Automatic budget auditing method and system

Abstract

The invention provides an automatic budget auditing method and system, and relates to the technical field of artificial intelligence, and the method comprises the steps: receiving a to-be-audited budget file and an auditing rule; based on the auditing rule and the file format and the professional attribute of the budget file, determining an analysis agent; performing structured analysis on the budget file based on the analysis agent to generate to-be-audited data; and calling the vertical domain large language model, and performing budget auditing on the to-be-audited data based on the auditing rule. According to the method and the system provided by the invention, structured analysis is performed on the budget file by using the analysis agent, so that the automation degree and the accuracy of auditing are improved; according to the budget auditing method, budget auditing is carried out by calling the vertical domain large language model subjected to vertical domain parameter fine adjustment, so that problems and abnormal conditions in the budget file can be identified more accurately, the budget auditing quality and efficiency are improved, and a more reliable auditing result is provided for a user; through automatic budget auditing, the budget auditing efficiency is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an automated budget auditing method and system. Background Technology

[0002] In the fields of telecommunications and municipal engineering construction, budget review still generally adopts the traditional model of "mainly offline manual review, supplemented by system registration." Taking telecommunications engineering as an example, its budget documents include multiple tables such as "Engineering Quantity Budget Table," "Material Quotation Table," and "Construction Cost Details Table," with a massive amount of data that must meet the requirements of multiple industry standards such as "Telecommunications Engineering Budget Quotas" and "Wireless Communication Base Station Construction Specifications." Budget data is scattered across various formats such as Excel and PDF, with significant differences in table structures among different design institutes. Furthermore, the review involves three major disciplines: transmission, wireless, and pipeline. The rules for each discipline are both independent and interconnected. For example, the quantity of fiber optic cable laying in the transmission discipline needs to be calculated in conjunction with the trench length in the pipeline discipline. In addition, when industry policies, market prices, or company regulations are updated, manual reviewers need to relearn the rules and manually adjust the process, resulting in a significant lag in rule response and a discrepancy between the budget review results and actual cost requirements.

[0003] To address these challenges, some companies have attempted to deploy rule engines based on regular expressions, but these engines can only handle simple logic such as "unit price > threshold" and cannot understand or execute review rules containing complex semantics. While traditional machine learning models perform reasonably well in single-professional reviews, their performance drops significantly in cross-professional scenarios. Due to the lack of professional knowledge transfer capabilities, separate models need to be trained for each profession, leading to a substantial increase in operational costs.

[0004] Traditional manual review processes are inefficient, error-prone, and slow to respond, while existing automation technologies suffer from limitations such as insufficient expressive power, difficulty in migration, and poor adaptability, failing to truly solve the problem. Therefore, improving the efficiency of budget review has become a pressing technical challenge for the industry. Summary of the Invention

[0005] This invention provides an automated budget auditing method and system to solve the problems of low efficiency of manual review, fragmented rules of multiple disciplines, and delayed dynamic response in the prior art, thereby improving the efficiency of budget auditing.

[0006] This invention provides an automated budget auditing method, comprising: Receive budget documents and audit rules to be audited; Based on the aforementioned review rules, as well as the file format and professional attributes of the budget document, a parsing agent is determined; The budget document is structured and parsed based on the parsing agent to generate data to be reviewed; The vertical domain large language model is invoked to perform budget audit on the data to be audited based on the audit rules; the vertical domain large language model is obtained by fine-tuning the parameters of the large language model based on the historical case library of the vertical domain to which the budget document belongs.

[0007] In some embodiments, the audit rules include fixed rules stored in a knowledge base and dynamic rules input by the user in natural language; the knowledge base includes a professional rule base, a quota database, and a historical case database.

[0008] In some embodiments, the method further includes: Based on the audit rules, and the file format and professional attributes of the budget document, an audit mode is determined; wherein, the audit mode includes at least one of static audit and dynamic audit; the static audit refers to auditing the budget document based on the fixed rules; the dynamic audit refers to auditing based on the dynamic rules input by the user in natural language.

[0009] In some embodiments, auditing the budget document based on the fixed rules includes: Call the fixed rules in the professional rule base and convert the fixed rules into logical instructions that can be executed by the vertical domain large language model; The vertical domain large language model is invoked, and the audit data to be reviewed is performed based on the logical instructions.

[0010] In some embodiments, the auditing based on the dynamic rules input by the user in natural language includes: Based on the dynamic rules input by the user in natural language, natural language instructions and floating parameters are determined; The natural language instructions are broken down into instruction elements; The floating parameters are standardized and calibrated by calling the quota database. Match similar scenarios from the historical case library to generate an audit path enhanced by transfer learning; Based on the instruction elements, the review path, and the standardized and calibrated floating parameters, a review instruction chain is generated through prompt word engineering; the prompt word engineering includes a layered template design of professional layer, rule layer, and data layer; The budget document is audited by invoking the vertical domain large language model and based on the audit instruction chain.

[0011] In some embodiments, the method further includes: If the budget document's review type includes both static and dynamic review, the static review will be performed first for basic verification, and then the dynamic review will be initiated to conduct a special review of the discount terms.

[0012] This invention provides an automated budget auditing system, comprising: The input module is used to receive budget documents and audit rules to be audited; The agent scheduling module is used to determine the parsing agent based on the audit rules, as well as the file format and professional attributes of the budget file; The data processing module is used to perform structured parsing of the budget document based on the parsing agent to generate data to be reviewed; The large model processing module is used to call the vertical domain large language model and perform budget audit on the data to be audited based on the audit rules; the vertical domain large language model is obtained by fine-tuning the parameters of the large language model based on the historical case library of the vertical domain to which the budget document belongs.

[0013] The present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the automated budget auditing method.

[0014] The present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the aforementioned automated budget auditing method.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the automated budget auditing method.

[0016] The automated budget auditing method and apparatus provided in this invention determine the parsing agent based on audit rules, file format, and professional attributes. This enables the system to select appropriate parsing methods for budget documents of different formats and professional backgrounds, improving the accuracy and efficiency of parsing. By utilizing the parsing agent to perform structured parsing of budget documents, a clear, standardized, and easy-to-operate data format is provided for subsequent budget auditing, enhancing the automation and accuracy of the audit. By calling a vertical domain large language model with fine-tuned vertical domain parameters for budget auditing, problems and anomalies in budget documents can be identified more accurately, improving the quality and efficiency of budget auditing and providing users with more reliable audit results. Automated budget auditing significantly improves the efficiency of budget review. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

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

[0019] Figure 1 This is a flowchart illustrating the automated budget auditing method provided by the present invention.

[0020] Figure 2 This is a flowchart of the intelligent agent workflow of the automated budget auditing method provided by the present invention.

[0021] Figure 3 This is a schematic diagram of the automated budget auditing system provided by the present invention.

[0022] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0023] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps, units, or modules is not necessarily limited to those explicitly listed, but may include other steps, units, or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0025] Figure 1 This is a flowchart illustrating the automated budget auditing method provided by the present invention, as shown below. Figure 1 As shown, the method includes steps 110, 120, 130 and 140.

[0026] Step 110: Receive the budget documents and audit rules to be audited.

[0027] Specifically, the automated budget auditing method provided in this embodiment of the invention is executed by an automated budget auditing device. This device can be implemented in software, such as an automated budget auditing program running on a computer; or it can be implemented in hardware, such as a computer or server that executes the automated budget auditing method.

[0028] Budget documents can be electronic documents in various formats, including but not limited to: Excel spreadsheets, PDF documents, Word documents, etc. These documents typically include a bill of quantities, a list of equipment and materials, a list of labor costs, etc.

[0029] Audit rules are a set of rules used to check the compliance and reasonableness of budget documents.

[0030] In this embodiment of the invention, the system receives budget documents to be audited and corresponding audit rules uploaded by the user.

[0031] Step 120: Based on the audit rules, as well as the file format and professional attributes of the budget document, determine the parsing agent.

[0032] Specifically, an intelligent agent is a software entity with autonomy, perception, and decision-making capabilities, capable of performing tasks and achieving predetermined goals in a specific environment. In an automated budget auditing system, a suitable parsing intelligent agent is selected based on the file format and professional attributes of the budget document. The parsing intelligent agent perceives user input and document content, invokes relevant modules and resources, performs complex parsing tasks, and generates corresponding results.

[0033] It should be noted that each specialized analytical agent is familiar with the professional terminology, table structures, and data relationships within its respective field. Analytical agents include at least specialized review agents for communication transmission, wireless, and pipeline technologies.

[0034] In this embodiment of the invention, the system first identifies the format type and professional attributes of the budget document, and parses the audit rules. Then, based on the audit rules, the file format and professional attributes of the budget document, a parsing agent is matched.

[0035] In addition, sub-agents can be dynamically switched based on the type of review rule (e.g., dynamic rules, static rules), file format, and professional attributes to achieve parallel review by multiple professions.

[0036] Step 130: Perform structured parsing of the budget document based on the parsing agent to generate data to be reviewed.

[0037] Specifically, structured parsing is a process of converting unstructured or semi-structured data (such as text, tables, etc.) into structured data (such as key-value pairs, data frames, etc.). It identifies and extracts key information from the data, organizing it into a format that facilitates subsequent processing and analysis. In automated budget auditing systems, structured parsing refers to converting data in budget documents into a format that the system can understand and process.

[0038] In this embodiment of the invention, the parsing agent can perform structured parsing of budget documents and generate data to be reviewed through steps such as document recognition, key field extraction, data standardization, and relationship construction.

[0039] Step 140: Call the vertical domain large language model and perform budget audit on the data to be audited based on the audit rules; the vertical domain large language model is obtained by fine-tuning the parameters of the large language model based on the historical case library of the vertical domain to which the budget document belongs.

[0040] Specifically, a vertical domain large language model is a large language model that has been fine-tuned for a specific domain.

[0041] In this embodiment of the invention, the vertical domain large language model is obtained by fine-tuning the parameters of the QWQ-32B model based on a historical case library of the vertical domain to which the budget document belongs. Through supervised fine-tuning based on a large amount of historical data and industry standards in that domain, it enhances the semantic understanding of budget rules, terminology, and logic in that domain, thereby enabling efficient and accurate processing of budget review tasks.

[0042] In this embodiment of the invention, by calling the vertical domain large language model and based on the audit rules, semantic matching and logical verification are performed on the data to be audited obtained by the parsing agent through structured parsing, so as to realize budget auditing and generate audit reports and early warning information.

[0043] It should be noted that the intelligent audit report provided by the system includes precise location of abnormal data, multi-level early warning classification, and optimization suggestions. The audit path and decision-making basis are intuitively displayed through a visual interface, supporting multiple output formats such as PDF and Excel, meeting the needs of machine-readable data interaction and facilitating manual review and audit traceability.

[0044] The automated budget auditing method provided in this invention determines the parsing agent based on audit rules, file format, and professional attributes. This allows the system to select appropriate parsing methods for budget documents of different formats and professional backgrounds, improving the accuracy and efficiency of parsing. By utilizing the parsing agent to perform structured parsing of budget documents, a clear, standardized, and easy-to-operate data format is provided for subsequent budget auditing, enhancing the automation and accuracy of the audit. Furthermore, by calling a vertical domain-specific large language model with fine-tuned parameters for budget auditing, problems and anomalies in budget documents can be identified more accurately, improving the quality and efficiency of budget auditing and providing users with more reliable audit results. Automated budget auditing significantly improves the efficiency of budget review.

[0045] In some embodiments, the audit rules include fixed rules stored in a knowledge base and dynamic rules input by the user in natural language; the knowledge base includes a professional rule base, a quota database, and a historical case database.

[0046] Specifically, fixed rules are a set of rules pre-stored in a knowledge base. These rules are based on mandatory industry standards in the field of communications engineering, corporate cost control standards, and are derived from a large number of historical audit cases. They are stable and authoritative, providing a fundamental and crucial judgment standard for budget audits.

[0047] For example, in pipeline engineering reviews, rules such as "trench excavation volume = design length × cross-sectional area × excavation coefficient" and "contingency fee = total price excluding tax × 3%" are specific examples of fixed rules. These rules are stored in a rule base, and the system automatically calls these fixed rules during the review process to compare and verify key fields in the budget table, thereby achieving a basic review of the budget documents.

[0048] Dynamic rules are rules that users input into the system in natural language based on the specific circumstances of a project. These rules are highly flexible and can adapt to special circumstances such as dynamic discount clauses in construction contracts and regional policy-driven price adjustments, thus overcoming the limitations of fixed rules.

[0049] For example, users can input dynamic parameters such as "construction discount rate 26%" or "contract discount reduced by 3%". The system uses a semantic parsing engine to decompose these dynamic rules described in natural language into executable instruction elements, and then combines them with other data in the knowledge base for review.

[0050] A professional rule base is a rule repository built for different professions (such as transmission, wireless, pipeline, etc.); it covers the detailed rule logic of each profession, which is based on industry norms, enterprise standards and professional practice experience.

[0051] The quota database stores quota standards for various materials, quantities, and costs. These quota standards are important references for budget review, and the system frequently accesses data from the quota database during the review process.

[0052] For example, when reviewing the unit prices of materials in the budget sheet, the unit prices in the sheet are compared with the standard unit prices of the corresponding numbers in the quota database to determine if there are any discrepancies. The quota database not only provides a data benchmark for fixed-rule reviews but also plays a crucial role in dynamic-rule reviews by standardizing and calibrating the floating parameters input by users to ensure the accuracy of the review.

[0053] The historical case library is a crucial component of the system's self-evolution and optimization, storing a vast amount of case data accumulated from past review processes. This data includes review results, anomalies, and handling suggestions for various budget documents. During the review process, the system uses transfer learning to match similar scenarios from the historical case library, generating transfer-learning-enhanced review paths. The historical case library provides rich practical references for the dynamic adjustment and optimization of review rules, enabling the system to continuously learn and adapt to new review scenarios, thus improving the intelligence level of the review process. Simultaneously, all review results are also stored in the historical case library, forming a closed loop of self-evolution.

[0054] In this embodiment of the invention, the audit rules include fixed rules stored in a knowledge base and dynamic rules input by the user in natural language; the knowledge base includes a professional rule base, a quota database, and a historical case database.

[0055] The automated budget auditing method provided in this invention is based on a modular design concept, decoupling the core auditing engine from the domain knowledge base. Relying on a three-layer structured knowledge base (rule base, quota base, and historical case base), the system possesses industry scalability. By adjusting the knowledge base content, the same technical architecture can quickly adapt to the budget auditing needs of different engineering fields, reducing the development and maintenance costs of multi-industry applications. By employing a natural language interaction interface and an intelligent rule engine, a dual-mode parallel mechanism of fixed rule auditing and dynamic rule auditing is achieved. Auditing parameters are dynamically adjusted through simple commands, automatically completing rule calibration and logic updates. Compared to traditional methods requiring recoding, this shortens maintenance response time and reduces maintenance costs.

[0056] In some embodiments, the method further includes: Based on the audit rules, as well as the file format and professional attributes of the budget documents, an audit mode is determined; wherein, the audit mode includes at least one of static audit and dynamic audit; static audit refers to auditing the budget documents based on the fixed rules; dynamic audit refers to auditing based on dynamic rules input by the user in natural language.

[0057] Specifically, to better adapt to the review needs of different budget documents and improve the accuracy and efficiency of the review, the system will determine the corresponding review mode based on the review rules, the file format, and professional attributes of the budget documents. The review modes mainly include two types: static review and dynamic review.

[0058] Static auditing refers to auditing the budget documents based on the established rules. When the audit task primarily involves verifying these established rules, the system will determine to use the static auditing mode. The static auditing mode focuses on accurately verifying the basic rules in the budget documents to ensure that the budget preparation complies with industry norms and company standards.

[0059] For example, for a communication engineering budget document that conforms to a standard format and is mainly compiled according to industry-standard rules, the system will prioritize the static audit mode. By calling fixed rules in the rule base, the system will perform batch comparison and verification of key fields in the budget document to quickly complete the basic audit task.

[0060] When the review task requires consideration of special rules or parameters input by the user, the system will determine to adopt the dynamic review mode. The dynamic review mode can flexibly handle various personalized needs, such as dynamic discount clauses in construction contracts and regional policy-based price adjustments.

[0061] For example, if a user enters dynamic parameters such as "construction discount rate 26%" or "contract discount decrease 3%" during the review process, the system will activate the dynamic review mode based on these dynamic rules. The system will use a semantic parsing engine to convert the dynamic rules described in natural language into executable review instructions and combine them with other data in the knowledge base for comprehensive verification to meet the project-specific review requirements.

[0062] In this embodiment of the invention, when a user uploads a budget file, the system automatically starts the file type recognition engine through the intelligent agent workflow. Based on the file title, header features and content keywords (such as "optical cable laying engineering quantity table" or "pipeline import material list"), the system determines the professional attribute (transmission / wireless / pipeline) and intelligently selects the review mode in combination with the review rules.

[0063] For example, the prompts for an agent workflow could be: Role: Automated Approval System Task objective: Based on the user input rules "Variable Settings > Review Rules", review the information in Table 3 of the user-uploaded files and output the review results according to the knowledge base.

[0064] Constraints: The knowledge base is a global knowledge base; it is essential to scan the entire knowledge base to ensure that no data is missed. Strictly adhere to the user-selected review content; content not specified will not be reviewed. Southern Nuclear Industry Rules and Processes (selected based on each review rule entered by the user): 1. Verification of quota number and name: Check whether the “quota number” in Table 3 is consistent with the standard “project name” in the knowledge base. Differences in punctuation are allowed, but the core content must be completely consistent.

[0065] If there is no matching number or the core content of the name does not match in the knowledge base, it will be marked as a Level 1 warning, and the abnormal situation will be recorded in detail.

[0066] 2. Verification of man-day values: Compare the "Unit Quota Value (Skilled Worker / General Worker)" in Table 3 with the standard value in the knowledge base, allowing for an error of ±5%.

[0067] If the error exceeds the error range or there is no corresponding value in the knowledge base, it will be marked as a level 2 warning and the specific anomaly will be recorded.

[0068] 3. Consistency check of total value: Calculate the "Quantity * Unit Quota Value (Skilled Workers / General Workers)" in Table 3, round it to 4 decimal places, and compare it with the "Total Value (Skilled Workers / General Workers)".

[0069] If the difference between the calculated value and the total value exceeds ±0.01, it will be marked as abnormal and recorded in detail.

[0070] Output requirements: Following the order of "Variable Settings > Audit Rules", provide feedback on each audit result, indicating the level of abnormality and details.

[0071] If all reviewed items are found to be normal, output "Review passed".

[0072] If any issues are found, all problems will be summarized and a "Report of Approval Failure" will be generated.

[0073] The automated budget auditing method provided in this invention combines static and dynamic auditing, enabling the auditing process to better meet the personalized needs of actual projects. This not only ensures the standardization and normalization of the auditing process but also improves its accuracy and adaptability.

[0074] In some embodiments, auditing budget documents based on fixed rules includes: Call the fixed rules in the professional rule base and convert the fixed rules into logical instructions that can be executed by the vertical domain large language model; The system invokes a large language model for the vertical domain and performs audits on the data to be reviewed based on logical instructions.

[0075] Specifically, Figure 2This is a flowchart of the intelligent agent workflow of the automated budget auditing method provided by the present invention, as follows: Figure 2 As shown, in this embodiment of the invention, auditing budget documents based on fixed rules includes: parsing the budget documents based on a parsing agent to determine the data to be audited; calling fixed rules in a professional rule base and converting the fixed rules into logical instructions that can be executed by a vertical domain large language model; and calling the vertical domain large language model to audit the data to be audited based on the logical instructions.

[0076] Taking pipeline audit as an example, the intelligent agent parsing module automatically extracts key fields from the budget table (such as "pipe length" and "number of joints"), then calls fixed rules from the rule base (such as "trench earthwork volume = design length × cross-sectional area × excavation coefficient" and "contingency fee = total price excluding tax × 3%)). The QWQ-32B large model converts these rules into executable code, performing batch comparisons of the table data. When it detects that "the concrete grade in Table 3A deviates from the C30 standard unit price in the quota library by more than ±5%", the system automatically marks the anomaly and generates a structured audit report, achieving fully automated auditing with zero human intervention. The report includes a list of issues, rectification suggestions, and the basis for the rules, and highlights the discrepancies in the original table.

[0077] The automated budget auditing method provided by this invention provides clear standards for auditing based on fixed rules in a professional rule base, and ensures the accuracy of audit results based on the precise verification capability of a vertical domain large language model, effectively reducing errors caused by human negligence. Through this fixed rule-based auditing method, an efficient, accurate, and standardized automated solution is provided for budget auditing, significantly improving the quality and efficiency of budget auditing.

[0078] In some embodiments, auditing is based on dynamic rules input by the user in natural language, including: Based on dynamic rules input by the user in natural language, natural language instructions and floating parameters are determined; Decompose natural language instructions into instruction elements; The floating parameters are standardized and calibrated by calling the quota database; Match similar scenarios from a historical case library to generate review paths enhanced by transfer learning; Based on instruction elements, review paths, and standardized and calibrated floating parameters, a review instruction chain is generated through prompt word engineering; prompt word engineering includes layered template design of professional layer, rule layer, and data layer; The budget documents are audited by invoking the vertical domain's large language model and based on the audit instruction chain.

[0079] Specifically, while audits based on fixed rules ensure the efficiency and accuracy of basic audits, they are limited by the static nature of the rule base and are difficult to adapt to dynamic discount clauses or regional policy-based price adjustments in construction contracts. This invention introduces a dynamic rule audit mode to overcome this limitation, enabling audits to be performed based on dynamic rules input by users in natural language.

[0080] like Figure 2 As shown, in this embodiment of the invention, auditing based on dynamic rules input by the user in natural language includes: determining natural language instructions and floating parameters based on the dynamic rules input by the user in natural language; decomposing the natural language instructions into instruction elements; calling a quota database to standardize and calibrate the floating parameters; matching similar scenarios from a historical case library to generate an audit path enhanced by transfer learning; generating an audit instruction chain based on the instruction elements, the audit path, and the standardized and calibrated floating parameters through prompt word engineering; prompt word engineering includes a layered template design of a professional layer, a rule layer, and a data layer; and calling a large-scale language model in a vertical domain to audit the budget document based on the audit instruction chain. Prompt word engineering is a technique and art involving the design, optimization, and iteration of prompt words, aiming to guide the large-scale language model to produce the desired, high-quality output results.

[0081] For example, after a user uploads a budget file, they input dynamic rules described in natural language through the system interface. The intelligent agent workflow first identifies and determines the natural language instructions and floating parameters within them. For instance, if the user inputs dynamic parameters such as "construction discount rate 26%" or "contract discount reduced by 3%", the system will identify them as floating parameters and extract the natural language instructions from them. The system can also decompose natural language instructions into specific instruction elements; for example, "report an error if the contingency fund column in audit form 1 is empty" can be broken down into (target table: form 1, target column: contingency fund, validation logic: non-empty judgment).

[0082] Furthermore, the system simultaneously activates the knowledge base decision-making mechanism. On the one hand, it calls the quota library to standardize and calibrate the floating parameters input by the user (such as associating "material unit price down 8%" with the price benchmark of the corresponding number in the quota library). On the other hand, it uses the reinforcement learning module to match similar scenarios from the historical case library, generate an audit path enhanced by transfer learning, and use the experience in the historical case library to provide reference and optimization suggestions for the current audit task.

[0083] At this point, the QWQ-32B large-scale model, fine-tuned in the field of communications engineering, plays a core decision-making role: based on instruction elements, review paths, and standardized and calibrated floating parameters, it integrates professional specifications, dynamic parameters, and quota library benchmark data into a machine-executable review instruction chain through prompt word engineering. Example instruction: "If the 'cable unit price' in Table 3A is greater than the price of quota library number GD-042 × 0.95, then trigger the secondary warning 'price exceeds the threshold'." Based on this, the large-scale model performs semantic-level logical verification on the budget table and outputs an intelligent report containing discrepancy location, cross-professional conflict analysis (such as the contradiction in the concrete allocation of the shared tower base for pipelines and wireless), and optimization suggestions.

[0084] It should be noted that the prompt word project includes a layered template design with professional, rule, and data layers to ensure the comprehensiveness and accuracy of the review instruction chain.

[0085] The automated budget auditing method provided in this invention performs audits based on user-input dynamic rules, adjusts audit parameters in real time, quickly adapts to policy adjustments and market changes, and can handle complex situations such as dynamic discount clauses in construction contracts and regional policy-driven price adjustments. Through a reinforcement learning module, it matches similar scenarios from a historical case library to generate audit paths enhanced by transfer learning, improving audit accuracy. Using hierarchical prompt word technology, it integrates professional specifications, dynamic parameters, and quota library benchmark data into a machine-executable audit instruction chain. The large model can perform semantic-level logic verification and output intelligent reports containing discrepancy location, cross-professional conflict analysis, and optimization suggestions, effectively identifying and handling complex logical relationships and improving the level of audit intelligence. The dynamic rule auditing mode dynamically adjusts audit parameters through simple instructions, automatically completing rule calibration and logic updates without frequent manual modification of the rule library or retraining of the audit model, improving the efficiency, accuracy, and intelligence of budget auditing while reducing audit costs.

[0086] In some embodiments, the method further includes: If the budget document review includes both static and dynamic reviews, the static review should be performed first for basic verification, and then the dynamic review should be initiated to conduct a special review of the discount terms.

[0087] Specifically, the dynamic and static audit modes in this invention can be coordinated through an intelligent agent state scheduling engine.

[0088] In this embodiment of the invention, the agent state scheduling engine can automatically select the review type (static review or dynamic review) based on user input and input parameters. Furthermore, in the process of selecting the review mode, the agent state scheduling engine comprehensively considers multiple factors and constructs a scientifically sound decision tree.

[0089] Among them, the decision tree is a tree-structured decision model used to select the optimal path from multiple possible review paths. It dynamically constructs the review process by analyzing user-input review requirements, the professional attributes of budget documents (such as transmission, wireless, pipeline, etc.), file formats (such as Excel, PDF, etc.), and the relevance of the knowledge base (such as the matching of rule base, quota base, and historical case base). Each node represents a decision point, and each branch represents a possible review path. Ultimately, the most suitable review mode and path are determined through a series of decision points.

[0090] For example, when a user uploads a fixed-format pipeline budget sheet and dynamic discount parameters at the same time, the engine first performs basic checks on the fixed rules, and then starts the dynamic rules module to conduct a special review of the discount terms.

[0091] All audit results are stored in the historical case library in real time, forming a self-evolving closed loop of "data input → rule parsing → knowledge base reinforcement → model decision-making → case storage".

[0092] Ultimately, the system generates a structured report that includes machine-readable coordinates of abnormal data, first- and second-level warning labels based on knowledge base rules (first-level for abnormal names and numbers, and second-level for abnormal data values), and a cross-professional budget integration solution, achieving a generational leap from fragmented manual review to full-domain intelligent auditing.

[0093] The automated budget auditing method provided in this invention achieves efficient collaboration between dynamic and static auditing through an intelligent agent state scheduling engine. This not only optimizes the auditing process and improves auditing efficiency and accuracy, but also forms a self-evolutionary closed loop through the accumulation and application of auditing results, ultimately generating a structured report and realizing an intelligent upgrade of budget auditing.

[0094] The apparatus provided in the embodiments of the present invention will be described below. The apparatus described below can be referred to in correspondence with the method described above.

[0095] Figure 3 This is a schematic diagram of the automated budget auditing system provided by the present invention, as shown below. Figure 3 As shown, the device includes an input module 310, an agent scheduling module 320, a data processing module 330, and a large model processing module 340 connected in sequence.

[0096] Input module 310 is used to receive budget documents to be audited and audit rules; The intelligent agent scheduling module 320 is used to determine the parsing intelligent agent based on the audit rules, as well as the file format and professional attributes of the budget file; Data processing module 330 is used to perform structured parsing of the budget document based on the parsing agent to generate data to be reviewed; The large model processing module 340 is used to call the vertical domain large language model and perform budget audit on the data to be audited based on the audit rules; the vertical domain large language model is obtained by fine-tuning the parameters of the large language model based on the historical case library of the vertical domain to which the budget document belongs.

[0097] The automated budget auditing system provided in this invention determines the parsing agent based on audit rules, file format, and professional attributes. This allows the system to select appropriate parsing methods for budget documents of different formats and professional backgrounds, improving the accuracy and efficiency of parsing. By utilizing the parsing agent to perform structured parsing of budget documents, a clear, standardized, and easy-to-operate data format is provided for subsequent budget auditing, enhancing the automation and accuracy of the audit. Furthermore, by calling a vertical domain-specific large language model with fine-tuned parameters for budget auditing, problems and anomalies in budget documents can be identified more accurately, improving the quality and efficiency of budget auditing and providing users with more reliable audit results. Through automated budget auditing, the efficiency of budget review is significantly improved.

[0098] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 4 As shown, the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communications bus 440, wherein the processor 410, the communications interface 420, and the memory 430 communicate with each other via the communications bus 440. The processor 410 can call logical commands stored in the memory 430 to execute the methods described in the above embodiments, for example: The system receives budget documents and audit rules to be audited; based on the audit rules, as well as the file format and professional attributes of the budget documents, it determines a parsing agent; based on the parsing agent, it performs structured parsing of the budget documents to generate data to be audited; it calls a vertical domain large language model to perform budget auditing on the data to be audited based on the audit rules; the vertical domain large language model is obtained by fine-tuning the parameters of the large language model based on the historical case library of the vertical domain to which the budget documents belong.

[0099] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part 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 the present 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.

[0100] The processor in the electronic device provided in this embodiment of the invention can call logical instructions in the memory to implement the above method. Its specific implementation method is the same as the aforementioned method implementation method and can achieve the same beneficial effects, which will not be repeated here.

[0101] This invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the methods provided in the above embodiments.

[0102] The specific implementation method is the same as the aforementioned method implementation method and can achieve the same beneficial effects, so it will not be repeated here.

[0103] This invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0104] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0105] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications 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.

Claims

1. An automated budget auditing method, characterized in that, include: Receive budget documents and audit rules to be audited; Based on the aforementioned review rules, as well as the file format and professional attributes of the budget document, a parsing agent is determined; The budget document is structured and parsed based on the parsing agent to generate data to be reviewed; The vertical domain large language model is invoked to perform budget audit on the data to be audited based on the audit rules; the vertical domain large language model is obtained by fine-tuning the parameters of the large language model based on the historical case library of the vertical domain to which the budget document belongs.

2. The automated budget auditing method according to claim 1, characterized in that, The audit rules include fixed rules stored in the knowledge base and dynamic rules input by the user in natural language; the knowledge base includes a professional rule base, a quota database, and a historical case database.

3. The automated budget auditing method according to claim 2, characterized in that, Before invoking the vertical domain large language model, the method further includes: Based on the audit rules, and the file format and professional attributes of the budget document, an audit mode is determined; wherein, the audit mode includes at least one of static audit and dynamic audit; the static audit refers to auditing the budget document based on the fixed rules; the dynamic audit refers to auditing based on the dynamic rules input by the user in natural language.

4. The automated budget auditing method according to claim 3, characterized in that, The auditing of the budget document based on the fixed rules includes: Call the fixed rules in the professional rule base and convert the fixed rules into logical instructions that can be executed by the vertical domain large language model; The vertical domain large language model is invoked, and the audit data to be reviewed is performed based on the logical instructions.

5. The automated budget auditing method according to claim 3, characterized in that, The audit based on the dynamic rules input by the user in natural language includes: Based on the dynamic rules input by the user in natural language, natural language instructions and floating parameters are determined; The natural language instructions are broken down into instruction elements; The floating parameters are standardized and calibrated by calling the quota database. Match similar scenarios from the historical case library to generate an audit path enhanced by transfer learning; Based on the instruction elements, the review path, and the standardized and calibrated floating parameters, a review instruction chain is generated through prompt word engineering; the prompt word engineering includes a layered template design of professional layer, rule layer, and data layer; The budget document is audited by invoking the vertical domain large language model and based on the audit instruction chain.

6. The automated budget auditing method according to claim 3, characterized in that, If the budget document's review type includes both static and dynamic reviews, the static review will be performed first for basic verification, and then the dynamic review will be initiated to conduct a special review of the discount terms.

7. An automated budget auditing system, characterized in that, include: The input module is used to receive budget documents and audit rules to be audited; The agent scheduling module is used to determine the parsing agent based on the audit rules, as well as the file format and professional attributes of the budget file; The data processing module is used to perform structured parsing of the budget document based on the parsing agent to generate data to be reviewed; The large model processing module is used to call the vertical domain large language model and perform budget audit on the data to be audited based on the audit rules; the vertical domain large language model is obtained by fine-tuning the parameters of the large language model based on the historical case library of the vertical domain to which the budget document belongs.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the automated budget auditing method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the automated budget auditing method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the automated budget auditing method as described in any one of claims 1 to 6.

Images