Intelligent order checking method, system and device based on artificial intelligence and medium

Abstract

The invention belongs to the technical field of order examination, and discloses an intelligent order examination method, system and device based on artificial intelligence and a medium, and the method comprises the steps: obtaining unstructured business data; performing information extraction and conversion on the unstructured business data based on a pre-trained small model to obtain structured text data; identifying a pre-made approval system file, generating an approval rule, and constructing a rule engine based on the approval rule; performing preliminary compliance auditing on the structured text data based on a rule engine to obtain a first auditing result; finding out associated knowledge features associated with the text data based on the structuralization; performing semantic analysis on the structured text data based on a pre-trained large model to obtain semantic features; comparing and auditing the associated knowledge features and the semantic features to obtain a second auditing result; and generating an audit report based on the first audit result and the second audit result. According to the invention, the accuracy and consistency of auditing are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of document examination, and specifically relates to an intelligent document examination method, system, equipment and medium based on artificial intelligence. Background Art

[0002] In the field of intelligent document review systems, existing technologies mainly rely on traditional rule engines and manual review processes. These systems usually use preset business rules to automatically check whether the submitted documents meet the established standards, but in actual business, they can only handle simple logical judgments and cannot handle complex background review; for example: 1. Judgment can only be made based on structured information: For example, it can support the judgment of whether the invoice amount is equal to the reimbursement amount, but it cannot judge whether the place where the invoice is issued is consistent with the business trip itinerary; 2. Judgment can only be made based on the company's internal data: For example, it can only support the judgment whether the invoice amount of the computer is consistent with the reimbursement amount, but cannot judge the specific product type in the invoice service content, such as whether "Lenovo Savior Y9000P" is a laptop; 3. The judgment logic is hard-coded, making it difficult to automatically check and update. Manual checking and updating are costly. For example, for self-driving business trip reimbursement, the original logic only applies to gasoline vehicles. After the emergence of electric vehicles, the reimbursement system had to be adjusted, so the code had to be modified to add a calculation method for the reimbursement standard for electric vehicles. 4. It is difficult to process data generated by non-enterprise internal systems: For example, when applying for reimbursement for self-driving business trips, employees are generally required to attach a screenshot of the mileage recorded by the mobile phone navigation system. However, traditional systems cannot recognize the total mileage in the image and must rely on manual calculation by financial staff. 5. It is difficult to make judgments based on common sense: For example, the reimbursement system requires that employees can only stay in national chain hotels when on business trips. Traditional systems find it difficult to determine whether the hotel the employee is staying in belongs to a national chain hotel.

[0003] In summary, traditional document review systems often rely too much on data from the company's internal information system, lack flexibility and adaptability, are unable to cope with complex and changing business scenarios, and are heavily dependent on complex manpower input. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent document review method, system, device and medium based on artificial intelligence, so as to solve the problem that the traditional document review system is often too dependent on the data of the internal information system of the enterprise, lacks flexibility and adaptability, has difficulty in coping with complex and changing business scenarios, and is also heavily dependent on complicated manpower input.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides an intelligent document examination method based on artificial intelligence, the method comprising: Obtain unstructured business data; Extract and transform unstructured business data based on pre-trained small models to obtain structured text data; Identify pre-established approval system documents, generate approval rules, and build a rule engine based on the approval rules; Conduct a preliminary compliance review of structured text data based on the rule engine to obtain the first review result; Find relevant knowledge features associated with structured text data from the pre-built knowledge base; Perform semantic analysis on structured text data based on a pre-trained large model to obtain semantic features; Compare and review the associated knowledge features and the semantic features to obtain a second review result; Based on the first audit result and the second audit result, a document audit report is generated and the document audit report is visualized.

[0006] Preferably, the knowledge base includes at least: an industry knowledge base and an external knowledge base.

[0007] Preferably, the small model includes at least: a text recognition model, a layout analysis model, a bill classification model and an information extraction model.

[0008] Preferably, the unstructured business data is extracted and transformed based on the pre-trained small model to obtain structured text data, including: Perform text recognition on unstructured business data based on the text recognition model to obtain text content; Perform layout analysis on unstructured business data based on the layout analysis model to obtain content elements; Based on the bill classification model, the unstructured business data is classified and the bill type is obtained; Extract key information from text content based on the information extraction model to obtain specific key information; Build structured text data based on text content, content elements, bill type, and specific key information.

[0009] Preferably, the method further comprises: constructing a first training data set, comprising: Acquire the first historical business data, and perform data labeling on the first historical business data to obtain structured data with labels; A first training data set is constructed based on structured data with labels, where the first training data set is used to train a small model.

[0010] Preferably, the method further includes: creating synthetic data and synthetic labels corresponding to the synthetic data based on the large model, using the synthetic data and the synthetic labels corresponding to the synthetic data as new training data, and adding the new training data to the first training data set.

[0011] Preferably, the method further comprises: constructing a second training data set, comprising: Acquire second historical business data, input the second historical business data into a pre-trained small model, and the small model outputs sample data; Conduct compliance review of sample data based on the rule engine to obtain sample review results; A second training data set is constructed based on the sample data and the sample audit results, and the second training data set is used to train the large model.

[0012] In a second aspect, the present invention provides an intelligent document review system based on artificial intelligence, which is used to implement the above-mentioned intelligent document review method based on artificial intelligence, and the system includes: Data acquisition module, used to acquire unstructured business data; The data conversion module is used to extract and convert unstructured business data based on a pre-trained small model to obtain structured text data; The rule building module is used to identify the pre-established approval system documents, generate approval rules, and build a rule engine based on the approval rules; A first review module, used to perform a preliminary compliance review on the structured text data based on a rule engine to obtain a first review result; A knowledge search module is used to search for associated knowledge features associated with structured text data from a pre-built knowledge base; The semantic analysis module is used to perform semantic analysis on structured text data based on a pre-trained large model to obtain semantic features; The second review module is used to compare and review the associated knowledge features with the semantic features to obtain a second review result; The report generation module is used to generate a document review report based on the first review result and the second review result, and to visually display the document review report.

[0013] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned intelligent document review method based on artificial intelligence when executing the computer program.

[0014] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned intelligent order review method based on artificial intelligence.

[0015] Beneficial effects: 1. The present invention no longer relies on structured information, and can make judgments based on various types of data (including but not limited to pictures, videos, PDFs, etc.); 2. The present invention greatly reduces the document processing time and improves the overall work efficiency by automating the review process; 3. The present invention reduces the reliance on manual review, thereby reducing labor costs and related operating expenses; 4. The present invention utilizes artificial intelligence technologies such as small models and large models to improve the accuracy and consistency of the audit. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings: Figure 1 It is a flow chart of an intelligent document examination method based on artificial intelligence provided by one embodiment of the present invention; Figure 2 It is a block diagram of an intelligent document review system based on artificial intelligence provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0018] Embodiment 1 Figure 1 FIG. 1 is a flow chart of an intelligent document examination method based on artificial intelligence provided by an embodiment of the present invention. Figure 1 As shown, this embodiment provides an intelligent document review method based on artificial intelligence, which runs on a document review server and completes the following steps on the document review server; therefore, the method includes: Step S10: Obtain unstructured business data. The business data mainly includes: invoices, receipts, expense reports and contracts. The types of these data can be pictures, videos, word or PDF, etc. The above business data can be uploaded to the review server through the application on the mobile phone or PC, so that the review server can obtain the relevant unstructured business data.

[0019] Step S20: extracting and transforming information from unstructured business data based on the pre-trained small model to obtain structured text data; In this embodiment, the small model at least includes: a text recognition model, a layout analysis model, a bill classification model and an information extraction model; therefore, the unstructured business data is extracted and transformed based on the pre-trained small model to obtain structured text data, including: Step S201: Perform text recognition on unstructured business data based on a text recognition model to obtain text content; wherein the text recognition model may be an OCR (Optical Character Recognition) model or a visual model, and the OCR model is used to recognize text content in images or scans and convert them into editable text; through training, the OCR model learns how to accurately locate text areas in images, including recognizing text in various fonts, font sizes, and colors, and correctly locating text in complex backgrounds or blurred situations; the visual model can understand and analyze information in images, not limited to text, but also including graphics and other visual elements; for documents involving images (such as map screenshots), specific information is extracted through the visual model, such as calculating the kilometers in multiple screenshots.

[0020] Step S202: Perform layout analysis on unstructured business data based on the layout analysis model to obtain content elements; layout analysis is to identify the layout structure of the document, that is, the content elements, including the position and relationship of elements such as paragraphs, titles, tables, and pictures. The document structure is identified through layout analysis technology, and key information areas (such as invoice number, amount, date, table, etc.) are distinguished to ensure the accuracy of data extraction.

[0021] Step S203: Based on the bill classification model, the unstructured business data is categorized and extracted to obtain the bill type; the bill classification model is used to automatically identify and classify different types of bills and documents, such as invoices, receipts, expense reports and contracts; through training, the bill classification model learns to extract key features from the content and layout features of the bill to distinguish different types of bills.

[0022] Step S204: extract key information from the text content based on the information extraction model to obtain specific key information; the information extraction model is used to extract specific key information from the text content, such as amount, date, invoice number and supplier name, etc.; through training, the information extraction model learns to recognize important entities in the text, understand the context of the text, distinguish similar fields, and identify the relationship between entities.

[0023] Step S205: construct structured text data according to the text content, content elements, bill type and specific key information.

[0024] Among them, structured text data is stored in the database to ensure data security and query efficiency; the database has the following functions: Structured storage: Store text data in a fixed format (such as a table), including clear fields and data types; through structured storage, data is more standardized and easier to query and analyze. For example, store invoice information in a table in a database, and each record contains fields such as invoice number, amount, date, etc. Data table design: Design corresponding data tables for different types of text data to ensure data standardization, such as invoice information tables including invoice number, amount, date and other fields; Indexing and query optimization: Creating indexes for commonly used query fields is like the table of contents of a book, which enables quick location of required information. Indexing can significantly increase the speed of database queries, ensuring that the system can quickly respond to user query requests. Data security and backup: Take measures to protect data from unauthorized access or loss, including data encryption, access control and regular backup; ensure that stored data is safe and reliable, prevent data loss due to system failure or accidents, and ensure business continuity.

[0025] In this embodiment, efficient information retrieval is supported through optimized database query technology: (1) Keyword search: Users can search for relevant records using keywords (such as invoice number, invoice type, etc.), allowing users to quickly find the information they need without having to browse through a large number of records; (2) SQL query optimization: Use optimized database query statements and strategies to improve query efficiency, optimize the query process in the background, speed up data retrieval, and enhance user experience; (3) Multi-condition combination query: Supports multi-condition combination query, such as invoice information within a specific time period, to quickly filter out the required data; (4) Fast return and matching: When the amount of data is large, the matching results are returned quickly.

[0026] Step S30: Identify the pre-formulated approval system document, generate approval rules, and build a rule engine based on the approval rules; in this embodiment, the pre-formulated approval system document is an internal approval rule document of the company, which is pre-formulated by the enterprise, and the approval system document contains a variety of approval rules; a pre-trained big model can also be used to identify the pre-formulated approval system document, and the big model can automatically read and understand the approval system document, and then generate approval rules that meet business needs.

[0027] Step S40: Based on the rule engine, a preliminary compliance review is conducted on the structured text data to obtain the first review result; in this embodiment, the rule engine checks whether the text data meets the requirements according to the corresponding approval rules, such as whether the amount, date, and reimbursement items are compliant. The rule engine can only determine whether simple data is compliant, and cannot determine more details, such as whether the location and itinerary on the invoice are consistent, and whether the contract terms comply with regulations; in order to improve the accuracy of the compliance review of the details of the text data, steps S50 to S70 are used for review.

[0028] Step S50: Find out the associated knowledge features associated with the structured text data from the pre-built knowledge base; the pre-built knowledge base includes at least: industry knowledge base and external knowledge base, wherein the industry knowledge base contains relevant policies and regulations, industry standards, company rules and regulations, etc. For example, the system can determine whether the hotel where the employee stays belongs to a national chain hotel by querying the hotel information database; it can connect to external knowledge bases, such as search engines such as Baidu and Google, to obtain more extensive information; for example, the system can determine whether the products in the invoice service content are within the scope of reimbursement through search engines; and the knowledge base also supports users to upload and maintain custom knowledge bases, Step S60: semantic analysis is performed on the structured text data based on the pre-trained large model to obtain semantic features; wherein the large model adopts a large language model (Large Language Model, LLM for short).

[0029] Step S70: Compare and review the associated knowledge features with the semantic features to obtain a second review result; in this embodiment, the semantic understanding technology of the large language model is used to understand the text description on the invoice and compare it with the itinerary information to determine whether the location and itinerary are consistent.

[0030] This example takes the case where an employee's business trip reimbursement is required and it is necessary to determine whether the location and itinerary on the invoice are consistent: First, extract key information from invoices and itineraries, such as invoice location, travel date, and itinerary destination; Use the approval rules corresponding to the rule engine to determine whether the invoice date is within the travel date range. The judgment result here is used as the first review result; Then, by connecting to the external knowledge base, querying the map database, obtain the latitude and longitude information of the invoice location and the itinerary destination; Use natural language processing technology to understand the text description in the itinerary, such as "departing from Beijing to Shanghai"; Compare the text description in the itinerary with the map information to determine whether the itinerary passes through the invoice location. The judgment result at this time is used as the second review result.

[0031] Step S80: Generate a document review report based on the first review result and the second review result, and visualize the document review report.

[0032] In this embodiment, after two audits, an audit report is automatically generated, which includes the basis for decision-making, compliance analysis, etc., which is convenient for auditors and management to review and refer to.

[0033] The present invention combines the capabilities of a large language model with a rule engine to form an intelligent and accurate review mechanism. This combination can not only perform basic rule judgments, but also handle complex approval scenarios, provide in-depth analysis and decision support, thereby significantly improving approval efficiency and accuracy.

[0034] In this embodiment, continuous optimization and feedback learning of the large language model and rule engine are also included: (1) Continuous improvement: The system will be continuously optimized based on user feedback and historical data. Through continuous learning and adjustment, the rule engine can improve the efficiency and accuracy of approval and adapt to changes in business needs; (2) Adaptability to new scenarios: The large language model is flexible and can adapt to different types of review tasks. Whether it is a new document type or a new approval rule, the system can handle it efficiently and accurately.

[0035] As a further optimization of this embodiment, the method also includes: for training the large model and the small model, it is necessary to construct a first training data set and a second training data set, the first training data set is used to train the small model, and the second training data set is used to train the large model.

[0036] In this embodiment, a model training and data management platform is built on the order review server. The model training and data management platform supports the collaborative training and optimization of large and small models to ensure the efficient operation of various models. During the training process, the platform can automatically adjust the model parameters and optimization strategies to achieve the best training effect. In addition, the platform also provides a wealth of data management and visualization tools to help users process and analyze data more conveniently. Through these tools, users can more intuitively understand the model training progress and performance, thereby optimizing the model more effectively.

[0037] The model training and data management platform includes the following functions: Data management and labeling: Data is the basis of model training. The platform supports the processing and labeling of multiple types of data through effective knowledge base management and data labeling tools.

[0038] (1) Knowledge base management: The platform has a built-in industry knowledge base that contains commonly used rules, policies, and expertise. It also supports users to upload and maintain custom knowledge bases. By providing rich learning materials, the model can master the latest business rules and policies, improve its ability to handle complex tasks, and ensure accurate support in various scenarios.

[0039] (2) Data labeling: The platform provides data labeling tools to help users label data, such as marking key information such as amounts and dates in documents. High-quality labeled data significantly improves the accuracy of model training, enabling the model to more accurately identify and extract required information, reduce error rates, and improve overall performance.

[0040] In the process of training the large model and the small model, since the small model is trained for specific tasks, such as OCR recognition, layout analysis, bill classification, etc., the task processing efficiency is improved; therefore, the method further includes: constructing a first training data set, including: Step a10: Acquire the first historical business data, perform data annotation on the first historical business data, and obtain structured data with labels; the first historical business data is the company's historical approval data, and the data annotation of the first historical business data is completed through the data annotation tool provided by the platform; Step a20: construct a first training data set based on the structured data with labels, where the first training data set is used to train the small model.

[0041] In this embodiment, the large model is mainly responsible for processing complex tasks, such as rule parsing and text summarization; therefore, the method further includes: constructing a second training data set, including: Step b10: obtaining second historical business data, and inputting the second historical business data into a pre-trained small model, wherein the small model outputs sample data; Step b20: Perform compliance review on the sample data based on the rule engine to obtain sample review results; Step b30: construct a second training data set based on the sample data and the sample review results, wherein the second training data set is used to train the large model.

[0042] In this embodiment, the output results and historical data of the rule engine are used to train the big model so that it can better understand complex document review scenarios. The big model improves its analysis and judgment capabilities of documents and enhances the accuracy and efficiency of the review by learning the judgment logic of the rule engine and the experience of human review.

[0043] Secondly, the big model can be fine-tuned through user feedback and new data to ensure that it continues to adapt to changes in document review needs; when business rules are updated or new types of documents appear, the big model can be fine-tuned in a timely manner to maintain optimal performance and accurately handle the latest document review tasks.

[0044] In this embodiment, the large model and the small model can be collaboratively trained and optimized as follows: (1) Data support: Small models can quickly process large amounts of data, filter and preprocess high-quality data for use in large model training; this reduces the burden of data processing on large models, allowing them to focus on learning complex tasks and improve training efficiency and effectiveness.

[0045] (2) Generation assistance: Large models have powerful generation capabilities and can create high-quality synthetic data or automatically generate synthetic labels corresponding to data to assist in the training of small models. This provides rich training resources for small models, especially when training data is insufficient, improving their performance on specific tasks. (3) Joint optimization: Through collaborative training, the small model and the large model cooperate with each other, learn from each other, and optimize the overall performance. This joint optimization ensures that the system can operate efficiently in various tasks, giving full play to their respective advantages and achieving better processing results.

[0046] Therefore, the artificial intelligence (AI)-based document review of the present invention can break through the technical barriers of the traditional document review system and truly realize unattended intelligent document review. For traditional intelligent document review, the AI ​​intelligent document review system can handle it perfectly.

[0047] 1. No longer rely on structured information, judgments can be made based on various types of data (including but not limited to pictures, videos, PDFs, etc.): For example, it can not only support the judgment of whether the invoice amount is equal to the reimbursement amount, but also judge whether the place where the invoice is issued is consistent with the business trip itinerary by comparing the place where the invoice is issued with the place where the business trip is. 2. Don’t just make judgments based on internal company data: For example, it can not only support the judgment of whether the invoice amount of the computer is consistent with the reimbursement amount, but also judge through the knowledge base and online search engine that the "Lenovo Legion Y9000P" in the invoice service content is a laptop computer and meets the reimbursement requirements; 3. Judgment logic can be described verbally without writing code to change the system: For example, the reimbursement system had to be adjusted after the emergence of electric vehicles. For the review of the reimbursement system for electric vehicles, it is only necessary to upload the updated reimbursement system text to AI, without modifying the code or adjusting the process; 4. Perfect support for data generated by non-enterprise internal systems: For example, when applying for reimbursement for a self-driving business trip, employees are generally required to attach a screenshot of the mileage displayed on their mobile phone navigation system. AI can then identify the mileage in the image and calculate the allowable reimbursement amount. It can even automatically obtain the mileage based on the departure and destination in the navigation software without the need for employees to upload screenshots. 5. Good at making judgments based on common sense: For example, if the reimbursement system requires that employees on business trips can only stay in national chain hotels, the AI ​​intelligent order review method can accurately determine whether the hotel the employee is staying in is a national chain hotel by calling the knowledge base, search engine, etc.

[0048] In summary, AI intelligent order review methods can solve problems that traditional order review software cannot solve. Unattended AI intelligent order review can be achieved through AI tools, which can bring unlimited improvements to the operation of enterprises: 1. Improve efficiency: By automating the review process, the document processing time is greatly reduced and the overall work efficiency is improved; 2. Reduce costs: Reduce reliance on manual review, thereby reducing labor costs and related operating expenses; 3. Enhanced accuracy: Using AI technology for pattern recognition and risk prediction to improve the accuracy and consistency of audits; 4. Flexibility and scalability: The system design takes into account the needs of different business scenarios and supports flexible configuration and expansion to adapt to the ever-changing market environment; 5. Security and compliance: Ensure that the audit process complies with relevant laws, regulations, industry standards and system requirements to reduce compliance risks.

[0049] Embodiment 2 Figure 2 FIG. 1 is a block diagram of an intelligent document review system based on artificial intelligence provided by an embodiment of the present invention. Figure 2 As shown, this embodiment provides an intelligent document review system based on artificial intelligence, which is used to implement the intelligent document review method based on artificial intelligence in Embodiment 1. The system includes: Data acquisition module, used to acquire unstructured business data; The data conversion module is used to extract and convert unstructured business data based on a pre-trained small model to obtain structured text data; The rule building module is used to identify the pre-established approval system documents, generate approval rules, and build a rule engine based on the approval rules; A first review module, used to perform a preliminary compliance review on the structured text data based on a rule engine to obtain a first review result; A knowledge search module is used to search for associated knowledge features associated with structured text data from a pre-built knowledge base; The semantic analysis module is used to perform semantic analysis on structured text data based on a pre-trained large model to obtain semantic features; The second review module is used to compare and review the associated knowledge features with the semantic features to obtain a second review result; The report generation module is used to generate a document review report based on the first review result and the second review result, and to visually display the document review report.

[0050] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the intelligent document review method based on artificial intelligence in Embodiment 1 is implemented.

[0051] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the intelligent document review method based on artificial intelligence in Embodiment 1 is implemented.

[0052] The present invention no longer relies on structured information and can make judgments based on various types of data (including but not limited to pictures, videos, PDFs, etc.); through automated review processes, it greatly reduces document processing time and improves overall work efficiency; reduces reliance on manual review, thereby reducing labor costs and related operating expenses; and uses artificial intelligence technologies such as small models and large models to improve the accuracy and consistency of review.

[0053] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0054] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A system that specifies the functions of a box or multiple boxes.

[0055] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. An intelligent bill review method based on artificial intelligence, characterized in that: The method comprises: Obtain unstructured business data; Extract and transform unstructured business data based on pre-trained small models to obtain structured text data; Identify pre-established approval system documents, generate approval rules, and build a rule engine based on the approval rules; Conduct a preliminary compliance review of structured text data based on the rule engine to obtain the first review result; Find relevant knowledge features associated with structured text data from the pre-built knowledge base; Perform semantic analysis on structured text data based on a pre-trained large model to obtain semantic features; Compare and review the associated knowledge features and the semantic features to obtain a second review result; Based on the first audit result and the second audit result, a document audit report is generated and the document audit report is visualized.

2. The intelligent document examination method based on artificial intelligence according to claim 1 is characterized in that: The knowledge base includes at least: an industry knowledge base and an external knowledge base.

3. The intelligent document examination method based on artificial intelligence according to claim 1 is characterized in that: The small model at least includes: a text recognition model, a layout analysis model, a bill classification model and an information extraction model.

4. The intelligent document examination method based on artificial intelligence according to claim 3 is characterized in that: Based on the pre-trained small model, unstructured business data is extracted and converted to obtain structured text data, including: Perform text recognition on unstructured business data based on the text recognition model to obtain text content; Perform layout analysis on unstructured business data based on the layout analysis model to obtain content elements; Based on the bill classification model, the unstructured business data is classified and the bill type is obtained; Extract key information from text content based on the information extraction model to obtain specific key information; Build structured text data based on text content, content elements, bill type, and specific key information.

5. The intelligent document examination method based on artificial intelligence according to claim 1 is characterized in that: The method further comprises: constructing a first training data set, comprising: Acquire the first historical business data, and perform data labeling on the first historical business data to obtain structured data with labels; A first training data set is constructed based on structured data with labels, where the first training data set is used to train a small model.

6. The intelligent document examination method based on artificial intelligence according to claim 5 is characterized in that: The method also includes: creating synthetic data and synthetic labels corresponding to the synthetic data based on the large model, using the synthetic data and the synthetic labels corresponding to the synthetic data as new training data, and adding the new training data to the first training data set.

7. The intelligent document examination method based on artificial intelligence according to claim 1 is characterized in that: The method further includes: constructing a second training data set, including: Acquire second historical business data, input the second historical business data into a pre-trained small model, and the small model outputs sample data; Conduct compliance review of sample data based on the rule engine to obtain sample review results; A second training data set is constructed based on the sample data and the sample audit results, and the second training data set is used to train the large model.

8. An intelligent document examination system based on artificial intelligence, used to implement the intelligent document examination method based on artificial intelligence described in any one of claims 1 to 7, characterized in that: The system comprises: Data acquisition module, used to acquire unstructured business data; The data conversion module is used to extract and convert unstructured business data based on a pre-trained small model to obtain structured text data; The rule building module is used to identify the pre-established approval system documents, generate approval rules, and build a rule engine based on the approval rules; A first review module, used to perform a preliminary compliance review on the structured text data based on a rule engine to obtain a first review result; A knowledge search module is used to search for associated knowledge features associated with structured text data from a pre-built knowledge base; The semantic analysis module is used to perform semantic analysis on structured text data based on a pre-trained large model to obtain semantic features; The second review module is used to compare and review the associated knowledge features with the semantic features to obtain a second review result; The report generation module is used to generate a document review report based on the first review result and the second review result, and to visually display the document review report.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the intelligent document review method based on artificial intelligence described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the intelligent document review method based on artificial intelligence described in any one of claims 1 to 7 is implemented.

Images