of computing resources, avoid the time delay caused by serial processing, and significantly improve the comprehensiveness, accuracy and overall processing efficiency of information extraction.

[0049] In one implementation, the document to be identified in this embodiment includes a contract document or a purchase order document; The preset patterns include valid signature patterns, the predefined key fields include the contract start date and the contract end date, and the preset keywords include setting the term name.

[0050] Input the standard document image corresponding to the contract document or purchase order document into the image recognition unit, and the image recognition unit outputs a Boolean value indicating whether the contract document or purchase order document contains valid signatures of both parties; The standard document image corresponding to the contract document or purchase order document is input into the key field extraction model. The key field extraction model performs optical character recognition (OCR) on the standard document image to obtain the full text. Using a rule-based and keyword-based field extraction algorithm, it accurately locates and extracts predefined key fields such as purchase date, contract start date and contract end date from the text, and outputs them as standardized key-value pair structured data objects.

[0051] The full text of the contract document or purchase order document obtained by the keyword judgment model is fed into the keyword recognition unit. The keyword judgment model applies a set of predefined keyword matching rules to quickly scan the full text of the contract document or purchase order document, such as automatically judging whether the document contains preset keywords such as "affiliated company clause (or risk clause or other specific clauses)", and outputs the corresponding mark (Boolean value) of the preset keyword.

[0052] In one embodiment, the review module provided in this embodiment is further used to parse the structured judgment result of the preset pattern, the structured data of the key field and the structured judgment result of the preset keyword from the data packet of the preset format, and to query the key information input data corresponding to the document to be identified from the database of the business system, and to perform consistency verification on the parsing result of the data packet of the preset format and the key information input data.

[0053] After receiving a data packet in a preset format, the business system automatically triggers the review module to begin data verification and review, parsing the data packet to extract the structured judgment results of the preset pattern, the structured data of the key fields, and the structured judgment results of the preset keywords.

[0054] For example, the system can parse a Boolean value from the data packet to indicate whether a preset pattern exists. Based on the order identifier of the document to be identified corresponding to the data packet, the system can query the database of the business system to find the Boolean value of whether the preset pattern exists, which was manually entered into the order. The parsed Boolean value is then compared and verified with the Boolean value of whether the preset pattern exists, which was manually entered into the database of the business system, to determine whether the recognition value of the intelligent recognition module is consistent with the corresponding value entered into the business system.

[0055] For example, key fields such as contract start and end dates are parsed from data packets (structured data of key fields extracted from the key field extraction model in the intelligent recognition module). Based on the order identifier of the document to be recognized corresponding to the data packet, the system queries the database of the business system for the manually entered data (i.e., the system record values) corresponding to the key fields such as contract start and end dates of the order. The business system calls its internal comparison logic to compare the key fields such as contract start and end dates parsed from the data packet with the corresponding entered data in the business system to perform consistency verification. In practical applications, date verification and extended calculation logic can be implemented based on the time processing package (Java java.time) in the business system.

[0056] In one embodiment, the review module provided in this embodiment is further used to generate an intelligent review conclusion for the document to be identified after reviewing whether the key information input data corresponding to the document to be identified pre-stored in the review business system is consistent with the parsing result of the data packet in the preset format, and then storing the intelligent review conclusion in the database of the business system.

[0057] The aforementioned intelligent review conclusions may include the intelligent identification value of the business order corresponding to the document to be identified (i.e., the parsing result of the data packet), the data entered by the business system, and the consistency verification result. The data stored in the intelligent review conclusions may be in the form of structured data, and the intelligent review conclusions may be persistently stored in the database of the business system.

[0058] In one implementation, such as Figure 1 As shown, the intelligent review device for the business system provided in this embodiment further includes: a display module 50, which is installed in the business system and is communicatively connected to the review module 30; The display module 50 is used to present the intelligent review conclusions in the form of a comparison table or a structured list.

[0059] In the order management interface of the business system, a display module (i.e., display panel) is embedded to show the intelligent review results. This display panel can visualize key fields in the form of comparison tables or structured lists.

[0060] For example, for key information already stored in the business database (including preset patterns, key fields (such as manually entered contract start dates), and preset keywords (such as markers for risk clauses), the display module presents the intelligent recognition module's identification value (i.e., the parsing result of the data packet, originating from the original document to be identified) and the key information entered into the business system (originating from the business database) in a parallel comparison manner. If the intelligent recognition module's identification value matches the corresponding key information entered data, it can be marked with a checkmark or other icon to highlight the comparison result; if the intelligent recognition module's identification value does not match the corresponding key information entered data, it can be highlighted with color (such as red font or background color) or a difference warning icon (such as a warning triangle) to highlight the abnormal review result.

[0061] For critical information whose records are not stored in the business database, the display module can also show non-comparison information extracted from the original document to be identified by the intelligent recognition module. This information may include, but is not limited to, Boolean results for signature completeness and risk markers such as whether related company clauses are included. For non-comparison information, the recognition results of the intelligent recognition module can be presented intuitively using clear icons or status color blocks such as checkmarks (indicating "yes / present") and crosses (indicating "no / absent").

[0062] By embedding a display module within the business system to showcase intelligent review conclusions, users can access all intelligent review conclusions at once and perform verification and subsequent business operations based on discrepancy prompts. The display module automatically parses the data difference types and field attributes, generating differentiated display styles (such as side-by-side comparison tables, color highlighting, difference icons, or status labels). If presented only as plain text or a simple list, users would still need to manually compare AI-recognized values ​​with data entered into the business system, making it difficult to intuitively and quickly locate differences. The differentiated display design of the module allows users to obtain all intelligent review conclusions and difference information within a single interface, significantly reducing the cognitive burden and time cost of manual verification. This provides clear and intuitive visual support for efficient decision-making, thereby improving collaborative efficiency.

[0063] In one embodiment, the intelligent review device for the business system provided in this embodiment further includes: a message queue and a cluster scheduler, and the intelligent recognition module includes multiple parallel AI models; The message queue is used to receive multiple document recognition tasks; The cluster scheduler is used to distribute multiple document recognition tasks to multiple parallel AI models for key information recognition.

[0064] To enable batch review of orders corresponding to multiple documents to be identified, the AI ​​model can receive review tasks for contracts or purchase orders through a message queue. The cluster scheduler dynamically distributes the review tasks to multiple parallel AI models for key information identification, achieving high-concurrency processing. The recognition results of the AI ​​models are encapsulated and transmitted to the business system, ensuring high efficiency in processing massive amounts of documents and seamless user experience.

[0065] By deploying message queues and a cluster scheduler, and containerizing the cluster and message queue, the multimodal analysis engine is containerized using a microservice architecture and deployed in a container orchestration cluster (such as Kubernetes). Simultaneously, a message queue is introduced as the asynchronous task scheduling hub to achieve peak shaving and dynamic scheduling of batch tasks. Related technologies only describe the processing logic of a single document and do not address the performance requirements of enterprise-level applications for high concurrency, high availability, and elastic scaling. Without the deployment of message queues and a cluster scheduler, the intelligent review device will experience task backlog, response timeouts, or even service crashes when dealing with batches of documents (such as thousands of orders during month-end centralized settlement), making it unusable in actual production. This invention, by adopting a containerized cluster and message queue architecture, enables the system to linearly scale its processing capacity during peak task periods, possessing the practicality and scalability for industrial deployment.

[0066] The intelligent review device for the aforementioned business system provided in this embodiment constructs a highly cohesive and loosely coupled system boundary. High cohesion is manifested in the fact that the perception layer AI model completely encapsulates all functions related to multimodal perception (such as image preprocessing, signature detection, field extraction, and risk scanning) internally, with a single responsibility: outputting only standardized, non-business semantic structured data (such as Boolean values ​​or date strings). The decision-making layer business system, on the other hand, encapsulates all business logic (such as data comparison rules, date verification, and conclusion fusion) within itself, relying only on standardized data input externally. Loose coupling is manifested in the fact that the perception layer (intelligent recognition module) and the decision-making layer (review module of the business system) interact through predefined API interfaces and standardized JSON data contracts, relying only on the interface protocol rather than each other's internal implementations. When the perception layer changes its AI model or upgrades its algorithm, as long as the output data format remains unchanged, the business system requires no modification; similarly, when the business system adjusts its verification rules, the perception layer is completely unaffected. The intelligent review device of the business system features a highly cohesive and loosely coupled design, which allows each module of the system to evolve independently and be flexibly replaced, significantly improving the system's maintainability, scalability, and technical adaptability.

[0067] The intelligent review device for the above-mentioned business system provided in this embodiment also realizes the triggering and data flow design of "reverse verification": the core innovation lies in "using AI (intelligent recognition module) to extract data as a trigger to drive the business system to verify its own data". The business system receives the recognition result of the intelligent recognition module, calls the key information stored in the business database to input data, performs data consistency verification, and displays the verification result on the front end.

[0068] The intelligent review device for the business system provided in this embodiment also achieves non-intrusive deep integration of the front end. It integrates an AI intelligent review conclusion display module into the existing business system, ensuring that its loading, rendering and interaction do not affect the original business functions, and can dynamically generate operation buttons according to different intelligent review conclusions.

[0069] Corresponding to the intelligent review device for business systems provided in the above embodiments, this embodiment provides an intelligent review method for business systems, see below. Figure 2 The flowchart shown below illustrates the intelligent review method for the business system. This method mainly includes the following steps: Step S202: Based on the intelligent recognition module, key information of the document to be recognized is identified, and the recognition result is output; wherein, the data type of the recognition result is structured data; Step S204: Based on the interface communication module, the recognition result is encapsulated into a data packet of a preset format, and the data packet of the preset format is transmitted to the review module through the preset interface; Step S206: Upon receiving a data packet in a preset format, the business system review is initiated. The review module checks whether the key information input data corresponding to the document to be identified pre-stored in the business system is consistent with the data packet in the preset format.

[0070] The intelligent review method for the business system provided in this embodiment identifies key information of the document to be identified by an independent intelligent recognition module outside the business system. The data object encapsulated according to the recognition result is transmitted to the review module through the interface provided by the business system. This can trigger the review module in the business system to automatically review the pre-stored key information data of the document to be identified, thereby improving the review efficiency and accuracy of the business system. At the same time, the intelligent recognition module and the business system can be independently upgraded and iterated, which has good maintainability and scalability, and improves the reliability of the business system's review.

[0071] In one implementation, the document to be identified provided in this embodiment includes a contract document or a purchase order document; The preset patterns include valid signature patterns, the predefined key fields include the contract start date and the contract end date, and the preset keywords include setting the term name.

[0072] The intelligent review method for the above-mentioned business system provided in this embodiment fundamentally eliminates the processing breakpoint between the unstructured PDF scanned version of the purchase order document and the business system. It establishes a highly cohesive, loosely coupled, and sustainably evolving technical architecture for reviewing PDF scanned purchase orders, achieving a leapfrog improvement in review efficiency and process restructuring. It has clear hierarchical boundaries and standardized interface design, enabling the perception AI model, business rules, and front-end experience to be independently iterated and upgraded, and possesses excellent maintainability and scalability.

[0073] The method provided in this embodiment has the same implementation principle and technical effect as the aforementioned embodiments. For the sake of brevity, any parts not mentioned in the method embodiment can be referred to the corresponding content in the aforementioned device embodiment.

[0074] Based on the foregoing embodiments, this embodiment provides an example of automatically auditing a business system using the intelligent review method described above. See, for example... Figure 3 The flowchart shown below illustrates the automated review process of the business system. Please refer to the following steps for execution: Step S301, Document Receiving and Standardization Processing: The preprocessing module receives the original PDF document of the purchase order or contract uploaded by the user, and converts the original PDF document into a standard document image with uniform size, resolution and color space through format parsing and image processing technology.

[0075] See also Figure 4 The diagram shows the structure of an intelligent review device for another business system. This intelligent review device mainly consists of two parts: an intelligent review processing cluster and a business system. The intelligent review processing cluster includes a preprocessing module 40 (also known as a document access and preprocessing module), an intelligent recognition module 10 (also known as a multimodal parallel intelligent perception and information structuring model), and an interface communication module 20. The business system includes a review module 30 and a display module 50. The review module 30 can interact with the business system database to obtain key information input data stored in the business system for data consistency verification.

[0076] The preprocessing module is mainly used to convert the original PDF document into an image and perform image standardization to obtain a standard document image. This standard document image is a standardized digital image sequence, which can provide a high-quality, unambiguous input source for the subsequent intelligent recognition module.

[0077] Step S302: Multimodal parallel perception and information structuring; the intelligent recognition module synchronously receives standard document images, and the image recognition unit, key field extraction model and keyword judgment model are executed in parallel.

[0078] The intelligent recognition module 10 synchronously receives standard document images and drives its three internal dedicated analysis units to work in parallel: the image recognition unit analyzes the valid signature elements in the image based on a deep learning visual-language model and outputs a completeness Boolean value, such as... Figure 4 As shown, the signature verification is performed by calling the VL model; the key field extraction model extracts the effective start and end dates of the purchase order and generates key-value pairs through optical character recognition and rule engine, that is, extracting the date field through keyword matching; the keyword judgment model scans the entire text according to the preset rule library and outputs the related company clause markers, that is, extracting the related company clauses through keyword matching; the above parallel analysis results are integrated and encapsulated in real time to form a structured JSON data object containing all the perception results.

[0079] Step S303, System Integration and Business Logic Verification: The interface communication module actively and securely pushes the structured JSON data object to the backend of the relevant core business system through a predefined API interface, triggering the review module to automatically start the review.

[0080] The interface communication module 20 encapsulates the recognition results of the intelligent recognition module into a JSON data object, and calls the API interface of the business system to actively push the JSON data object to the business system. After receiving the JSON data object, the business system binds it with the corresponding order recorded in the business system and automatically triggers the review module 30 to perform data consistency verification logic.

[0081] The review module 30 parses the recognition results of the intelligent recognition module 10 in the JSON, such as... Figure 4 As shown, the review module 30 retrieves manually entered data of purchase orders or contracts from the business system database, automatically compares the manually entered data of the corresponding orders in the business database with the recognition results and verifies them according to business rules, thereby performing data consistency verification. Based on the fusion of signature status, risk markers and data consistency results, the final comprehensive review conclusion is generated and persistently stored in the business database.

[0082] Step S304, interactive comparison presentation and closed-loop business processing: The core business system integrates a display module into its user interface to clearly present the results extracted by artificial intelligence, the data stored in the system, and the data verification conclusions in a parallel comparison form, thereby completing the closed loop from intelligent review to business data update.

[0083] The intelligent review results will be compared and displayed in the business system, such as... Figure 4 As shown, the display module 50 interacts with the application server to obtain the verification results from the review module and displays the recognition results of the intelligent recognition module 10 and the intelligent review conclusion of the data entered into the business system.

[0084] The automatic review method for business systems provided in this embodiment has the following effects: A highly cohesive, loosely coupled system architecture with decoupled "perception-decision-presentation" layers was constructed. The solution innovatively clarifies the technical boundaries: the front-end AI engine (intelligent recognition module) focuses on perception and understanding in uncertain environments (CV and NLP); the back-end business system focuses on deterministic logic and data management; and the presentation layer (display module) focuses on user experience and interaction. Each layer communicates through standardized interfaces (structured data and APIs), enabling independent iteration and upgrades of AI models, changes to business rules, and optimization of the front-end UI, greatly improving the system's maintainability, scalability, and technological evolution capabilities. It has achieved automation and efficiency improvement in the review process: Through the technical chain of "multimodal perception - structured extraction - system-triggered verification", it has systematically solved the gap in automation processing between unstructured PDF orders and structured business systems, compressing the time-consuming manual sequential review to the second level, improving efficiency by hundreds of times; An automated reverse verification mechanism was built to ensure the data quality of business systems: using the original PDF information extracted by AI as a benchmark, the business system is driven to automatically verify the accuracy of its internal manually entered data, turning passive error correction into proactive error prevention; This has created a decision-making closed loop that deeply integrates intelligent review with core business processes: Through standardized data interfaces, the structured review results of AI are seamlessly pushed and embedded into the display module (i.e., operation interface) of the business system. This interface is presented as a comparative visualization page and provides direct operation entry points such as "confirm whether the data is accurate". This enables the discovery of intelligent review to be transformed into executable business instructions in an instant, realizing a business closed loop from "risk identification" to "risk handling", which greatly improves operational compliance and decision-making efficiency. It provides an enterprise-grade, scalable, high-performance support architecture: a cloud-native architecture based on multi-container clusters and asynchronous messaging, which enables the system to have high concurrency and batch processing capabilities, and can stably support enterprise-level application scenarios.

[0085] This invention provides a computer-readable medium storing computer-executable instructions. When these computer-executable instructions are invoked and executed by a processor, they cause the processor to implement the methods described in the above embodiments.

[0086] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the foregoing embodiments, and will not be repeated here.

[0087] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0088] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0089] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0090] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An intelligent review device for a business system, characterized in that, include: The system includes an intelligent identification module, an interface communication module, and an inspection module. The inspection module is located within the business system, and the interface communication module communicates with the inspection module within the business system through a preset interface. The intelligent recognition module is used to identify key information in the document to be recognized and output the recognition result; wherein the data type of the recognition result is structured data; The interface communication module is used to encapsulate the recognition result into a data packet of a preset format, and transmit the data packet of the preset format to the review module in the business system through the preset interface; The review module is used to initiate a business system review when the data packet of the preset format is received, and to review whether the key information input data corresponding to the document to be identified pre-stored in the business system is consistent with the data packet of the preset format.

2. The intelligent review device for a business system according to claim 1, characterized in that, Also includes: Preprocessing module; The preprocessing module is used to convert the document to be recognized into a standard document image in a preset format, and input the standard document image into the intelligent recognition module to identify key information of the document to be recognized.

3. The intelligent review device for a business system according to claim 1, characterized in that, The intelligent recognition module includes an image recognition unit and a keyword recognition unit; the key information includes preset patterns and keywords. The image recognition unit is used to recognize the preset pattern in the document to be recognized and output the recognition result of the preset pattern; The keyword recognition unit is used to locate and recognize keywords in the document to be recognized, and output the keyword recognition results.

4. The intelligent review device for a business system according to claim 3, characterized in that, The keyword recognition unit includes a key field extraction model and a keyword judgment model; the keywords include predefined key fields and preset keywords. The image recognition unit is used to identify whether the document to be recognized contains a preset pattern, and outputs the structured determination result of the preset pattern; The key field extraction model is used to perform character recognition on the document to be identified to obtain the full text of the document to be identified, locate and extract predefined key fields from the full text, and output the structured data of the predefined key fields. The keyword judgment model is used to perform full-text scanning of the full text of the document to be identified based on preset keywords, determine whether the full text contains the preset keywords, and output the structured judgment result of the preset keywords.

5. The intelligent review device for a business system according to claim 4, characterized in that, The document to be identified includes contract documents or purchase order documents; The preset pattern includes a valid signature pattern, the predefined key fields include the contract start date and the contract end date, and the preset keywords include the set clause name.

6. The intelligent review device for a business system according to claim 4, characterized in that, The review module is also used to parse the structured judgment result of the preset pattern, the structured data of the key field, and the structured judgment result of the preset keyword from the data packet of the preset format, and to query the key information input data corresponding to the document to be identified from the database of the business system, and to perform consistency verification on the parsing result of the data packet of the preset format and the key information input data.

7. The intelligent review device for a business system according to claim 1, characterized in that, The review module is also used to generate an intelligent review conclusion for the document to be identified after reviewing whether the key information input data corresponding to the document to be identified pre-stored in the business system is consistent with the parsing result of the data packet in the preset format, and to store the intelligent review conclusion in the database of the business system.

8. The intelligent review device for a business system according to claim 1, characterized in that, Also includes: A display module, which is set within the business system; The display module is used to present the intelligent review conclusions in the form of a comparison table or a structured list.

9. The intelligent review device for a business system according to claim 1, characterized in that, Also includes: The message queue and cluster scheduler, the intelligent recognition module includes multiple parallel AI models; The message queue is used to receive multiple document recognition tasks; The cluster scheduler is used to assign multiple document recognition tasks to multiple parallel AI models for key information recognition.

10. An intelligent review method for a business system, characterized in that, The intelligent review apparatus applied to the business system according to any one of claims 1-9, wherein the intelligent review method of the business system includes: The intelligent recognition module identifies key information in the document to be recognized and outputs the recognition results; wherein the data type of the recognition results is structured data. The identification result is encapsulated into a data packet of a preset format based on the interface communication module, and the data packet of the preset format is transmitted to the review module through the preset interface; Upon receiving the data packet in the preset format, the business system initiates a review, and the review module examines whether the key information input data corresponding to the document to be identified, which is pre-stored in the business system, is consistent with the data packet in the preset format.

Images