A robot auditing operation automation process management method and system

Abstract

The present invention relates to the field of robot audit technology, and discloses a robot audit work automation process management method and system, including: collecting data information of audit target objects through robots, building an internal control matrix and a node map of major misstatement risks; using an improved annealing algorithm to optimize the path in the node map of major misstatement risks to obtain the audit path of major misstatement risks; identifying the audit path of major misstatement risks to obtain the robot's audit results for audit nodes or audit node combinations. The identification of major misstatement risks is optimized, and the audit path is optimized by an improved annealing algorithm to ensure the consistency and standardization of the audit process. At the same time, the present invention provides seamless integration with other systems, improves collaborative work efficiency, reduces audit costs, and ensures the stability and reliability of audit quality.

Description

Technical Field

[0001] The present invention relates to the field of robot auditing technology, and in particular to a method and system for automated process management of robot auditing operations. Background Art

[0002] Audit standards may vary from project to project, but some audit systems fail to provide clear guidance on audit standards, resulting in a lack of unified standards for auditors in actual operations. At the same time, the audit system may need to be integrated with other systems, such as financial systems and business systems, but some current audit systems have technical barriers to integration. In scenarios where multiple departments and teams work together, some audit systems may not be able to provide effective collaborative tools or platforms, resulting in low efficiency in collaborative work.

[0003] Therefore, it is necessary to improve the efficiency and accuracy of audit work, reduce audit costs, and ensure the stability and consistency of audit quality through automated processes. At present, there is an urgent need for a robot technology to automate data collection, analysis, processing and reporting, including total project investment estimates, infrastructure operation and maintenance service estimates, etc. It is used to audit the construction, operation and maintenance, and fund use of the project. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for managing the automated process of robot auditing operations, comprising:

[0006] Collect data information of the audit target object through the robot, extract features of the data information, and construct audit nodes using the extracted features;

[0007] The characteristics of each audit node are analyzed through a neural network model, and the nodes identified as internal control deficiencies and nodes identified as material misstatement risks are used to construct an internal control matrix and a node map of material misstatement risks;

[0008] By analyzing the internal control matrix, the decision factor of the risk of material misstatement is obtained; taking multiple nodes of the risk of material misstatement as starting points, the improved annealing algorithm is used to perform path optimization in the node map of the risk of material misstatement to obtain the audit path of the risk of material misstatement;

[0009] Identify the audit path of the material misstatement risk, obtain the robot's audit result on the audit node or audit node combination, and generate an audit result mark;

[0010] The decision factor represents the impact of internal control deficiencies on the risk of material misstatement; the size of the decision factor is proportional to the degree of internal control deficiencies, and the decision factor acts on the analysis process of the risk of material misstatement, so that the risk of occurrence of material misstatement is proportional to the size of the decision factor.

[0011] As a preferred solution of the method for automated process management of robot auditing operations described in the present invention, wherein: the audit target objects include the current and historical reports, internal controls and other affairs of the audit object;

[0012] The other matters are related matters that need to be reviewed during the audit work, including but not limited to major transactions and special matters, the industry criticality of the audited entity, the results of the confirmation letters, and the time of occurrence of each transaction;

[0013] The data information includes all the data and information in the audit target object.

[0014] As a preferred solution of the method for automated process management of robot auditing operations described in the present invention, the feature extraction of the data information includes establishing the features of the single entry for each single entry in the accounting entry: ;

[0015] in, represents the amount of the jth single entry in the ith compound entry, Indicates the internal control involved in the jth single entry in the ith compound entry. Indicates other transaction information involved in the jth single entry in the ith compound entry;

[0016] The construction of the audit node includes taking the single entry as a node in the audit process, assigning the characteristics of the single entry to each node, and obtaining the smallest unit of the audit work.

[0017] As a preferred solution of the method for automated process management of robot auditing operations described in the present invention, the internal control matrix includes rows representing business types and columns representing internal control schemes, and constructing a matrix:

[0018] Features business processes and control activities, define a matrix To represent the various elements of internal control:

[0019]

[0020] in, Indicates whether the nth control activity in the mth business process is effective, with effective control recorded as 1 and ineffective control recorded as 0; It takes the records of each control activity of each business as input and obtains the analysis results through the trained random forest;

[0021] By analyzing the internal control matrix, specifically including obtaining the Item, will Match with the business process to get The business process in which it is located;

[0022] set up The business process is located in the yth row of the matrix. Through the comprehensive assessment of whether the control activities in the yth row are effective, the assessment result is Y:

[0023]

[0024] in, ; represents the number of control activities required in the business process in row y; Represents a selection function that selects the control effectiveness in the business process in row y. If represents the control activities required in the business process of row y, the output result is ;like It represents the control activities that are not required in the business process of row y, and the output result is 0.

[0025] As a preferred solution of the method for automated process management of robot audit work of the present invention, wherein: the node map of the risk of material misstatement includes aggregating all audit nodes, connecting them to indicate the association relationship between them, and forming a map of audit nodes;

[0026] According to the accounting order of the single entries in the compound entry, the connection between the audit nodes is given a direction, the positive direction is the accounting order, and the negative direction is the opposite direction of the accounting order.

[0027] As a preferred solution of the method for automated process management of robot auditing operations of the present invention, the decision factor specifically includes: - minus the evaluation result of the business process;

[0028] Decision factors for the business process of the jth single entry in the ith compound entry equal ;

[0029] The audit path for the risk of material misstatement includes evaluating each audit node through a deep feedforward neural network; the input is: , CB; Output: Estimation result of the probability of material misstatement risk at each audit node ;like , then make the current audit node a starting point of the audit path; if , no processing is done; obtain all audit nodes that are the starting points of the audit path, and for each path starting point, generate a path according to the negative direction of the connection line in the node map;

[0030] CB is the financial statements of previous years. represents the risk probability threshold of material misstatement of the jth single entry in the ith compound entry;

[0031] The improved annealing algorithm includes setting an initial temperature and cooling rate α;

[0032] Starting from the path starting point, let the current path be , represents a path containing several nodes;

[0033] Expand from the end point of the current path to the next possible node to form a new path candidate , added to the current path to form a new path candidate;

[0034] During the path selection process, the GBDT model is used to analyze the risk of material misstatement reflected by all nodes in the path. At the same time, the identification metric of the initial path is obtained, including the loss function and accuracy of the model.

[0035] Calculate new path The degree of improvement of the GBDT model: if the accuracy of the model is improved or the loss is reduced, the new path is accepted; otherwise, the new path is rejected;

[0036] The exploration process is controlled using the temperature parameter in the annealing algorithm, and the temperature is updated according to the formula:

[0037]

[0038] in, Indicates the updated temperature, Indicates the temperature before updating;

[0039] When the temperature drops to the threshold , or when the model performance has not been improved in multiple iterations, or there is no path to choose from, stop the annealing process;

[0040] In all iterations of the annealing algorithm, record the path that helps the GBDT model the most ;

[0041] By optimizing multiple starting points, multiple paths are obtained, which are expressed as ;

[0042] in, Represents the path obtained by optimizing the qth starting point.

[0043] As a preferred solution of the robot auditing automation process management method of the present invention, the audit results include: using the trained GBDT model, inputting the , and financial statements for previous years;

[0044] The output is: the robot’s evaluation result for each audit path, indicating whether it is identified as a material misstatement;

[0045] Insert a mark into the audit node or audit node combination contained in the path; the mark content includes the combination information of the audit node and the assessment result of the material misstatement risk.

[0046] A robot auditing automation process management system using the method of the present invention, wherein:

[0047] A collection unit collects data information of the audit target object through a robot, extracts features of the data information, and constructs an audit node using the extracted features;

[0048] The map construction unit analyzes the characteristics of each audit node through a neural network model, and constructs an internal control matrix and a node map of material misstatement risks using nodes identified as internal control deficiencies and nodes identified as material misstatement risks, respectively;

[0049] A path optimization unit, which obtains a decision factor of a material misstatement risk by analyzing the internal control matrix; takes multiple nodes of a material misstatement risk as starting points, and uses an improved annealing algorithm to perform path optimization in a node map of the material misstatement risk to obtain an audit path of the material misstatement risk;

[0050] The identification unit identifies the audit path of the material misstatement risk, obtains the robot's audit result on the audit node or audit node combination, and generates an audit result mark.

[0051] A computer device comprises: a memory and a processor; the memory stores a computer program, wherein: the processor implements the steps of any one of the methods of the present invention when executing the computer program.

[0052] A computer-readable storage medium stores a computer program, wherein: when the computer program is executed by a processor, the steps of any one of the methods of the present invention are implemented.

[0053] Beneficial effects of the present invention: The robot audit work automation process management method provided by the present invention realizes the automation process of audit work, including data collection, analysis, processing and automatic generation of reports, which significantly improves the audit efficiency and accuracy. Through the combination of neural network and GBDT model, the identification of material misstatement risks is optimized, and the audit path is optimized by improved annealing algorithm to ensure the consistency and standardization of the audit process. At the same time, the present invention provides seamless integration with other systems, improves collaborative work efficiency, reduces audit costs, and ensures the stability and reliability of audit quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0055] Figure 1 An overall flow chart of a method for automated process management of robot audit tasks provided for the first embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0057] Example 1, reference Figure 1 , as an embodiment of the present invention, provides a method for automated process management of robot auditing operations, comprising:

[0058] S1: Collect data information of the audit target object through the robot, extract features of the data information, and construct audit nodes using the extracted features.

[0059] Furthermore, the audit target object includes the audit object's current and historical financial statements, internal controls and other affairs; the other affairs are related matters that need to be reviewed in the audit work, including but not limited to major transactions and special matters, the industry criticality of the audited unit, the results of confirmations, and the time of occurrence of each transaction; the data information includes all data and information in the audit target object.

[0060] It should be said that collecting financial statements is one of the core tasks of auditing. The purpose is to analyze the accuracy, completeness and compliance of financial statements and understand the financial performance and trends of enterprises in different years. By comparing historical data, potential anomalies and major misstatement risks can be identified. The purpose of collecting internal control-related data is to evaluate the effectiveness and soundness of the internal control of the audited unit. Effective internal control can reduce the risk of misstatement in financial statements and ensure the safety of assets and the normal operation of business. By reviewing internal control deficiencies, potential major risks in financial statements can be identified. Major transactions and special matters often have a significant impact on financial status. The purpose of collecting these data is to confirm whether these matters are compliant and transparent, and whether there are potential financial misstatements or fraud risks. The particularity of the industry may affect the preparation standards and risk assessment standards of financial statements. Therefore, collecting key industry data helps to provide audit background information; the confirmation results are used to verify the authenticity and accuracy of the accounts. By collecting the above data and information, auditors can fully understand the financial and operating status of the enterprise, so as to accurately assess the risk of major misstatements and ensure the reliability and accuracy of the audit results.

[0061] Extracting features from the data information includes establishing features of each single entry in the accounting entries: .in, represents the amount of the jth single entry in the ith compound entry, Indicates the internal control involved in the jth single entry in the ith compound entry. Represents other transaction information involved in the jth single entry in the ith compound entry.

[0062] It should be noted that the accounting entries are automatically obtained from the system. If there are no accounting entries, the neural network is used to build accounting entries for each transaction. And the single entries are disassembled from the built compound entries.

[0063] The construction of the audit node includes taking the single entry as a node in the audit process, assigning the characteristics of the single entry to each node, and obtaining the smallest unit of the audit work.

[0064] S2: Analyze the characteristics of each audit node through the neural network model, and use the nodes identified as internal control deficiencies and the nodes identified as material misstatement risks to construct the internal control matrix and the node map of material misstatement risks.

[0065] Furthermore, the internal control matrix includes rows representing business types and columns representing internal control plans, constructing the matrix:

[0066] Features business processes and control activities, define a matrix To represent the various elements of internal control:

[0067]

[0068] in, Indicates whether the nth control activity in the mth business process is effective, with effective control recorded as 1 and ineffective control recorded as 0; The record of each control activity of each business is taken as input, and the analysis results are obtained through the trained random forest.

[0069] By analyzing the internal control matrix, specifically including obtaining the Item, will Match with the business process (each entry must have a corresponding business process, and matching can be achieved through preset matching rules), and get The business process in which it is located.

[0070] set up The business process is in the yth row of the matrix. Through a comprehensive assessment of whether the control activities in the yth row are effective, the assessment result is Y:

[0071] .

[0072] in, ; represents the number of control activities required in the business process in row y; Represents a selection function that selects the control effectiveness in the business process in row y. If represents the control activities required in the business process of row y, the output result is ;like The output is 0 if it represents the control activities that are not required in the business process of row y. (It should be noted that some internal controls are invalid controls and cannot play the role of supervising transactions or financial statement data. For example, when measuring inventory, regular inventory counting is an effective control, but the sign-in policy has no control effect in inventory counting. At this time, only the effective part is measured by selection).

[0073] It is worth mentioning that by inputting the control activity records of each business into the trained random forest model, the effectiveness of each control activity can be automatically evaluated. Random forests can handle complex and nonlinear relationships and automatically identify whether certain control activities are effective, thereby improving audit efficiency and reducing errors in manual judgment. By matching audit nodes with business processes, it is possible to accurately locate which business processes have failures or defects in control activities. By analyzing each business process (row) in the matrix, the effectiveness distribution of control activities in each business process can be systematically understood, and potential problems can be quickly found. By comprehensively evaluating the effectiveness of control activities in row y (i.e., a specific business process), the overall control effectiveness index Y is calculated. This design provides a quantitative standard for evaluating the overall internal control effectiveness of a business process by integrating the effectiveness of all relevant control activities into a single result.

[0074] The node map of material misstatement risk includes aggregating all audit nodes, indicating the correlation between them through connecting lines, and forming a map of audit nodes; according to the accounting order of single entries in the compound entry, the connecting lines between the audit nodes are given directions, the positive direction is the accounting order, and the negative direction is the opposite direction of the accounting order.

[0075] By aggregating the audit nodes and using lines to represent the associations between the nodes, a clear risk association diagram is constructed. This visualization method helps auditors quickly understand the associations between different audit nodes. For example, a node may have direct or indirect associations with multiple other nodes, which may mean that a local problem may affect multiple business links or transactions. By using the accounting order of single entries in compound entries as a reference, the lines between nodes are given directionality. The distinction between the positive and reverse directions of the accounting order enables auditors to clearly understand how risks are transmitted from one node to the next in the business process. If a node has a significant risk of misstatement, this design can help determine the propagation path of the risk throughout the business process, thereby identifying which subsequent nodes may be affected.

[0076] The map design not only shows the relationship between nodes, but also helps the system or auditors determine the optimal audit path through directional connections. Using this map, you can optimize the path through optimization algorithms (such as annealing algorithms), ensure that the audit process covers all potential high-risk nodes, and reasonably arrange the order of audits according to the accounting order to improve audit efficiency.

[0077] S3: Analyze the internal control matrix to obtain the decision factors of material misstatement risk; take multiple nodes of material misstatement risk as starting points, use the improved annealing algorithm to perform path optimization in the node map of material misstatement risk, and obtain the audit path of material misstatement risk.

[0078] The decision factor represents the impact of internal control deficiencies on the risk of material misstatement; the size of the decision factor is proportional to the degree of internal control deficiencies, and the decision factor acts on the analysis process of the risk of material misstatement, so that the risk of occurrence of material misstatement is proportional to the size of the decision factor.

[0079] The decision factor specifically includes: minus the assessment result of the business process. It should be noted that the judgment of the risk of material misstatement needs to refer to "whether there are defects in internal control" to a certain extent. Through the robot audit process, this factor is calculated through the decision factor, which can reflect the rigor and logic of the audit process.

[0080] Decision factors for the business process of the jth single entry in the ith compound entry equal The audit path for the risk of material misstatement includes evaluating each audit node through a deep feedforward neural network; the input is: , CB; Output: Estimation result of the probability of material misstatement risk at each audit node ;like , then make the current audit node a starting point of the audit path; if , no processing is done; obtain all audit nodes that are the starting points of the audit path, and for each path starting point, generate a path in the negative direction of the line in the node map. CB is the financial statements of previous years, It represents the risk probability threshold of material misstatement of the jth single entry in the i-th compound entry.

[0081] The purpose of designing the decision factor is to assess the impact of internal control deficiencies on the risk of material misstatement by quantifying the degree of deficiencies in internal control. The size of the decision factor is proportional to the internal control deficiencies, which means that the weaker the internal control, the larger the decision factor, which represents a higher potential risk of material misstatement. Through this design, the audit process can comprehensively and accurately assess the contribution of control deficiencies to the risk of misstatement, thereby identifying high-risk nodes in a targeted manner.

[0082] By evaluating the risk probability of material misstatement of each audit node through a deep feedforward neural network model, it is possible to automatically determine which nodes are potential high-risk points. This is a preliminary judgment on the risk of misstatement of the audit node. Neural networks perform well in identifying complex nonlinear risk characteristics, and can dynamically generate risk assessment results for each node by integrating historical data (such as financial statements over the years). When the risk probability of an audit node is greater than the set threshold (that is, it exceeds a certain risk probability threshold), the system uses the node as the starting point of the audit path for material misstatement risk. Combined with the negative direction of the connection in the node map (indicating the accounting order of compound entries), a path can be automatically generated to ensure that the audit covers all potential high-risk areas.

[0083] The improved annealing algorithm includes setting an initial temperature and cooling rate α.

[0084] Starting from the path starting point, let the current path be , indicating a path containing several nodes.

[0085] Expand from the end point of the current path to the next possible node to form a new path candidate , added to the current path to form a new path candidate.

[0086] During the path selection process, the GBDT model is used to analyze the significant misstatement risks reflected by all nodes in the path; at the same time, the identification measure of the initial path is obtained, including the loss function and accuracy of the model.

[0087] Calculate new path For the improvement of the GBDT model, if the accuracy of the model is improved or the loss is reduced, the new path is accepted; otherwise, the new path is rejected.

[0088] The exploration process is controlled using the temperature parameter in the annealing algorithm, and the temperature is updated according to the formula:

[0089]

[0090] in, Indicates the updated temperature, Indicates the temperature before updating.

[0091] When the temperature drops to the threshold , or when the model performance has not been improved in multiple iterations, or when there is no selectable path (this is generally the earliest accounting entry, such as the accounting entry when the transaction occurs. There is no other transaction-related content before this entry. At this time, no new path can be found by tracing back in the negative direction of the accounting entry), the annealing process is stopped.

[0092] In all iterations of the annealing algorithm, record the path that helps the GBDT model the most .

[0093] By optimizing multiple starting points, multiple paths are obtained, which are expressed as in, Represents the path obtained by optimizing the qth starting point.

[0094] The audit path is optimized through the improved annealing algorithm, thereby improving the efficiency and accuracy of identifying the risk of material misstatement. The specific design combines the evaluation ability of the GBDT model with the global search ability of the annealing algorithm to explore multiple audit paths in an automated manner to improve the quality and efficiency of the audit.

[0095] First, the initial temperature and cooling rate are set, and the risk of material misstatement of each new path is calculated through path search and node expansion with the support of the GBDT model. The GBDT model analyzes the node information in the path and uses accuracy and loss functions to quantify the quality of the path to ensure that the selected path can effectively identify potential misstatement risks. This performance-based path selection mechanism ensures that the model focuses on analyzing the most risky paths during the audit process.

[0096] Secondly, the exploration process is controlled by the temperature parameter in the annealing algorithm, which allows suboptimal solutions to be accepted in the early stages to avoid falling into the local optimal solution. As the temperature decreases, the exploration gradually focuses on the path with the best performance. Through repeated iterations and combined with the relationship between audit nodes, the algorithm can effectively avoid path duplication and unnecessary exploration.

[0097] Finally, the algorithm can automatically identify the boundary conditions of accounting entries and find the optimal audit path among multiple paths. This multi-path optimization design ensures the comprehensiveness and pertinence of the audit work, ensures that high-risk nodes are covered to the maximum extent with limited audit resources, and improves the overall audit efficiency and effectiveness.

[0098] S4: Identify the audit path of the material misstatement risk, obtain the robot's audit results for the audit node or audit node combination, and generate an audit result mark.

[0099] Using the trained GBDT model, input the , as well as financial statements for previous years; GBDT is an ensemble learning method that gradually improves the performance of the model by building a series of weak learners (usually decision trees). Each newly added tree is designed to reduce the prediction error (residual) of the previous round of models. By gradually optimizing the loss function through the gradient descent algorithm, GBDT can learn complex patterns and improve the accuracy of classification or regression tasks.

[0100] The output is: the robot's evaluation results for each audit path, indicating whether it is identified as a material misstatement. A mark is inserted into the audit node or audit node combination contained in the path; the mark content includes the combination information of the audit node and the evaluation results of the risk of material misstatement.

[0101] By leveraging the trained GBDT model, the robot can automatically evaluate each node in the audit path and identify the risk of material misstatement. This reduces the need for human involvement, makes the audit process highly automated, and greatly improves audit efficiency. Auditors only need to focus on high-risk areas marked by the system, thereby optimizing the workflow.

[0102] The design allows for combined analysis of audit nodes and their combined information, and flexible adjustments for different business scenarios and audit paths. GBDT is able to learn complex risk patterns from complex, multi-dimensional audit data, which enables the method to adapt to different types of audit tasks and provide dynamic risk assessment.

[0103] The generated audit result tags bind the risk assessment and node combination information to ensure the traceability of the audit process. The assessment of each audit node and path is clearly recorded, and auditors can quickly view high-risk areas and understand the analytical basis behind them, which improves the transparency of audit decisions. Automated risk assessment not only improves efficiency, but also reduces manual input, thereby reducing audit costs. Robots are able to process large amounts of audit data and guide audit work through precise tags, avoiding possible redundant operations or errors when manually processing data.

[0104] On the other hand, this embodiment also provides a robot audit job automation process management system, which includes: a collection unit, which collects data information of the audit target object through a robot, extracts features of the data information, and constructs an audit node using the extracted features.

[0105] The map construction unit analyzes the characteristics of each audit node through a neural network model, and constructs an internal control matrix and a node map of material misstatement risks using nodes identified as internal control deficiencies and nodes identified as material misstatement risks, respectively.

[0106] The path optimization unit obtains the decision factors of the risk of material misstatement by analyzing the internal control matrix; takes multiple nodes of the risk of material misstatement as starting points, uses an improved annealing algorithm to perform path optimization in the node map of the risk of material misstatement, and obtains the audit path of the risk of material misstatement.

[0107] The identification unit identifies the audit path of the material misstatement risk, obtains the robot's audit result on the audit node or audit node combination, and generates an audit result mark.

[0108] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0109] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0110] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0111] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0112] Example 2 is an embodiment of the present invention, which provides a method for automated process management of robot audit operations. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0113] The comparative experiment is based on the financial statements of a manufacturing company in 2021 and 2022, and the audit contents involved include major financial items such as sales revenue, procurement expenditure, assets and liabilities. The experiment mainly evaluates the performance of three audit methods: manual audit, rule-driven audit system and the automated audit method of the present invention, focusing on testing the number of misstatements identified, the time required for audit and the overall audit efficiency.

[0114] The main process of the experiment is as follows:

[0115] Financial statement data preparation: The company's 2021 and 2022 financial statement data are used for testing. The data contains the company's sales revenue, procurement expenditure, assets and liabilities and other major financial data. Each annual financial statement contains approximately 2,500 financial transaction records, involving complex transaction structures and multiple business processes.

[0116] Manual audit: Two senior auditors will manually review the financial statements for 2021 and 2022 in accordance with international auditing standards. The audit process includes checking and analyzing financial data such as revenue, costs, accounts receivable, accounts payable, etc., to find misstatements and unusual transactions. Each auditor needs to check each item in the financial statements one by one and manually check whether there are any financial misstatements.

[0117] Rule-driven audit system: The rule-driven audit system automatically detects abnormal data by comparing fixed thresholds based on set audit rules, such as key indicators such as revenue-to-cost ratio, accounts receivable turnover, and debt-to-asset ratio. Although this method can quickly detect common anomalies, it may not be able to identify complex financial misstatements and nonlinear anomalies due to the fixed nature of the rules.

[0118] The automated audit method of the present invention: The present invention conducts a comprehensive analysis based on the GBDT model combined with financial data and internal control deficiency information, optimizes the audit path through an improved annealing algorithm, and evaluates the misstatement risk of each accounting entry. The system automatically generates the best audit path, prioritizes the review of high-risk nodes, and gradually iterates to the optimal result. Through the combination of machine learning models and annealing algorithms, the risk of misstatement can be accurately identified in a relatively short period of time.

[0119] The goal of the experiment is to verify the innovation and advantages of the present invention in terms of efficiency and accuracy by comparing the number of misstatement identification and audit time of the three audit methods. The recorded data are shown in Table 1.

[0120] Table 1 Data comparison table

[0121]

[0122] By comparing the data in the table, it can be clearly seen that the present invention has significantly improved the number of misstatement identification and audit time.

[0123] As can be seen from the data table, the present invention identified 11 and 12 misstatements in the financial statements for 2021 and 2022, respectively, while manual audits identified 8 and 9, respectively, and the rule-driven audit system only identified 5 and 6. The present invention can more accurately identify the risk of misstatement, mainly due to the multi-dimensional analysis capabilities of the GBDT model combined with financial data, historical data and internal control. This shows that compared with manual audits and rule-driven audit systems, the present invention can more comprehensively detect complex misstatements.

[0124] In terms of audit time, the automated audit system of the present invention significantly improves efficiency. For the 2021 financial statements, manual auditing took 15 hours, rule-driven auditing took 8 hours, and the present invention only took 2.5 hours. In the 2022 financial statements, manual auditing took 16 hours, rule-driven auditing took 7 hours, and the present invention only took 2.7 hours. This shows that the present invention can quickly complete the audit work of large-scale data through automated path generation and optimization, significantly reducing the time required for auditing and reducing the redundant workload in manual operations.

[0125] Although manual auditing can cover more business processes through manual experience, it is time-consuming and prone to human omissions because it requires checking each item one by one. Rule-driven auditing systems rely on preset fixed rules and are difficult to capture complex misreporting patterns, resulting in many omissions. The present invention significantly improves audit efficiency while maintaining high-precision misreporting identification by combining a machine learning model with an improved annealing algorithm.

[0126] The audit system of the present invention can not only automatically generate audit paths and intelligently assess misstatement risks, but also adaptively adjust audit strategies, making the audit process more efficient and intelligent. Compared with the existing technology, it significantly reduces human intervention and human errors, and can cover more high-risk areas, improving the overall quality of the audit.

[0127] By testing the path optimization technology based on the improved annealing algorithm and GBDT model in the present invention, its effectiveness in path selection, internal control assessment, and identification of material misstatement risks in the audit process is verified. The experiment will collect and record key data in the invention process, with a special focus on internal control assessment, decision factor calculation, and misstatement risk assessment of nodes in the path.

[0128] Experimental subjects: Financial data and business process audit records of a manufacturing company over the past three years, including internal control scores, node amounts, and business process information required for risk assessment.

[0129] Seven major audit nodes were obtained, and the characteristics of each node including internal control score, decision factor and amount were recorded, as shown in Table 2.

[0130] Using the GBDT model in the present invention, internal control evaluation is performed on each audit node, and its decision factor is calculated. According to the decision factor and control score of the node, the path is dynamically generated and optimized through the annealing algorithm. The initial temperature is 100, the cooling rate is 0.95, and the path is optimized for each iteration. The path generation rule follows the priority selection of high-risk nodes. The system evaluates the overall risk of the generated audit path, and comprehensively calculates the risk of material misstatement of the path through the risk factors of all nodes in the path. Select 2 paths and record the data as shown in Table 3.

[0131] Table 2 Node record table

[0132]

[0133] Table 3 Path record table

[0134]

[0135] First, it can be seen from Table 2 that the present invention scores the internal control of the audit nodes through the GBDT model, and combined with the calculation of the decision factor, it can accurately evaluate the risk contribution of each node to the overall audit path. The control score and decision factor of the node accurately reflect the effectiveness of the internal control in each business process and its impact on the risk of misstatement. For example, the internal control scores of nodes 1, 3, and 5 are relatively high, 0.75, 0.90, and 0.85, respectively, indicating that these nodes have large control deficiencies in the business process, and their decision factors further quantify the impact of control deficiencies on the overall path risk.

[0136] As can be seen from Table 3, the present invention can dynamically generate targeted high-risk audit paths through internal control scores and decision factors, rather than blindly covering all nodes. This path optimization strategy significantly improves audit efficiency and reduces unnecessary low-risk audit work. It can intelligently select important nodes and evaluate the misstatement risk of the entire path, allowing auditors to focus more on high-risk areas, thereby improving the accuracy and reliability of audit results.

[0137] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for automated process management of robot audit work, characterized in that: include: Collect data information of the audit target object through the robot, extract features of the data information, and construct audit nodes using the extracted features; The characteristics of each audit node are analyzed through a neural network model, and the nodes identified as internal control deficiencies and nodes identified as material misstatement risks are used to construct an internal control matrix and a node map of material misstatement risks; By analyzing the internal control matrix, the decision factors of material misstatement risk are obtained; Taking multiple nodes of material misstatement risk as starting points, using an improved annealing algorithm to perform path optimization in the node map of material misstatement risk, and obtaining an audit path of material misstatement risk; Identify the audit path of the material misstatement risk, obtain the robot's audit result on the audit node or audit node combination, and generate an audit result mark; The decision factor represents the impact of internal control deficiencies on the risk of material misstatement; the size of the decision factor is proportional to the degree of internal control deficiencies, and the decision factor acts on the analysis process of the risk of material misstatement so that the risk of material misstatement is proportional to the size of the decision factor; The improved annealing algorithm includes setting an initial temperature and cooling rate α; Starting from the path starting point, let the current path be , represents a path containing several nodes; Expand from the end point of the current path to the next possible node to form a new path candidate , added to the current path to form a new path candidate; During the path selection process, the GBDT model is used to analyze the risk of material misstatement reflected by all nodes in the path. At the same time, the identification metric of the initial path is obtained, including the loss function and accuracy of the model. Calculate new path The degree of improvement of the GBDT model: if the accuracy of the model is improved or the loss is reduced, the new path is accepted; otherwise, the new path is rejected; The exploration process is controlled using the temperature parameter in the annealing algorithm, and the temperature is updated according to the formula: ， in, Indicates the updated temperature, Indicates the temperature before updating; When the temperature drops to the threshold , or when the model performance has not been improved in multiple iterations, or there is no path to choose from, stop the annealing process; In all iterations of the annealing algorithm, record the path that helps the GBDT model the most ; By optimizing multiple starting points, multiple paths are obtained, which are expressed as ; in, Represents the path obtained by optimizing the qth starting point.

2. The method for automated process management of robot auditing operations according to claim 1, characterized in that: The audit objectives include the audit object's current and historical financial statements, internal controls and other matters; The other matters are related matters that need to be reviewed during the audit work, including but not limited to major transactions and special matters, the industry criticality of the audited entity, the results of the confirmation letters, and the time of occurrence of each transaction; The data information includes all the data and information in the audit target object.

3. The method for automated process management of robot auditing operations as claimed in claim 2, characterized in that: Extracting features from the data information includes establishing features of each single entry in the accounting entries: ; in, represents the amount of the jth single entry in the ith compound entry, Indicates the internal control involved in the jth single entry in the ith compound entry. Indicates other transaction information involved in the jth single entry in the ith compound entry; The construction of the audit node includes taking the single entry as a node in the audit process, assigning the characteristics of the single entry to each node, and obtaining the smallest unit of the audit work.

4. The method for automated process management of robot auditing operations according to claim 1, characterized in that: The internal control matrix includes rows representing business types and columns representing internal control plans, and constructs the matrix: Features business processes and control activities, define a matrix To represent the various elements of internal control: ， in, Indicates whether the nth control activity in the mth business process is effective, with effective control recorded as 1 and ineffective control recorded as 0; It takes the records of each control activity of each business as input and obtains the analysis results through the trained random forest; By analyzing the internal control matrix, specifically including obtaining the Item, will Match with the business process to get The business process in which it is located; set up The business process is in row y of the matrix. Through comprehensive evaluation of whether the control activities in row y are effective, the evaluation result is Y: ， in, ; represents the number of control activities required in the business process in row y; Represents a selection function that selects the control effectiveness in the business process in row y. If represents the control activities required in the business process of row y, the output result is ;like It represents the control activities that are not required in the business process of row y, and the output result is 0.

5. The method for automated process management of robot auditing operations according to claim 1, characterized in that: The node map of the risk of material misstatement includes aggregating all audit nodes, connecting them to indicate the association between them, and forming a map of audit nodes; According to the accounting order of the single entries in the compound entry, the connection between the audit nodes is given a direction, the positive direction is the accounting order, and the negative direction is the opposite direction of the accounting order.

6. The method for automated process management of robot auditing operations according to claim 1, characterized in that: The decision factor specifically includes: - minus the evaluation result of the business process; Decision factors for the business process of the jth single entry in the ith compound entry equal ; The audit path for the risk of material misstatement includes evaluating each audit node through a deep feedforward neural network; the input is: , CB; Output: Estimated probability of material misstatement risk for each audit node ;like , then make the current audit node a starting point of the audit path; if , no processing is done; obtain all audit nodes that are the starting points of the audit path, and for each path starting point, generate a path according to the negative direction of the connection line in the node map; CB is the financial statements of previous years. It represents the risk probability threshold of material misstatement of the jth single entry in the i-th compound entry.

7. The method for automated process management of robot auditing operations according to claim 1, characterized in that: The audit results include using the trained GBDT model to input the , and financial statements for previous years; The output is: the robot’s evaluation result for each audit path, indicating whether it is identified as a material misstatement; Insert a mark into the audit node or audit node combination contained in the path; the mark content includes the combination information of the audit node and the assessment result of the material misstatement risk.

8. A robot auditing automation process management system using the method according to any one of claims 1 to 7, characterized in that: A collection unit collects data information of the audit target object through a robot, extracts features of the data information, and constructs an audit node using the extracted features; The map construction unit analyzes the characteristics of each audit node through a neural network model, and constructs an internal control matrix and a node map of material misstatement risks using nodes identified as internal control deficiencies and nodes identified as material misstatement risks, respectively; A path optimization unit, which obtains decision factors of material misstatement risk by analyzing the internal control matrix; Taking multiple nodes of material misstatement risk as starting points, using an improved annealing algorithm to perform path optimization in the node map of material misstatement risk, and obtaining an audit path of material misstatement risk; The identification unit identifies the audit path of the material misstatement risk, obtains the robot's audit result on the audit node or audit node combination, and generates an audit result mark.

9. A computer device comprising: A memory and a processor; the memory stores a computer program, wherein the processor implements the steps of any method as claimed in claim 1 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images