Method and device for making auditing manuscript based on AI large model

Abstract

The invention discloses a method and device for making an audit manuscript based on an AI large model. The method comprises the steps that a manuscript auditing tool selection model of the manuscript auditing robot identifies a corresponding manuscript making intention based on an input manuscript auditing requirement, and a corresponding manuscript auditing tool is screened and extracted from an auditing tool library according to the manuscript making intention; the auditing manuscript creation model of the auditing manuscript robot creates an auditing manuscript by using the auditing manuscript tool based on the acquired auditing data of the related subject. The technical problem of low auditing efficiency in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method and device for producing audit working papers based on an AI big model. Background Art

[0002] In the traditional audit process, auditors usually rely on manual methods to prepare audit working papers. This process is not only time-consuming, but also easily affected by human factors, resulting in low audit efficiency. With the continuous development of information technology, the audit industry urgently needs to use automated tools to improve audit efficiency and accuracy. However, most existing audit tools lack intelligence and cannot automatically select and generate corresponding audit working papers according to specific audit needs. Auditors still need to perform a lot of repetitive manual operations. This not only increases the workload of auditors, but also increases the risk of errors in the audit process.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present invention provide a method and device for producing audit working papers based on an AI big model, so as to at least solve the technical problem of low audit efficiency in the prior art.

[0005] According to one aspect of an embodiment of the present invention, a method for producing audit working papers based on an AI big model is provided, comprising: the audit working paper tool selection model of the audit working paper robot identifies the corresponding audit working paper production intention based on the input audit working paper requirements, and screens and extracts the corresponding audit working paper tools from the audit tool library according to the audit working paper production intention; the audit working paper production model of the audit working paper robot produces audit working papers based on the acquired audit data of relevant subjects, using the audit working paper tools.

[0006] According to another aspect of an embodiment of the present invention, there is also provided a device for producing audit working papers based on an AI big model, namely, an audit working paper robot, comprising: an audit working paper tool selection model, configured to identify corresponding intentions based on input audit working paper requirements, and screen and extract corresponding audit working paper tools from an audit tool library according to the corresponding intentions; an audit working paper production model, configured to produce audit working papers using the audit working paper tools based on the acquired audit data of relevant subjects.

[0007] In an embodiment of the present invention, the audit working paper tool selection model of the audit working paper robot identifies the corresponding working paper production intention based on the input audit working paper demand, and screens and extracts the corresponding audit working paper tool from the audit tool library according to the working paper production intention; the audit working paper production model of the audit working paper robot uses the audit working paper tool to produce audit working papers based on the audit data of the relevant subjects obtained. Through the above scheme, the technical problem of low audit efficiency in the prior art is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0009] Figure 1 is a flowchart of a method for preparing audit working papers based on an AI big model according to an embodiment of the present invention;

[0010] Figure 2 This is an architectural diagram of an audit working paper robot based on an AI big model according to an embodiment of the present invention;

[0011] Figure 3 is a flowchart of another method for preparing audit working papers based on an AI big model according to an embodiment of the present invention;

[0012] Figure 4 It is the specific design and implementation process of the audit working paper making function according to the embodiment of the present invention;

[0013] Figure 5 is a flowchart of another method for preparing audit working papers based on an AI big model according to an embodiment of the present invention;

[0014] Figure 6 A schematic diagram of the structure of an electronic device suitable for implementing the embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0015] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0016] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0017] The audit working papers robot is used to execute the method of making audit working papers based on AI big model, such as Figure 1 As shown, the method includes:

[0018] Step S102, the audit working paper tool selection model of the audit working paper robot identifies the corresponding audit working paper preparation intention based on the input audit working paper requirements, and screens and extracts the corresponding audit working paper tools from the audit tool library according to the audit working paper preparation intention;

[0019] First, the audit working papers tool is constructed based on the subject categories and the contents of the audit working papers.

[0020] Next, the corresponding audit working paper production intention is identified based on the input audit working paper requirements, and the corresponding audit working paper tools are screened and extracted from the audit tool library according to the audit working paper production intention. This application identifies the corresponding audit working paper production intention based on the input audit working paper requirements, thereby improving the accuracy and efficiency of the audit process. By analyzing the input requirements through a machine learning model, different audit working paper requirements can be automatically identified, reducing the error rate of manual analysis and ensuring that relevant audit work can be carried out in a targeted manner.

[0021] The audit working paper tool selection model of the audit working paper robot identifies the working paper production intention based on the input audit working paper requirements to obtain a tuple including tool information and subject information; based on the tool information and subject information in the tuple, the corresponding audit working paper tool is screened and extracted from the audit tool library. Among them, the subject categories include main business income, main business cost, other income subjects, and other expense costs; the content of the audit working paper includes a detailed list, a monthly detailed list, a comparative list of the same period, an income cutoff test, and a voucher test. The present application combines tool information with subject information in the form of a tuple, further optimizing the accuracy of tool selection. The pairing mechanism provided by the present application enables the working paper tool and the audit content to be better linked, avoiding the risk of mismatch between tools and content, and improving the systematicness and accuracy of audit work.

[0022] Step S104: the audit working papers preparation model of the audit working papers robot prepares audit working papers using the audit working papers tool based on the acquired audit data of relevant subjects.

[0023] Specifically, first, based on the subject information in the binary group, the audit data is acquired through the audit data acquisition model; then, the data format of the audit data is set to JSON format by using the generation function of the large language model; finally, based on the audit data in the JSON format, the audit data in the JSON format is processed and calculated by using the audit working paper formulation function according to the output requirements of the audit working paper to obtain the processed audit data, and the processed audit data is displayed in the form of a table to obtain the audit working paper. The audit working paper production model includes different types of audit working paper formulation functions under each subject, and the input parameter of the audit working paper formulation function is the subject information in the binary group.

[0024] This application uses an audit data acquisition model to automatically collect audit data based on the subject information in the binary group, greatly improving the efficiency and accuracy of audit data acquisition. Compared with the traditional manual data collection method, this automated method not only reduces the possibility of human error, but also ensures the integrity and consistency of the data, especially showing significant advantages when processing large-scale and complex audit data. In addition, the audit data format is set to a standard JSON format using the generation function of the large language model, which is convenient for subsequent data processing and calculation. Finally, based on the audit data in JSON format, according to the output requirements of the audit working papers, the working papers are used to formulate functions to further process and calculate the data, and the data can be cleaned, classified, summarized and other operations can be accurately performed according to the audit rules, and finally generate audit results that meet the requirements. By displaying the processed audit data in a tabular form, not only the data is more intuitive and easy to understand, but also it is convenient for auditors to review and proofread the results, improving the efficiency and accuracy of audit working papers.

[0025] The following will take profit and loss accounts as an example to describe in detail the method by which the audit working papers robot produces audit working papers based on the AI ​​big model.

[0026] In view of the high repetitiveness of audit working papers in the current audit process of corporate financial statements, as well as the cumbersome number extraction involved in the production of audit working papers, this application designs an audit working paper robot to realize the automation of some audit procedures, and generates audit working papers to be output to auditors during the execution process. The audit working paper robot involves data extraction from multiple data tables such as account balance tables and chronological accounts in the process of executing tasks. The audit data acquisition model has been constructed in the above text, so the constructed audit data acquisition model will be used in the audit working paper robot. In addition, the audit working paper robot of this application will take profit and loss accounts as an example. Since there will be certain differences in the audit working papers between various accounts, the audit working paper production link needs to set the working papers of these different types of accounts in order to output the accurate audit working papers of the account. Moreover, profit and loss accounts are important performance indicators of corporate operating results and profitability, so profit and loss accounts are selected as an example to design the audit working paper robot. And in the subsequent work, the construction of audit working paper robots for asset, liability and equity accounts will be completed.

[0027] The audit working papers robot mainly includes two aspects. On the one hand, it is the construction of the audit working papers tool selection model, which is mainly based on the audit working papers requirements put forward by the auditors, identifying their intentions from the requirements, and screening and extracting audit working papers tools from the audit tool library according to the intentions. On the other hand, it is the construction of the audit working papers production model, whose main purpose is to obtain the data of relevant subjects through the audit data acquisition model for the selected audit tools, and to realize the production of audit working papers through the specified working papers preparation tools. The specific framework of the audit working papers robot is as follows: Figure 2 shown.

[0028] In this embodiment, the LangChain framework is still used to connect to the GPT-4 API to implement the call of the large language model, and the large language model is used to implement the draft tool selection model and obtain relevant audit data. The draft production involved in the draft production link mainly constructs the relevant draft generation function to realize the production of the audit draft form. Figure 3 As shown, the method comprises the following steps:

[0029] Step S302, audit working paper tool selection model construction.

[0030] The core of the work of the audit working paper robot is to generate working papers of what subjects and what contents to generate. Therefore, this application collects working paper generation tools of different subjects and different types by constructing an audit working paper tool library. By using a large language model to identify the intention of auditors' working paper requirements, a two-tuple in the form of "['tool', 'subject']" is obtained to call the corresponding tool from the audit working paper tool library. Since the audit working paper content requirements between different subjects are different, in the audit working paper tool library, "main business income" and "main business cost" are used as separate modules to construct the audit working paper tool, "other business income" is classified as "other income subjects", and "other business costs", "administrative expenses", "financial expenses" and "business taxes and surcharges" are classified as "other expense subjects". When constructing the audit working paper tool library, the corresponding audit working paper tool is constructed according to the subject category and the content of the audit working paper. In addition, the contents of the working papers in each subject will be added or deleted as the business content changes. Therefore, this application mainly focuses on constructing universal audit working papers such as "detailed list", "monthly detailed list", "same period comparison list", "income cutoff test" and "voucher test".

[0031] The present invention effectively integrates the generation tools of different subjects and working papers by constructing an audit working paper tool library, thereby improving the flexibility and accuracy of working paper generation. The large language model is used to identify the auditor's demand intentions, and the corresponding tools are automatically called in the form of a two-tuple in the form of "['tool', 'subject']", which significantly reduces the error rate of manual intervention. At the same time, according to the audit working paper content requirements between different subjects, the customized modular design ensures the matching and applicability of the content. The embodiment of the present application can dynamically adjust the content of the working papers to adapt to business changes, especially in the construction of universal audit working papers, which improves the systematicness and efficiency of audit work, thereby better supporting the work of auditors.

[0032] The correspondence between the subjects in the audit working papers tool library and the working papers contents is shown in Table 1 below.

[0033] Table 1 Correspondence between subjects and manuscript contents

[0034]

[0035] Based on the established main profit and loss accounts and the contents of the audit working papers for each account, the definition of the tools in the audit working paper tool library is completed, as shown in Table 2 below.

[0036] Table 2 Definition and main contents of audit working paper tools

[0037]

[0038] The audit working papers tool library will include detailed tables of main business income and costs, monthly details and audit working papers for cut-off tests. It also includes detailed tables of other income and other expenses, comparative tables for the same period and working papers related to voucher testing. The audit working papers also include income audit papers and expense audit papers.

[0039] The tool for generating audit working papers mainly combines several scattered working papers into a unified output. This application uses the API of GPT-4 to connect with the audit working paper tool library, and introduces the relevant tools and functions to the large language model in the form of prompt engineering for the large language model to learn. After learning and understanding various tools, the large language model will identify the intention of the audit working paper requirements input by the auditor into the large language model, and identify the tool types and main subject contents contained in the working paper requirements input by the auditor. For example, if the auditor's demand is "help me generate a detailed list of main business income", then the demand mainly includes two key points: "main business income" and "detailed list". Therefore, the large language model will focus on the main business income detailed list numbered "B" when selecting tools, and the subject is "main business income", so its output tuple is "['B', 'main business income']". When carrying out prompt engineering for the large language model, some examples are used to help the large language model better understand the relevant content, so as to generate tuples more accurately. The audit working papers generation tool of the present invention significantly improves the efficiency of audit working papers generation by merging multiple working papers and unifying the output. The GPT-4 API is used to connect with the audit working papers tool library, and the relevant tools and their functions are introduced to the large language model through the prompt project, so that it can effectively learn and identify the input needs of auditors. When the auditor makes a request such as "help me generate a detailed list of main business income", the large language model can quickly identify the two key elements of "main business income" and "detailed list", thereby generating the corresponding bigrams. In this way, it not only optimizes the accuracy of tool selection, but also helps the large language model to have a deeper understanding of the specific requirements of working paper generation through the guidance of examples. The intelligent processing method of the embodiment of the present application improves the efficiency of audit work, ensures the accuracy and relevance of the output results, and provides more reliable support for auditors.

[0040] The following Table 3 shows the prompt engineering content for the connection and interaction between the audit working paper tool library and the large language model.

[0041] Table 3 Interaction prompts between the audit working paper tool library and the big language model

[0042]

[0043]

[0044] Auditors only need to input the audit working papers they need to build into the model, and then use the audit working papers tool selection model to identify the tool type and the content of the subjects involved, and then select the corresponding tool to produce the audit working papers.

[0045] Step S304: constructing the audit working papers production model.

[0046] After the audit working paper robot uses the audit working paper tool selection model to obtain the audit tool binary, the corresponding audit working paper production function is selected according to the tool information and account information in the binary. The audit working paper production model is composed of different types of audit working paper production functions under each account. Table 4 below shows the audit working paper production functions corresponding to each audit working paper tool. The "sub_des" in each function is the value of the account name in the binary. The account name is used to obtain and calculate the relevant data of the account, and finally the audit working paper is generated through the construction of the table. In the tool, both "income audit working papers" and "expense audit working papers" are output as complete audit working papers. Therefore, when calling the two audit tools, the audit working papers will be fully output by calling various working paper production functions of the relevant sub-accounts.

[0047] Table 4 Audit working papers preparation function correspondence table

[0048]

[0049] The specific design and implementation process of the audit working paper production function is as follows Figure 4 As shown in the table above, the audit working papers preparation function for each account type is as follows Figure 4 The audit working papers production function mainly includes four parts. The first is to use the audit data acquisition model constructed in the previous section to automatically acquire audit-related data. The input of the data acquisition model is the subject in the binary group. Through the pre-set input template, you only need to input the subject into the data acquisition model to acquire the audit data related to the subject. Secondly, after retrieving the relevant data from the audit data acquisition model, use the generation function of the large language model to set the data format to JSON format for subsequent table conversion. After completing the conversion of JSON data into a table, use the output requirements of the audit working papers to realize the operation between data. Finally, the processed and completed data is displayed in the form of a table.

[0050] The main differences between different audit working papers production functions are in the data acquisition model and data processing and calculation. In the data acquisition model, the calculation method of account amounts is different. Income accounts use the method of total credit amount minus total debit amount to calculate the amount, while expense accounts use the method of debit amount minus credit amount to calculate the amount. At the same time, different types of audit working papers require different data extraction. The detailed table calculates the amount of the current period and the same period of the previous year according to different items in the account, and the same period comparison table and monthly detailed table will extract the amount of each month in the current period and the same period of the previous year according to different months. In terms of data processing and calculation, the specific calculation content and calculation formula of each audit working paper type are shown in Table 5.

[0051] Table 5 Data processing and calculation contents and formulas of various types of audit working papers

[0052]

[0053] Step S306, output the audit working papers.

[0054] Auditors input audit working papers related requirements into the audit working papers robot. The audit working papers tool selection model will identify the intention of the input requirements, generate tool and subject tuples, pass the tuples to the audit working papers production model to generate the corresponding audit working papers, and finally output the audit working papers in the form of a table.

[0055] The present application embodiment provides another method for producing audit working papers based on the AI ​​big model, such as Figure 5 As shown, the method comprises the following steps:

[0056] Step S502, obtaining audit data.

[0057] In the process of obtaining audit data, many types of tables are involved, such as financial statements, chronological accounts, account balance sheets, etc. Auditors need to extract corresponding information from these data sources. However, traditional audit data acquisition methods often face some challenges. First, due to the limited number of tokens of the GPT-4 API, when too many data acquisition task examples are input, it is easy to exceed the token limit and fail to fully output the content expected by the auditor. Secondly, due to the complexity of the task, GPT-4 may encounter difficulties in selecting appropriate examples, resulting in inaccurate output results.

[0058] In order to deal with these problems, this application proposes an example selection optimization strategy introduced in the LangChain framework. By adopting the semantic similarity example selector in LangChain, the examples most relevant to the questions asked can be intelligently screened out during the audit process. These examples are input into the GPT-4 model together with the specific data acquisition requirements provided by the auditors, thereby improving the accuracy of data extraction and effectively avoiding the token overlimit problem.

[0059] The core mechanism of this application is semantic similarity calculation. The auditor's query requirements are first converted into an embedding vector, denoted as Vquery. At the same time, each example in the sample library is also converted into an embedding vector, denoted as VSV. Then, the cosine similarity between the query embedding vector and the sample embedding vector is calculated to select the semantically closest example.

[0060] The formula is:

[0061]

[0062] Among them, S represents the sample set of the sample library, Vquery is the embedding vector converted from the query requirements provided by the auditor, and VS is the embedding vector corresponding to each sample in the sample library. By calculating the cosine similarity between the input query embedding vector and the embedding of each sample in the audit data acquisition sample library, the sample with the highest semantic similarity to the current audit requirement is selected. This method can ensure that the selected examples are highly relevant to the audit requirements, which helps to improve the accuracy of data acquisition.

[0063] In order to further improve the accuracy of sample selection, this scheme also introduces multi-dimensional similarity calculation and regularization mechanism. First, in the multiple dimensions of the embedding vector of each sample, the cosine similarity between the query embedding vector Vquery and the sample embedding vector VSiV in the jth dimension is calculated respectively. Then, the similarities of all dimensions are weighted and summed to obtain the comprehensive similarity. On this basis, the regularization term is also introduced to prevent the model from overfitting.

[0064] Specifically, the following method can be used to calculate the most relevant examples. First, the query embedding vector Vquery and the sample embedding vector V Si The cosine similarity of the jth dimension among the m total dimensions is calculated, and the cosine similarities of all the m total dimensions are weighted summed to obtain the comprehensive similarity between the embedding vector and the sample; then, based on the sample S i The embedding vector V Si The regularization term is calculated by the square of the L2 norm and the regularization coefficient; finally, the most relevant example is determined based on the comprehensive similarity and regularization. The specific calculation formula is as follows:

[0065]

[0066] Among them, W i It is sample S i is a weight factor used to reflect the importance or quality of the sample. λ is the regularization term coefficient, which controls the strength of the regularization term to prevent overfitting. γj is used to control the nonlinear activation function of each embedding dimension. σ is used to control the activation function of the final example score. Vquery is the embedding vector converted from the query requirement text provided by the auditor. V Si is the embedding vector of the i-th sample in the sample library, m represents the number of dimensions of the embedding vector, Similarity(V query,j ,V Si,j ) represents the query embedding vector Vquery and the sample embedding vector V Si The cosine similarity in the jth dimension, a j is the weight coefficient of the j-th similarity calculation dimension.

[0067] Through the above optimization, this solution can effectively capture the complex relationship between audit requirements and data and avoid overfitting problems, thereby improving the generalization ability and robustness of audit data acquisition. This example selection strategy based on semantic similarity not only improves the accuracy of the model in processing audit tasks, but also provides an innovative and efficient optimization approach for audit work.

[0068] Step S504, select the audit working paper tool.

[0069] First, after the audit data is acquired, the requirements input by the auditor usually involve a certain category of accounts (such as main business income, other business costs, etc.) and specific types of working papers (such as detailed tables, monthly detailed tables, comparative tables for the same period, etc.). The audit working paper robot parses the input audit requirements through the intention recognition function of the large language model and converts the requirements into a binary form, which consists of tool information and account information. For example, if the auditor requires the generation of a "detailed table of main business income", the robot will parse the requirement as "[Tool: Generate detailed table, account: main business income]" and use this information to find the most suitable audit tool from the tool library.

[0070] Next, the audit working paper tool selection model matches the functional characteristics of the tools in the tool library based on the audit requirements through semantic similarity calculation in the LangChain framework. During the matching process, the robot analyzes the subject categories in the audit requirements, the complexity of the working paper content, and the specific objectives of the audit work to ensure that the selected tools can accurately meet the requirements. By calculating the similarity between the embedding vector of the audit requirements and the embedding vectors of each tool in the tool library, the model can filter out the most relevant tools and automatically call them. This not only avoids the time and error risks of manual tool selection, but also improves the intelligence level of the audit process. The audit tool library is a dynamically updated module. As the business scope of the enterprise expands or the audit rules change, the content of the tool library can be adjusted and expanded in real time. For example, when the enterprise adds a new source of income or a new audit compliance requirement is introduced, the corresponding tool will be added to the tool library to ensure that auditors can always use the latest audit tools. This dynamic update mechanism ensures the flexibility and sustainability of audit work.

[0071] Finally, through the optimization selection process of the tool selection model, auditors’ needs can be responded to quickly and accurately. The selected tools can not only meet the current audit needs, but also provide technical support for the subsequent production of audit working papers, ensuring the efficiency and accuracy of the audit process.

[0072] Step S506: Based on the audit data, use the audit working paper tool to prepare the audit working paper.

[0073] To ensure data consistency and operability, the audit working paper robot uses the generation function of a large language model to convert various types of audit data into a unified JSON format. The JSON format not only facilitates structured storage of data, but also makes subsequent data processing and operations more concise and orderly. After the data is standardized, the robot processes it according to the needs of auditors and audit rules. For example, for financial statement items, the robot can perform operations including aggregation, comparison, and calculation of differences, ensuring the integrity and accuracy of the data.

[0074] Once the audit data is standardized, the audit working paper robot will further calculate and clean the JSON format data through the selected audit tool. Different audit working paper types correspond to different data processing logic. For example:

[0075] Detailed statement: When the audit task involves generating a detailed statement, the robot will automatically call the working paper tool to process the relevant financial data according to the prescribed audit standards. For profit and loss items, the tool will compare the current period data with the same period of the previous year and automatically calculate the difference to help auditors quickly identify anomalies or changing trends.

[0076] Comparison table for the same period: The robot can generate a detailed comparison table for the same period, compare data from multiple periods, analyze the extent of changes and trends, and automatically summarize the reasons for the changes for reference by auditors.

[0077] Revenue cutoff test: When conducting a revenue cutoff test, the audit tool will automatically filter out revenue data within a specific period and perform corresponding compliance tests to ensure that the data meets the company's financial reporting requirements.

[0078] During the data processing process, the audit working paper robot can not only perform routine aggregation and classification, but also automatically detect anomalies or potential audit clues in the data. For identified anomalies, the robot will generate prompt information to remind auditors to conduct further verification and review, thereby improving the accuracy of the audit work.

[0079] After data processing is completed, the audit working papers will be presented in a clear table format to ensure that auditors can review the data intuitively. At the same time, the generated audit working papers will remain consistent with the original data to ensure the traceability and reliability of the data.

[0080] This step ensures that the generation of audit working papers complies with industry standards by closely combining the acquisition, processing and application of audit data, thereby improving the automation and accuracy of audit work.

[0081] Figure 6 FIG. 1 shows a schematic diagram of the structure of an electronic device suitable for implementing the embodiment of the present disclosure. It should be noted that: Figure 6 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0082] like Figure 6 As shown, the electronic device includes a central processing unit (CPU) 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage part 1008 into a random access memory (RAM) 1003. In the RAM 1003, various programs and data required for system operation are also stored. The CPU 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0083] The following components are connected to the I / O interface 1005: an input section 1006 including a keyboard, a mouse, etc.; an output section 1007 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, a modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1010 as needed, so that a computer program read therefrom is installed into the storage section 1008 as needed.

[0084] In particular, according to an embodiment of the present disclosure, the process described below with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1009, and / or installed from the removable medium 1011. When the computer program is executed by the central processing unit (CPU) 1001, various functions defined in the method and apparatus of the present application are executed. In some embodiments, the electronic device may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0085] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for producing audit working papers based on AI big model, characterized in that: include: The audit working paper tool selection model of the audit working paper robot identifies the corresponding working paper production intention based on the input audit working paper requirements, and screens and extracts the corresponding audit working paper tools from the audit tool library according to the working paper production intention; The audit working papers preparation model of the audit working papers robot prepares audit working papers based on the acquired audit data of relevant subjects and using the audit working papers tool.

2. The method according to claim 1, characterized in that The audit working paper tool selection model of the audit working paper robot identifies the corresponding intention based on the input audit working paper requirements, and screens and extracts the corresponding audit working paper tools from the audit tool library according to the corresponding intention, including: The audit working paper tool selection model of the audit working paper robot identifies the working paper production intention based on the input audit working paper requirements to obtain a tuple including tool information and subject information; Based on the tool information and the subject information in the tuple, the corresponding audit working paper tools are screened and extracted from the audit tool library.

3. The method according to claim 2, characterized in that Before screening and extracting the corresponding audit working paper tool from the audit tool library according to the corresponding intention, the method further includes: constructing the audit working paper tool based on the subject category and the content of the audit working paper.

4. The method according to claim 3, characterized in that The subject categories include main business income, main business cost, other income subjects, and other expense costs; The contents of the audit working papers include detailed statements, monthly detailed statements, comparative statements for the same period, income cutoff tests, and voucher tests.

5. The method according to claim 2, characterized in that: The audit working paper preparation model includes different types of audit working paper preparation functions under each subject, wherein the input parameter of the audit working paper preparation function is the subject information in the binary group.

6. The method according to claim 2, characterized in that Use the audit working papers tool to prepare audit working papers, including: Based on the subject information in the binary group, acquiring audit data through an audit data acquisition model; Using the generation function of the large language model, the data format of the audit data is set to JSON format; Based on the audit data in the JSON format, the audit data in the JSON format is processed and calculated using the audit working paper formulation function according to the output requirements of the audit working paper to obtain the processed audit data; The processed audit data is presented in a table to obtain the audit working papers.

7. A device for producing audit working papers based on AI big model, characterized in that: include: An audit working paper tool selection model is configured to identify corresponding intentions based on input audit working paper requirements, and screen and extract corresponding audit working paper tools from an audit tool library according to the corresponding intentions; The audit working papers preparation model is configured to prepare audit working papers using the audit working papers tool based on the acquired audit data of relevant subjects.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 6.

9. A computer device, characterized in that: include: Memory and processor, The memory stores a computer program; The processor is used to execute the computer program stored in the memory, and when the computer program is run, the processor is enabled to execute the method according to 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, the steps of the method according to any one of claims 1 to 6 are implemented.

Images