A hospital internal contract and project intelligent auditing method and system

Abstract

This invention discloses a method and system for intelligent auditing of internal hospital contracts and projects. The method involves inputting documents to be reviewed into a multi-agent collaborative network model within a large audit domain model. A perception agent converts the document formats of multiple documents to be reviewed, resulting in multiple text files. Intent recognition is performed based on the documents to be reviewed, the audit scenario, and the audit type to obtain a list of audit rules. Based on the audit rule list, the content to be reviewed is extracted, document-sliced, and assembled to obtain atomic files corresponding to each audit rule. An analysis agent performs risk assessment on the atomic files corresponding to each audit rule, obtaining the corresponding atomic results. A summarizing agent then summarizes these results to obtain a risk warning list. Based on the risk warning list, an audit opinion is derived. This method improves the efficiency of audit results by processing documents of different formats through a tiered task decomposition and by integrating the processing results of various agents.

Description

Technical Field

[0001] This invention relates to the field of intelligent recognition model application technology, and in particular to a method, system, equipment and medium for intelligent auditing of internal contracts in hospitals. Background Technology

[0002] As hospital operations become increasingly complex and diversified, internal auditing, as a crucial means for hospitals to strengthen internal management and supervision, is playing an increasingly important role, especially in the compliance auditing of contracts and procurement. Current internal contract and procurement audits involve two scenarios: in-process auditing, where other functional departments typically submit contracts along with attachments, including requirement documents, project initiation documents, tender documents, bid documents, and award notices, and auditors conduct compliance audits based on these materials; and post-project special auditing, which audits the compliance of projects throughout the entire process from project initiation and procurement to acceptance, involving the verification and comparison of as many as 10-20 documents. Auditors often have to manually check each document one by one, and this mechanical and repetitive work consumes a significant amount of their time and energy. Furthermore, because audit documents are often unstructured long texts, even in different formats (tables, images, PDFs, etc.), and with diverse writing styles, previous technologies such as RPA robots and deep learning-based methods have struggled to achieve satisfactory results. Summary of the Invention

[0003] To address the aforementioned technical problems, embodiments of the present invention provide a method, system, device, and medium for intelligent auditing of internal hospital contracts, thereby resolving the technical issue of low audit efficiency caused by the inability of existing technologies to comprehensively analyze and understand internal hospital contracts in various formats.

[0004] A first aspect of this invention provides a method for intelligent auditing of internal contracts in hospitals, the method comprising:

[0005] Obtain multiple documents to be reviewed, including contract documents or project documents;

[0006] The documents to be reviewed are input into a multi-agent collaborative network model based on a large audit domain model. The perceptual agent is used to convert the document format of multiple documents to be reviewed, resulting in multiple text files. The intent of the documents to be reviewed is recognized to obtain a list of audit rules. Based on the list of audit rules, the position of the content to be reviewed corresponding to each audit rule is determined. Based on the position of each content to be reviewed, the content to be reviewed is extracted and the document is sliced ​​to obtain multiple sliced ​​documents. The multiple sliced ​​documents are assembled to obtain the atomic file corresponding to each audit rule.

[0007] The analytical agent performs risk assessment on the atomic files corresponding to each audit rule, and after obtaining the atomic results for each audit rule, the summarizing agent summarizes the atomic results to obtain a risk warning list. Based on the risk warning list, the atomic results with risk points are summarized to obtain audit opinions.

[0008] In one possible implementation of the first aspect, a perceptual agent is used to convert the document formats of multiple documents to be reviewed, resulting in multiple text files, including:

[0009] Using PDF parsing and conversion technology, each page of multiple documents to be reviewed is converted into an image, resulting in the corresponding converted file.

[0010] Based on the converted files, multimodal OCR technology is used to locate and extract computer text and handwritten text, obtaining the corresponding extraction results and text coordinates. The corresponding extraction results are then embedded into the corresponding files to be reviewed, resulting in a text file corresponding to each file to be reviewed.

[0011] In one possible implementation of the first aspect, after obtaining multiple text files, it further includes:

[0012] Based on punctuation and grammar rules, intelligent sentence segmentation is performed on each text file to obtain multiple segmented text files.

[0013] Correct erroneous and missing characters in each sentence of each segmented text file, and assign a location identifier.

[0014] In one possible implementation of the first aspect, multiple sliced ​​documents are assembled to obtain an atomic file corresponding to each audit rule, including:

[0015] The multiple sliced ​​documents corresponding to each audit rule are assembled to obtain the assembled file corresponding to each audit rule;

[0016] The atomic file corresponding to each audit rule is obtained based on the audit rule number, text file number, location identifier, assembly file, and next step task content.

[0017] In one possible implementation of the first aspect, after obtaining the audit opinion, it further includes:

[0018] The risk warning list is encapsulated into a first document using an execution intelligent agent, and the audit opinion is encapsulated into a second document.

[0019] Based on the first and second documents, annotations are made on the documents to be reviewed using their numbers and location identifiers, resulting in annotation results.

[0020] In one possible implementation of the first aspect, the audit vertical domain large model is obtained by pre-training a general large model using audit domain text data, including:

[0021] Acquire textual data in the audit field, including textual data of laws and regulations in the audit field, audit cases, and audit rules;

[0022] NLP technology is used to segment and identify entities in text data in the audit field to obtain key information, and a knowledge graph is constructed based on the entities and relationships between entities in the key information.

[0023] A general large model is selected as the base model. Entity vectors from the knowledge graph are inserted into the base model as key values. An adapter structure is used to isolate knowledge parameters and dynamically adjust the learning rate and mask ratio to obtain the initial audit vertical domain large model.

[0024] Based on the business needs and model optimization goals of the audit vertical domain, the model is divided into multiple tasks. Supervision data and instruction sets for each task are constructed. Based on the supervision data and instruction sets of each task, a reward model is designed using reinforcement learning and an optimization strategy based on the PPO algorithm is used to optimize the initial audit vertical domain model, thus obtaining the audit vertical domain model.

[0025] In one possible implementation of the first aspect, each agent in the multi-agent cooperative network model is obtained by fine-tuning the large audit domain model, including:

[0026] Based on the large-scale auditing model, a multi-agent collaborative network model is constructed by combining prompt word engineering and Python-based engineering functions. The multi-agent collaborative network model includes a perceptual agent, an analytical agent, a summarizing agent, and an executive agent.

[0027] A second aspect of this invention provides an intelligent auditing system for internal hospital contracts and projects, the system comprising:

[0028] The acquisition module is used to respond to the review operation, determine the audit scenario and audit type, and acquire multiple documents to be reviewed based on the audit scenario and audit type. These documents can be contract documents or project documents.

[0029] The conversion module is used to input the documents to be reviewed into a multi-agent collaborative network model based on the large-scale audit domain model. The perceptual agent is used to convert the document format of multiple documents to be reviewed, resulting in multiple text files. The module also performs intent recognition on the documents to be reviewed to obtain a list of audit rules. Based on the list of audit rules, the position of the content to be reviewed corresponding to each audit rule is determined. Based on the position of each content to be reviewed, the content to be reviewed is extracted and the document is sliced ​​to obtain multiple sliced ​​documents. The multiple sliced ​​documents are then assembled to obtain the atomic file corresponding to each audit rule.

[0030] The summary module is used to use the analysis agent to perform risk assessment on the atomic files corresponding to each audit rule. After obtaining the atomic results for each audit rule, the summary agent summarizes the atomic results to obtain a risk warning list. Based on the risk warning list, the atomic results with risk points are summarized to obtain audit opinions.

[0031] A third aspect of the present invention provides a computer device, comprising:

[0032] Memory, used to store computer programs;

[0033] A processor is used to execute computer programs to implement steps such as the intelligent auditing method for hospital internal contracts and projects, as described in the first aspect.

[0034] A fourth aspect of the present invention provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the intelligent auditing method for internal hospital contracts and projects as described in the first aspect.

[0035] The technical solution of this invention has the following advantages:

[0036] The intelligent auditing method for internal hospital contracts provided in this invention determines the audit scenario and audit type in response to the review operation. Multiple documents to be reviewed are obtained based on the audit scenario and audit type. These documents are input into a multi-agent collaborative network model based on a large-scale audit domain model. A perception agent converts the document formats of the multiple documents to be reviewed, resulting in multiple text files. Intent recognition is performed based on the documents to be reviewed, the audit scenario, and the audit type to obtain a list of audit rules. The location of the content to be reviewed corresponding to each audit rule is determined based on the list. Based on the location of each content to be reviewed, the content is extracted and sliced ​​into multiple sliced ​​documents. These sliced ​​documents corresponding to each audit rule are assembled to obtain atomic files corresponding to each audit rule. An analysis agent performs risk assessment on the atomic files corresponding to each audit rule, obtaining atomic results for each audit rule. Finally, a summarizing agent summarizes the atomic results to obtain a risk warning list. Based on the risk warning list, atomic results with risk points are summarized to obtain audit opinions. The above method processes files of different formats by decomposing them into tiered tasks and integrates the processing results of each agent to achieve intelligent auditing and improve the efficiency of audit results. Attached Figure Description

[0037] 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.

[0038] Figure 1 This is a flowchart illustrating the intelligent auditing method for internal hospital contracts and projects in this embodiment of the invention.

[0039] Figure 2 This is a diagram illustrating the functions of the intelligent auditing system for the intelligent auditing method of internal hospital contracts and projects in this embodiment of the invention.

[0040] Figure 3 This is a flowchart of the collaborative network process of the auditing intelligent agent in the intelligent auditing method for internal hospital contracts and projects in this embodiment of the invention;

[0041] Figure 4 This is a system block diagram of the intelligent auditing system for internal hospital contracts and projects in an embodiment of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Please refer to Figure 1 This is a flowchart illustrating one embodiment of the intelligent auditing method for internal hospital contracts and projects provided by the present invention, including steps S101 to S103.

[0044] S101. In response to the review operation, determine the audit scenario and audit type, and obtain multiple documents to be reviewed based on the audit scenario and audit type, wherein the documents to be reviewed are contract documents or project documents.

[0045] In this embodiment, a hospital internal contract and project intelligent auditing system based on an intelligent agent collaborative network is designed and deeply integrated with the hospital's internal operation system and project management contracts to achieve intelligent auditing of hospital internal contracts and projects. Figure 2 As shown, the audit system includes an AI consultation module, an AI contract audit module, an AI project-specific audit module, and AI knowledge base and rule base management modules. AI consultation retrieves relevant cases through natural language queries. The AI ​​then provides answers based on a knowledge base of audit-related laws, regulations, cases, and FAQs, summarizing the relevant materials. Specifically, the intelligent agent receives natural language queries, identifies the intent through a model, and determines the query type. Based on the query type, the intelligent agent retrieves relevant cases, laws, regulations, and FAQs from the audit knowledge base, integrates the search results, generates an answer, and returns it to the user. The system retrieves relevant cases through natural language queries, and the AI ​​provides answers based on a knowledge base of audit-related laws, regulations, cases, and FAQs, summarizing the relevant materials.

[0046] The AI ​​contract audit module is used when bidding and signing contracts outside the institution, which requires contract review. In this process, auditors click the smart audit button, select the documents to be submitted for review and the documents to be compared, and the audit intelligence collaborative network model reviews them and outputs review opinions, an Excel version of the risk review status table, inconsistency comparison results, and an annotated version of the contract review document.

[0047] The AI ​​project-specific audit module allows auditors to click on "intelligent audit" upon project completion. The audit intelligence agent collaborative network model then reviews the project, outputting an Excel-version risk review report and a Word-version review opinion. The AI ​​knowledge base and rule base management module allows auditors to customize their knowledge base and rule base, including information such as regulations, cases, and audit rules.

[0048] AI knowledge base and rule base management: Supports auditors to set up their own knowledge base and rule base, which includes information such as regulations, cases, and audit rules.

[0049] When using the AI ​​contract audit module and AI project audit module for auditing, multiple documents to be reviewed are first obtained and input into the audit agent collaborative network model for review. Information acquisition steps: Through API interfaces, message subscriptions, and other synchronous methods, the following information is obtained: the review scenario (contract audit / special audit), audit type (service, engineering, goods, information system), basic project / contract information (name, ID number), and submitted attachments (original documents). For example: when the review scenario is a contract audit, the audit type is information system, the basic project / contract information is H001 OA system upgrade project contract, and the submitted documents include a docx format contract file and a pdf format tender document.

[0050] When the review scenario is a special audit, the audit type is Project Z002, and the submitted attachments include the project initiation notice in PDF format, the budget allocation notice in JPG format, the tender documents in PDF format, the winning bid notice in DOC format, the contract in PDF format, the advance payment approval form in PDF format, the commencement order in JPG format, and the acceptance report in PDF format.

[0051] It should be noted that the documents to be reviewed refer to contract documents or project documents.

[0052] In one embodiment, a perceptual agent is used to convert the document formats of multiple documents to be reviewed, resulting in multiple text files, including:

[0053] Using PDF parsing and conversion technology, each page of multiple documents to be reviewed is converted into an image, resulting in the corresponding converted file.

[0054] Based on the converted files, multimodal OCR technology is used to locate and extract computer text and handwritten text, obtaining the corresponding extraction results and text coordinates. The corresponding extraction results are then embedded into the corresponding files to be reviewed, resulting in a text file corresponding to each file to be reviewed.

[0055] In this embodiment, firstly, PDF parsing and conversion technology is used to convert each page of multiple documents to be reviewed into images. Then, multimodal OCR technology is used to locate and extract computer text and handwritten text from the converted images. The text is then embedded into the original PDF review document to obtain a converted text file, and the text coordinates are used as the location identifiers for sorting all independent text segments. Using this method, for any PDF or Word document, it is possible to locate and trace the original text's risky locations, reducing the illusion of a large model and making it more convenient for auditors to view and use.

[0056] In one embodiment, after obtaining multiple text files, the method further includes:

[0057] Based on punctuation and grammar rules, intelligent sentence segmentation is performed on each text file to obtain multiple segmented text files.

[0058] Correct erroneous and missing characters in each sentence of each segmented text file, and assign a location identifier.

[0059] In this embodiment, the converted text file undergoes intelligent sentence segmentation. Using Natural Language Processing (NLP) technology, the text is divided into independent sentences based on punctuation and grammatical rules, and any errors or omissions are automatically corrected. Furthermore, to ensure text traceability, each sentence is assigned a unique identifier, i.e., a line number ID, which facilitates subsequent retrieval, comparison, and analysis.

[0060] S102. Input the documents to be reviewed into a multi-agent collaborative network model based on the audit vertical domain large model. Apply the perceptual agent to convert the document format of multiple documents to be reviewed, obtain multiple text files, and perform intent recognition on the documents to be reviewed to obtain a list of audit rules. According to the list of audit rules, determine the position of the content to be reviewed corresponding to each audit rule. Based on the position of each content to be reviewed, extract and slice the content to be reviewed to obtain multiple sliced ​​documents. Assemble the multiple sliced ​​documents to obtain the atomic file corresponding to each audit rule.

[0061] In this embodiment, as Figure 3As shown, the documents to be reviewed are input into the large-scale audit model. A perceptual agent is used to convert the documents into text files. Then, based on the documents to be reviewed, the audit scenario, and the audit type, intent recognition is performed to generate an audit rule list. For each rule, the document containing the content to be reviewed and its location within the document are located. Data extraction and document slicing are performed based on the location, and the files are assembled, outputting the atomic file for that rule {rule ID, document ID, line number ID, assembled file, next task content}. By processing multimodal data more flexibly and combining multi-stage, function-specific agents, complex tasks are broken down and processed, improving information extraction efficiency and achieving a comprehensive understanding and analysis of the content of documents in different formats.

[0062] As an example of this embodiment, based on the input content, the audit intent is identified as a special audit of an engineering project, generating a list of special audit rules for the engineering project, totaling 11 rules. Rule 1 identifies risk points: whether there is procurement and expenditure without a budget. The content to be reviewed includes: the notification time of the budget issuance notice, the contract signing time in the contract, and the approval time in the prepayment approval form. The content to be reviewed in the three documents is extracted to form assembly file 1, generating the next task content: analyze the chronological relationship of the three times, requiring that the contract signing time be later than the budget issuance notice time, and the time of the hospital leader's approval opinion in the prepayment approval form be later than the contract signing time; otherwise, it violates the regulations.

[0063] In one embodiment, the audit vertical domain large model is obtained by pre-training a general large model using audit domain text data, including:

[0064] Acquire textual data in the audit field, including textual data of laws and regulations in the audit field, audit cases, and audit rules;

[0065] NLP technology is used to segment and identify entities in text data in the audit field to obtain key information, and a knowledge graph is constructed based on the entities and relationships between entities in the key information.

[0066] A general large model is selected as the base model. Entity vectors from the knowledge graph are inserted into the base model as key values. An adapter structure is used to isolate knowledge parameters and dynamically adjust the learning rate and mask ratio to obtain the initial audit vertical domain large model.

[0067] Based on the business needs and model optimization goals of the audit vertical domain, the model is divided into multiple tasks. Supervision data and instruction sets for each task are constructed. Based on the supervision data and instruction sets of each task, a reward model is designed using reinforcement learning and an optimization strategy based on the PPO algorithm is used to optimize the initial audit vertical domain model, thus obtaining the audit vertical domain model.

[0068] In this embodiment, the audit vertical domain large model is obtained by pre-training a general large model using text data from the audit domain. The specific pre-training steps are as follows:

[0069] S21: Obtain audit domain text data, which includes audit laws and regulations, audit cases, and audit rules. Utilize NLP technology to segment and identify entities in the audit domain text data to obtain key information, including audit issues, violation types, and audit procedures. Construct a knowledge graph based on the entities and relationships within the key information.

[0070] S22: Select a general large model as the base model, such as llama3, Claude, Wenxin large model 3.5, Ali Tongyi large model, deepseek, doubao pro, etc., and continue pre-training. Insert the knowledge graph entity vectors established in S21 as key-value pairs into the transformer layer of the base model. Use an adapter structure to isolate knowledge parameters and dynamically adjust the learning rate (1e-5) and mask ratio (0.3) to obtain the initial audit vertical domain large model.

[0071] S23: Based on the business needs and model optimization goals of the audit vertical, the task is divided into 3 tasks, and the supervision data and instruction set for each task are constructed.

[0072] 1) Question and answer task: Question and answer pair, such as "instruction: What amount is considered government procurement; output: Goods and services projects are above 1 million yuan, and engineering projects are above 1.2 million yuan".

[0073] 2) Perception and classification task: The combination of document to be reviewed, audit scenario and audit type - audit intent - audit rule list, such as "instruction: Based on the submitted document and scenario, identify the intent and output the rule list; input: contract audit, information system, H001OA system upgrade project contract; output: special audit of information contract, the rule list has 11 items in total."

[0074] 3) Analysis Task: Each rule is transformed into a step-by-step reasoning chain using Whitelist-Guided Question Chain Generation (WGQG) technology, guiding the large model's thinking. "Input: Original contract text slice; Rule: Are the payment terms compliant?; Output 1: Reasoning chain process: What is the contract amount? - Is it a government procurement project (amounts greater than 1 million RMB are considered government procurement projects)? - Are all payment times within 10 working days (government procurement projects require less than 10 working days)?; Output 2: Risk exists, modification is needed."

[0075] S24: Use reinforcement learning (RLHF) to design a reward model, fine-tune it based on the PPO algorithm optimization strategy, optimize the large-scale audit vertical model, and improve task performance.

[0076] In one embodiment, each agent in the multi-agent cooperative network model is obtained by fine-tuning the large audit domain model, including:

[0077] Based on the aforementioned audit vertical domain model, a multi-agent collaborative network model is constructed by combining prompt word engineering and Python-based engineering functions. The multi-agent collaborative network model includes a perceptual agent, an analytical agent, a summarizing agent, and an executive agent.

[0078] In this embodiment, the process of constructing the intelligent agents is as follows: Based on S24, by combining the prompt word engineering and Python-based engineering functions, a perception intelligent agent, an analysis intelligent agent, a summarizing intelligent agent, and an execution intelligent agent are constructed, thereby building an intelligent agent collaborative network model. The multi-agent approach allows each agent to focus on a specific function, improving the efficiency of execution at each stage. Through tiered task decomposition and the integration of the results from each agent, the overall task is completed efficiently, solving the problem of poor performance in processing multi-format documents in existing technologies.

[0079] In one embodiment, multiple sliced ​​documents are assembled to obtain an atomic file corresponding to each audit rule, including:

[0080] The multiple sliced ​​documents corresponding to each audit rule are assembled to obtain the assembled file corresponding to each audit rule;

[0081] The atomic file corresponding to each audit rule is obtained based on the audit rule number, text file number, location identifier, assembly file, and next step task content.

[0082] In this embodiment, after data extraction and document slicing based on location, multiple sliced ​​documents are obtained. The multiple sliced ​​documents are then assembled into an atomic file for the rule. The atomic file contains the rule ID, document ID, line number ID, assembled file, and next task content.

[0083] S103. Using the analytical agent, perform risk assessment on the atomic files corresponding to each audit rule. After obtaining the atomic results corresponding to each audit rule, use the summarizing agent to summarize the atomic results to obtain a risk warning list. Based on the risk warning list, summarize the atomic results with risk points to obtain audit opinions.

[0084] In this embodiment, atomic files for each audit rule are obtained. An analytical agent, based on task requirements and a rule knowledge base, evaluates the content of each atomic file to determine if any risk points exist. If risk points are found, the agent outputs the risk point details, the violation, and rectification suggestions, forming the atomic result for each rule. The atomic result includes the rule ID, document ID, line number ID, and output result. Then, the atomic results for each rule are obtained, and all results are summarized to form a risk warning list, as shown in Table 1.

[0085] Table 1 Risk Warning List

[0086]

[0087]

[0088] After obtaining the atomic results of each of the above rules, the summarizing agent is used to summarize all the results to form a risk warning list. The risky items are further summarized, and combined with the basic situation of the project, an audit opinion draft is formed, i.e., the audit opinion.

[0089] In one embodiment, after receiving the audit opinion, the method further includes:

[0090] The audit vertical domain big model uses execution intelligence to encapsulate the risk warning list as a first document and the audit opinion as a second document;

[0091] Based on the first and second documents, annotations are made on the documents to be reviewed using their numbers and location identifiers, resulting in annotation results.

[0092] In this embodiment, an execution agent is used to encapsulate the generated risk warning list into a separate file. The list is encapsulated in Excel format, and the audit opinion draft is encapsulated in Word format. Based on the document ID, line number ID, and output result, the output result is annotated at the corresponding position in the original text for easy viewing. The files are then packaged and sent back to the native system via an API call. This method enables precise annotation of results at specific locations, facilitating further investigation of issues.

[0093] It should be noted that the first document refers to the Excel document, and the second document refers to the Word document.

[0094] The intelligent auditing method for internal hospital contracts proposed in this invention is tailored to specific business scenarios, organically integrating intelligent agents with the auditing system to achieve an end-to-end usage model. This close integration with the auditing system not only enhances the practical application effectiveness of the intelligent agents but also optimizes the overall performance of the auditing system. By deeply integrating the intelligent agents and the auditing system, it overcomes the shortcomings of existing technologies in system integration and optimization, thereby improving the intelligence level and efficiency of auditing work.

[0095] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0096] In one embodiment, such as Figure 4 The diagram shows a block diagram of a hospital internal contract and project intelligent audit system 400 provided in an embodiment of this application, including an acquisition module 401, a conversion module 402, and a summary module 403, wherein:

[0097] The acquisition module 401 is used to respond to the review operation, determine the audit scenario and audit type, and acquire multiple documents to be reviewed based on the audit scenario and audit type, wherein the documents to be reviewed are contract documents or project documents;

[0098] The conversion module 402 is used to input the documents to be reviewed into a multi-agent collaborative network model based on the audit vertical domain large model, apply a perceptual agent to convert the document format of multiple documents to be reviewed to obtain multiple text files, and perform intent recognition on the documents to be reviewed to obtain an audit rule list. According to the audit rule list, the position of the content to be reviewed corresponding to each audit rule is determined. Based on the position of each content to be reviewed, the content to be reviewed is extracted and document sliced ​​to obtain multiple sliced ​​documents. The multiple sliced ​​documents are assembled to obtain the atomic file corresponding to each audit rule.

[0099] The summary module 403 is used to perform risk assessment on the atomic files corresponding to each audit rule using the analysis agent. After obtaining the atomic results corresponding to each audit rule, the summary agent summarizes the atomic results to obtain a risk warning list. Based on the risk warning list, the atomic results with risk points are summarized to obtain audit opinions.

[0100] The specific implementation method of the intelligent auditing system for internal hospital contracts and projects is basically the same as the specific implementation method of the intelligent auditing method for internal hospital contracts and projects described above, and will not be repeated here.

[0101] In one embodiment of this application, a computer device is provided, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above steps. The implementation principle and technical effects of the computer device provided in this embodiment are similar to those of the above method embodiments, and will not be repeated here.

[0102] In one embodiment of this application, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, it performs the above steps; the implementation principle and technical effects of the computer-readable storage medium provided in this embodiment are similar to those of the above method embodiments, and will not be repeated here.

[0103] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0104] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for intelligent auditing of in-house contracts and projects in a hospital, characterized in that, include: In response to the review operation, the audit scenario and audit type are determined, and multiple documents to be reviewed are obtained according to the audit scenario and audit type, wherein the documents to be reviewed are contract documents or project documents; The documents to be reviewed are input into a multi-agent collaborative network model based on a large audit domain model. A perceptual agent is applied to convert the document formats of multiple documents to be reviewed, resulting in multiple text files. Intent recognition is performed based on the documents to be reviewed, the audit scenario, and the audit type to obtain an audit rule list. Based on the audit rule list, the location of the content to be reviewed corresponding to each audit rule is determined. Based on the location of each content to be reviewed, the content to be reviewed is extracted and sliced ​​into multiple sliced ​​documents. The multiple sliced ​​documents corresponding to each audit rule are assembled to obtain an assembled file corresponding to each audit rule. Atomic files corresponding to each audit rule are obtained based on the audit rule number, text file number, location identifier, the assembled file, and the next task content. The analysis agent performs risk assessment on the atomic files corresponding to each audit rule, and obtains the atomic results corresponding to each audit rule. Then, the summarizing agent summarizes the atomic results to obtain a risk warning list. Based on the risk warning list, the atomic results with risk points are summarized to obtain audit opinions.

2. The intelligent auditing method for internal hospital contracts and projects as described in claim 1, characterized in that, The process involves using a perceptual agent to convert the document formats of multiple documents to be reviewed, resulting in multiple text files, including: Using PDF parsing and conversion technology, each page of the multiple documents to be reviewed is converted into an image to obtain the corresponding converted file; Based on each of the converted files, multimodal OCR technology is used to locate and extract computer text and handwritten text to obtain the corresponding extraction results and text coordinates. The corresponding extraction results are then embedded into the corresponding files to be reviewed to obtain the text file corresponding to each file to be reviewed.

3. The intra-hospital contracts and projects intelligent auditing method of claim 2, wherein, After obtaining the multiple text files, the process also includes: Based on punctuation and grammar rules, each of the text files is intelligently segmented to obtain multiple segmented text files. Correct erroneous and missing characters in each sentence of each segmented text file, and assign a location identifier.

4. The intra-hospital contracts and projects intelligent auditing method of claim 1, wherein, After receiving the aforementioned audit opinion, the following is also included: The risk warning list is encapsulated into a first document using an execution intelligent agent, and the audit opinion is encapsulated into a second document. Based on the first document and the second document, annotations are made on the documents to be reviewed using the numbers of each document to be reviewed and the location identifiers, and annotation results are obtained.

5. The intra-hospital contracts and projects intelligent auditing method of claim 1, wherein, The audit-specific large-scale model is obtained by pre-training a general large-scale model using textual data from the audit domain, including: Acquire audit-related text data, which includes audit-related legal and regulatory text data, audit cases, and audit rules; NLP technology is used to segment and identify entities in the text data of the audit domain to obtain key information, and a knowledge graph is constructed based on the entities and relationships between entities in the key information. A general large model is selected as the base model. The entity vectors in the knowledge graph are inserted into the base model as key values. An adapter structure is used to isolate the knowledge parameters and dynamically adjust the learning rate and mask ratio to obtain the initial audit vertical domain large model. Based on the business needs and model optimization objectives of the audit vertical domain, the model is divided into multiple tasks. Supervision data and instruction sets for each task are constructed. Based on the supervision data and instruction sets of each task, a reward model is designed using reinforcement learning and an optimization strategy based on the PPO algorithm is used to optimize the initial audit vertical domain model, thus obtaining the audit vertical domain large model.

6. The intra-hospital contracts and projects intelligent auditing method of claim 1, wherein, Each agent in the multi-agent cooperative network model is obtained by fine-tuning the large audit domain model, including: Based on the aforementioned audit vertical domain model, a multi-agent collaborative network model is constructed by combining prompt word engineering and Python-based engineering functions. The multi-agent collaborative network model includes a perceptual agent, an analytical agent, a summarizing agent, and an executive agent.

7. A hospital internal contracts and projects intelligent auditing system, characterized in that, To implement the intelligent auditing method for hospital internal contracts and projects as described in any one of claims 1-6, the method includes: The acquisition module is used to respond to the review operation, determine the audit scenario and audit type, and acquire multiple documents to be reviewed based on the audit scenario and audit type, wherein the documents to be reviewed are contract documents or project documents; The conversion module is used to input the documents to be reviewed into a multi-agent collaborative network model based on the audit vertical domain large model, apply a perceptual agent to convert the document format of multiple documents to be reviewed to obtain multiple text files, and perform intent recognition on the documents to be reviewed to obtain an audit rule list. According to the audit rule list, the position of the content to be reviewed corresponding to each audit rule is determined. Based on the position of each content to be reviewed, the content to be reviewed is extracted and document sliced ​​to obtain multiple sliced ​​documents. The multiple sliced ​​documents are assembled to obtain the atomic file corresponding to each audit rule. The aggregation module is used to perform risk assessment on the atomic files corresponding to each audit rule using the analysis agent, obtain the atomic results corresponding to each audit rule, and then use the summarizing agent to aggregate the atomic results to obtain a risk warning list. Based on the risk warning list, the atomic results with risk points are summarized to obtain audit opinions.

8. A computer device, comprising: include: Memory, used to store computer programs; A processor, configured to implement the intelligent auditing method for internal hospital contracts and projects as described in any one of claims 1 to 6 when executing the computer program.

9. A storage medium, characterized by The storage medium stores a computer program, which, when executed by a processor, implements the steps of the intelligent auditing method for internal hospital contracts and projects as described in any one of claims 1 to 6.

Images