Financial auditing method and device, electronic equipment and storage medium

Abstract

The invention discloses a financial auditing method and device, electronic equipment and a storage medium. The method comprises the steps of determining a task type corresponding to a financial audit question in response to the financial audit question input by a user; under the condition that the task type corresponding to the financial audit problem is a structured query language query task, determining a structured query language query statement corresponding to the financial audit problem, and executing the structured query language query statement to obtain a structured query language query result; and when the task type corresponding to the financial audit problem is the text understanding task, retrieving the financial audit problem in the financial audit document to obtain an associated text corresponding to the financial audit problem, and generating a document query result based on the associated text corresponding to the financial audit problem. According to the technical scheme, the input financial audit problem is subjected to structured query language query or document query in a classified manner, so that the query result of the financial audit problem is quickly obtained, and the financial audit efficiency is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a financial auditing method, device, electronic equipment and storage medium. Background Art

[0002] In traditional financial auditing, auditing work mainly relies on manual processing. With the increase in business volume, traditional financial auditing methods can no longer meet the demand.

[0003] In the process of implementing the present invention, it is found that there are at least the following technical problems in the prior art: the existing financial auditing method has the problem of low auditing efficiency. Summary of the invention

[0004] The present invention provides a financial audit method, device, electronic equipment and storage medium to improve the efficiency of financial audit.

[0005] According to one aspect of the present invention, there is provided a financial audit method, comprising:

[0006] In response to a financial audit question input by a user, determining a task type corresponding to the financial audit question;

[0007] In the case where the task type corresponding to the financial audit question is a structured query language query task, determining a structured query language query statement corresponding to the financial audit question, executing the structured query language query statement corresponding to the financial audit question, and obtaining a structured query language query result corresponding to the financial audit question;

[0008] When the task type corresponding to the financial audit question is a text understanding task, the financial audit question is searched in the financial audit document to obtain the associated text corresponding to the financial audit question, and a document query result corresponding to the financial audit question is generated based on the associated text corresponding to the financial audit question.

[0009] According to another aspect of the present invention, there is provided a financial auditing device, comprising:

[0010] A financial audit question type determination module, used to determine the task type corresponding to the financial audit question in response to the financial audit question input by the user;

[0011] A structured query language query result determination module is used to determine the structured query language query statement corresponding to the financial audit question, execute the structured query language query statement corresponding to the financial audit question, and obtain the structured query language query result corresponding to the financial audit question when the task type corresponding to the financial audit question is a structured query language query task;

[0012] The document query result determination module is used to search the financial audit question in the financial audit document when the task type corresponding to the financial audit question is a text understanding task, obtain the associated text corresponding to the financial audit question, and generate a document query result corresponding to the financial audit question based on the associated text corresponding to the financial audit question.

[0013] According to another aspect of the present invention, there is provided an electronic device, the electronic device comprising:

[0014] at least one processor;

[0015] and a memory communicatively coupled to the at least one processor;

[0016] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the financial audit method described in any embodiment of the present invention.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the financial audit method described in any embodiment of the present invention when executed.

[0018] The technical solution of the embodiment of the present invention determines the task type corresponding to the financial audit question in response to the financial audit question input by the user. When the task type corresponding to the financial audit question is a structured query language (SQL) query task, the structured query language query statement corresponding to the financial audit question is determined, and the structured query language query statement corresponding to the financial audit question is executed to obtain the structured query language query result corresponding to the financial audit question; when the task type corresponding to the financial audit question is a text understanding task, the financial audit question is searched in the financial audit document to obtain the associated text corresponding to the financial audit question, and the document query result corresponding to the financial audit question is generated based on the associated text corresponding to the financial audit question. The above technical solution realizes the rapid acquisition of the query results of the financial audit question by classifying the financial audit question input by the user and performing SQL query or document query, thereby effectively improving the efficiency of financial audit.

[0019] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 is a flow chart of a financial audit method provided according to Embodiment 1 of the present invention;

[0022] Figure 2 is a flow chart of a financial audit method provided according to Embodiment 2 of the present invention;

[0023] Figure 3 is a flow chart of a financial audit method provided according to Embodiment 3 of the present invention;

[0024] Figure 4 is a flow chart of a financial audit method provided according to a fourth embodiment of the present invention;

[0025] Figure 5 is a structural diagram of a financial auditing device provided according to Embodiment 5 of the present invention;

[0026] Figure 6 It is a schematic diagram of the structure of an electronic device for implementing the financial auditing method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] 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.

[0028] 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 interchangeable 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 of their variations 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. The acquisition, storage, use, processing, etc. of data in the technical solution of this application comply with the relevant provisions of national laws and regulations.

[0029] Embodiment 1

[0030] Figure 1 This is a flowchart of a financial audit method provided in the first embodiment of the present invention. This embodiment is applicable to the case of financial auditing through a large language model. The method can be executed by a financial audit device. The financial audit device can be implemented in the form of hardware and / or software. The financial audit device can be configured in electronic devices such as terminals and / or servers. Figure 1 As shown, the method includes:

[0031] S110 . In response to a financial audit question input by a user, determine a task type corresponding to the financial audit question.

[0032] In the embodiment of the present invention, financial audit issues refer to issues related to financial audit.

[0033] Exemplarily, users can input questions related to financial auditing through the question-and-answer interactive interface. For example, financial auditing questions may be "What are the top three projects in terms of cash flow in 2021?" or "In the photovoltaic power generation forecasting project, how to determine the allocation ratio of project funds?" etc.

[0034] In an embodiment of the present invention, the task type is a structured query language (SQL) query task or a text understanding task. Among them, the SQL query task refers to the type of question that converts a financial audit question into an SQL query statement and then queries it in an SQL database through the SQL query statement. The text understanding task refers to the type of question that retrieves a financial audit question from a financial audit document. It should be noted that the answer to a financial audit question may exist in an SQL database or in a financial audit document. By determining the task type corresponding to the financial audit question, it can be clearly determined whether to search for the answer to the question in the SQL database or in the financial audit document, thereby improving the efficiency of financial audit queries.

[0035] S120. When the task type corresponding to the financial audit question is a structured query language query task, determine a structured query language query statement corresponding to the financial audit question, execute the structured query language query statement corresponding to the financial audit question, and obtain a structured query language query result corresponding to the financial audit question.

[0036] The structured query language query result refers to the query result of the SQL query statement in the SQL database.

[0037] Specifically, when the task type corresponding to the financial audit problem is an SQL query task, the financial audit problem is converted into an SQL statement to obtain the SQL query statement corresponding to the financial audit problem, and the SQL query statement corresponding to the financial audit problem is executed to make the query structure of the SQL query statement in the SQL database to obtain the SQL query result corresponding to the financial audit problem.

[0038] S130. When the task type corresponding to the financial audit question is a text understanding task, the financial audit question is searched in the financial audit document to obtain the associated text corresponding to the financial audit question, and a document query result corresponding to the financial audit question is generated based on the associated text corresponding to the financial audit question.

[0039] Among them, financial audit documents refer to files that need to be found and analyzed during the financial audit process. For example, financial audit documents can be photovoltaic power generation forecast project documents or project contracts, etc. Related texts refer to words, sentences or paragraphs related to financial audit questions retrieved from financial audit documents, and are not specifically limited here. Document query results refer to the answers to financial audit questions retrieved and analyzed in financial audit documents.

[0040] For example, a financial audit question may be "What is the depreciation method for the company's fixed assets?", and the associated text retrieved in the financial audit document may be "In 2019, the company used the straight-line method to depreciate fixed assets, with an estimated residual value of 5% and a depreciation period of 5 years", and the document query result may be "The company uses the straight-line method to depreciate fixed assets."

[0041] In some optional embodiments, the financial audit document may be a pdf file or a word file, etc. Text extraction and data cleaning are performed on the pdf file or the word file to obtain a Markdown format table, so that financial audit issues can be retrieved in the Markdown format table to improve the retrieval speed and retrieval quality.

[0042] Optionally, a document query result corresponding to the financial audit question is generated based on the associated text corresponding to the financial audit question, including: constructing an audit document retrieval prompt based on the financial audit question and the associated text corresponding to the financial audit question; inputting the audit document retrieval prompt into a large language model to obtain a document query result corresponding to the financial audit question.

[0043] The audit document retrieval prompt word refers to reference data that provides guidance for the large language model to output answers. The large language model can be a Qwen large model or other large models, which are not specifically limited here.

[0044] It should be noted that audit document retrieval prompts can guide the large language model to make full use of related texts to provide efficient and accurate answers, thereby improving the accuracy of document query results.

[0045] The technical solution of the embodiment of the present invention determines the task type corresponding to the financial audit question in response to the financial audit question input by the user. When the task type corresponding to the financial audit question is a structured query language query task, the structured query language query statement corresponding to the financial audit question is determined, and the structured query language query statement corresponding to the financial audit question is executed to obtain the structured query language query result corresponding to the financial audit question; when the task type corresponding to the financial audit question is a text understanding task, the financial audit question is searched in the financial audit document to obtain the associated text corresponding to the financial audit question, and the document query result corresponding to the financial audit question is generated based on the associated text corresponding to the financial audit question. The above technical solution realizes the rapid acquisition of the query results of the financial audit question by classifying the financial audit question input by the user and performing SQL query or document query, thereby effectively improving the efficiency of financial audit.

[0046] Embodiment 2

[0047] Figure 2 A flowchart of a financial audit method provided for Embodiment 2 of the present invention, the method of this embodiment can be combined with the various optional schemes in the financial audit method provided in the above embodiments. The financial audit method provided in this embodiment has been further optimized. Optionally, determining the task type corresponding to the financial audit question includes: constructing a query target determination prompt word based on the historical question and answer information and the financial audit question; inputting the query target determination prompt word into a large language model to obtain a query target corresponding to the financial audit question; classifying the query target corresponding to the financial audit question to obtain the task type corresponding to the financial audit question.

[0048] like Figure 2 As shown, the method includes:

[0049] S210 . In response to a financial audit question input by a user, construct a query target and determine prompt words based on historical question and answer information and the financial audit question.

[0050] Here, historical question and answer information refers to one or more financial audit question and answer examples.

[0051] Exemplarily, historical question and answer information may include: Question 1: "In the photovoltaic power generation forecast project, how to determine the allocation ratio of project funds?", Answer 1: "Photovoltaic power generation forecast project file". Question 2: "Use photovoltaic power generation forecast project funds to reimburse the purchase of a hard disk, the reimbursement amount is 1900", Answer 2: "SQL database".

[0052] In the embodiment of the present invention, the query target determination prompt word refers to reference data that provides guidance for the large language model to determine the query target. The query target can be a project file or an SQL database, etc.

[0053] For example, the In Context Learning (ICL) method can be used to construct query target determination prompt words based on historical question and answer information and financial audit questions to help the large language model understand which query target the financial audit questions raised by users should be queried from.

[0054] S220: Input the query target determination prompt word into the large language model to obtain the query target corresponding to the financial audit question.

[0055] Exemplarily, the formula for determining the query target corresponding to the financial audit question is:

[0056] C = M (ICL (question));

[0057] Among them, question represents a financial audit question, ICL(·) represents the query target determination prompt word obtained through context learning, M(·) represents the query target obtained through large language model prediction, and C represents the query target, for example, C can be a SQL database or project file name, etc.

[0058] S230: Classify the query targets corresponding to the financial audit questions to obtain the task types corresponding to the financial audit questions.

[0059] For example, a decision tree (DT) or other classification algorithms may be used to classify the query target. The specific mathematical expression is as follows:

[0060] A=DT(C);

[0061] Among them, C represents the query target, DT(·) represents classification through decision tree, and A represents the task type corresponding to the financial audit problem.

[0062] S240. When the task type corresponding to the financial audit question is a structured query language query task, determine a structured query language query statement corresponding to the financial audit question, execute the structured query language query statement corresponding to the financial audit question, and obtain a structured query language query result corresponding to the financial audit question.

[0063] S250. When the task type corresponding to the financial audit question is a text understanding task, the financial audit question is searched in the financial audit document to obtain the associated text corresponding to the financial audit question, and a document query result corresponding to the financial audit question is generated based on the associated text corresponding to the financial audit question.

[0064] The technical solution of the embodiment of the present invention constructs query target determination prompt words based on historical question and answer information and financial audit questions, and then inputs the query target determination prompt words into the large language model to obtain the query target corresponding to the financial audit question, and then classifies the query target corresponding to the financial audit question to obtain the task type corresponding to the financial audit question. The above technical solution determines the task type corresponding to the financial audit question by constructing prompt words, predicting query targets, and classifying operations, so that it can be clear whether to go to the SQL database or the financial audit document to find the answer to the question, thereby improving the efficiency of financial audit queries.

[0065] Embodiment 3

[0066] Figure 3A flowchart of a financial audit method provided in the third embodiment of the present invention, the method of this embodiment can be combined with each optional scheme in the financial audit method provided in the above embodiment. The financial audit method provided in this embodiment is further optimized. Optionally, determining the structured query language query statement corresponding to the financial audit problem includes: obtaining multiple financial audit problem samples and the structured query language query statement corresponding to each financial audit problem sample; determining the Jaccard similarity between each financial audit problem sample and the financial audit problem; sorting the multiple financial audit problem samples based on the Jaccard similarity between each financial audit problem sample and the financial audit problem; based on the sample sorting result, screening out financial audit problem samples similar to the financial audit problem and structured query language query statements corresponding to the financial audit problem samples similar to the financial audit problem; constructing structured query language conversion prompt words based on the financial audit problem, the financial audit problem samples similar to the financial audit problem and the structured query language query statements corresponding to the financial audit problem samples similar to the financial audit problem; inputting the structured query language conversion prompt words into the large language model to obtain the structured query language query statement corresponding to the financial audit problem.

[0067] like Figure 3 As shown, the method includes:

[0068] S310: In response to a financial audit question input by a user, determine a task type corresponding to the financial audit question.

[0069] S320: When the task type corresponding to the financial audit question is a structured query language query task, obtain a plurality of financial audit question examples and a structured query language query statement corresponding to each financial audit question example.

[0070] The financial audit problem examples are pre-constructed financial audit problem examples. The SQL query statements and the financial audit problem examples exist in pairs, that is, each financial audit problem example has its corresponding SQL query statement.

[0071] For example, multiple financial audit question examples can be obtained from the QuerytoSQL annotation library. The financial audit question example can be: Which project of the company has the largest turnover in XX Bank in 2024? The SQL query statement corresponding to the financial audit question example is:

[0072] SELECT project name, SUM (transaction amount) AS total amount;

[0073] FROM flow table;

[0074] WHERE Bank Name = 'XX Bank';

[0075] AND YEAR(transaction date) = 2024;

[0076] GROUP BY item name;

[0077] ORDER BY total amount DESC;

[0078] LIMIT 1.

[0079] S330, determining the Jaccard similarity between each financial audit question sample and the financial audit question.

[0080] Among them, the Jaccard similarity calculation formula is:

[0081]

[0082] Among them, J i represents the Jaccard similarity between the financial audit problem and the i-th financial audit problem example, q represents the financial audit problem, Q i represents the i-th financial audit problem example.

[0083] S340. Sort the multiple financial audit question samples based on the Jaccard similarity between each financial audit question sample and the financial audit question; and based on the sample sorting results, screen out financial audit question samples similar to the financial audit question and structured query language query statements corresponding to the financial audit question samples similar to the financial audit question.

[0084] The financial audit problem samples similar to the financial audit problem refer to the top k financial audit problem samples ranked by Jaccard similarity to the financial audit problem, and k is a custom parameter value, which is not specifically limited here.

[0085] Exemplarily, based on the Jaccard similarity between each financial audit problem sample and the financial audit problem, multiple financial audit problem samples are sorted from large to small, the first k financial audit problem samples in the sample sorting results are taken as financial audit problem samples similar to the financial audit problem, and the SQL query statements corresponding to the financial audit problem samples similar to the financial audit problem are obtained.

[0086] S350: constructing structured query language conversion prompt words based on the financial audit problem, financial audit problem examples similar to the financial audit problem, and structured query language query statements corresponding to the financial audit problem examples similar to the financial audit problem.

[0087] The structured query language conversion prompt refers to reference data that provides guidance for the large language model to generate SQL statements.

[0088] For example, a financial audit question may be: "Query the top three projects with the highest turnover in XX Bank in 2024?". Examples of financial audit questions similar to financial audit questions may be: "Find the top two projects with the highest transaction amount in XX Bank in 2019", "Which project has the highest total turnover in XX Bank in 2021?", "Query the top five projects with the largest turnover in XYZ Bank in 2021". The corresponding SQL query statement for "Find the top two projects with the highest transaction amount in XX Bank in 2019" is: "SELECT project name, SUM (transaction amount) AS total amount FROM flow WHERE bank name = 'XX Bank' AND YEAR (transaction date) = 2019 GROUP BY project name ORDER BY total amount DESC LIMIT 2"; "Which project has the highest total annual flow in XX Bank in 2021?" The corresponding SQL query statement is: "SELECT project name, SUM (transaction amount) AS total amount FROM flow table WHERE bank name = 'XX Bank' AND YEAR (transaction date) = 2021 GROUP BY project name ORDER BY total amount DESC LIMIT 1"; "Query the top five projects with the largest flow in XYZ Bank in 2021" The corresponding SQL query statement is: "SELECT project name, SUM (transaction amount) AS total amount FROM flow WHERE bank name = 'XYZ Bank' AND YEAR (transaction date) = 2021 GROUP BY project name ORDER BY total amount DESC LIMIT 1 BY total amount DESC LIMIT 5".

[0089] Furthermore, the structured query language conversion prompt words may be:

[0090] “Here are the financial audit problems I want to solve:

[0091] "Query the top three items with the most transactions in XX Bank in 2024?"

[0092] Here are similar questions and their corresponding SQL queries for reference:

[0093] Similar question 1: "Find the top two projects with the highest transaction amount in XX Bank in 2019";

[0094] The corresponding SQL query statement is: "SELECT project name, SUM (transaction amount) AS total amount FROM transaction data WHERE bank name = 'XX Bank' AND YEAR (transaction date) = 2019 GROUP BY project name ORDER BY total amount DESC LIMIT 2";

[0095] Similar question 2: "Which project has the highest total annual turnover in XX Bank in 2021?";

[0096] The corresponding SQL query statement is: "SELECT project name, SUM (transaction amount) AS total amount FROM flow table WHERE bank name = 'XX Bank' AND YEAR (transaction date) = 2021 GROUP BY project name ORDER BY total amount DESC LIMIT 1";

[0097] Similar question 3: "Query the top five items with the largest turnover in XYZ Bank in 2021";

[0098] The corresponding SQL query statement is: "SELECT project name, SUM (transaction amount) AS total amount FROM transaction data WHERE bank name = 'XYZ Bank' AND YEAR (transaction date) = 2021 GROUP BY project name ORDER BY total amount DESC LIMIT 5";

[0099] Based on the above references, please generate appropriate SQL query statements for financial audit problems:".

[0100] S360: Input the structured query language conversion prompt word into the large language model to obtain a structured query language query statement corresponding to the financial audit question, execute the structured query language query statement corresponding to the financial audit question, and obtain a structured query language query result corresponding to the financial audit question.

[0101] Exemplarily, according to the SQL conversion prompt word, the large language model can output the SQL query statement corresponding to the financial audit question, so as to use the SQL query statement corresponding to the financial audit question to query in the SQL data and obtain the SQL query result.

[0102] In an embodiment of the present invention, when the task type corresponding to the financial audit question is a text understanding task, the financial audit question is searched in the financial audit document to obtain the associated text corresponding to the financial audit question, and a document query result corresponding to the financial audit question is generated based on the associated text corresponding to the financial audit question.

[0103] In some optional embodiments, executing a structured query language query statement corresponding to the financial audit question to obtain a structured query language query result corresponding to the financial audit question includes: executing a structured query language query statement corresponding to the financial audit question to obtain multiple candidate query results; if the same candidate query result exists among the multiple candidate query results, determining the same candidate query result as the structured query language query result corresponding to the financial audit question; if the same candidate query result does not exist among the multiple candidate query results, selecting from the multiple candidate query results based on Jaccard similarity to obtain the structured query language query result corresponding to the financial audit question.

[0104] Among them, by using the SQL query statement corresponding to the financial audit question to query in the SQL data, multiple candidate query results can be obtained, and the candidate query results refer to the SQL query results to be confirmed.

[0105] It should be noted that by judging whether there is an identical candidate query result among multiple candidate query results, the candidate query results are screened, thereby determining an accurate and reasonable structured query language query result.

[0106] For example, the final SQL query result can be determined according to the following formula:

[0107]

[0108] Among them, Answer represents the structured query language query result corresponding to the financial audit question. 1 Indicates the same candidate query result, that is, the candidate query result with the most occurrences. 2 is the candidate query result of the SQL query statement corresponding to the sample with the maximum Jac in the SQL database, and Jac represents the Jaccard similarity.

[0109] The technical solution of the embodiment of the present invention obtains multiple financial audit problem samples and the structured query language query statement corresponding to each financial audit problem sample, and then determines the Jaccard similarity between each financial audit problem sample and the financial audit problem, and then sorts the multiple financial audit problem samples based on the Jaccard similarity between each financial audit problem sample and the financial audit problem; based on the sample sorting results, screens out financial audit problem samples similar to the financial audit problem and the structured query language query statements corresponding to the financial audit problem samples similar to the financial audit problem, and then constructs the structured query language conversion prompt words based on the financial audit problem, the financial audit problem samples similar to the financial audit problem, and the structured query language query statements corresponding to the financial audit problem samples similar to the financial audit problem; inputs the structured query language conversion prompt words into the large language model to obtain the structured query language query statement corresponding to the financial audit problem. The above technical solution realizes the accurate conversion of SQL query statements through Jaccard similarity sorting and screening, the construction of structured query language conversion prompt words, and the prediction of structured query language query statements, and effectively improves the accuracy of financial audit data query.

[0110] Embodiment 4

[0111] Figure 4 This is a flowchart of a financial audit method provided by the fourth embodiment of the present invention. The method of this embodiment can be combined with various optional solutions of the financial audit method provided in the above embodiments. The financial audit method provided by this embodiment is further optimized. Optionally, the financial audit question is retrieved in the financial audit document to obtain the associated text corresponding to the financial audit question, including: performing word segmentation on the financial audit question to obtain the word segmentation result corresponding to the financial audit question; embedding coding the word segmentation result corresponding to the financial audit question to obtain the word segmentation semantic vector corresponding to the financial audit question; performing word segmentation on multiple text blocks in the financial audit document to obtain the word segmentation result corresponding to each text block; embedding coding the word segmentation result corresponding to each text block to obtain the word segmentation semantic vector corresponding to each text block; determining the cosine similarity between the word segmentation semantic vector corresponding to each text block and the word segmentation semantic vector corresponding to the financial audit question; sorting the multiple text blocks in the financial audit document based on the cosine similarity between the word segmentation semantic vector corresponding to each text block and the word segmentation semantic vector corresponding to the financial audit question; based on the document block sorting result, screening to obtain multiple text blocks similar to the financial audit question; context sorting the multiple text blocks similar to the financial audit question to obtain the associated text corresponding to the financial audit question, wherein the associated text is text with context association.

[0112] like Figure 4As shown, the method includes:

[0113] S410: In response to a financial audit question input by a user, determine a task type corresponding to the financial audit question.

[0114] In an embodiment of the present invention, when the task type corresponding to the financial audit question is a structured query language query task, a structured query language query statement corresponding to the financial audit question is determined, the structured query language query statement corresponding to the financial audit question is executed, and a structured query language query result corresponding to the financial audit question is obtained.

[0115] S420. When the task type corresponding to the financial audit question is a text understanding task, the financial audit question is segmented to obtain a segmentation result corresponding to the financial audit question; and the segmentation result corresponding to the financial audit question is embeddedly encoded to obtain a segmentation semantic vector corresponding to the financial audit question.

[0116] The word segmentation result corresponding to the financial audit question refers to the basic processing unit obtained by decomposing the financial audit question. The word segmentation semantic vector corresponding to the financial audit question is a vector representation of the semantic meaning of the word segmentation result corresponding to the financial audit question.

[0117] For example, the financial audit questions can be tokenized through the tokenization layer to obtain the tokenization results corresponding to the financial audit questions so that the large language model can understand and process them. Then, the tokenization results corresponding to the financial audit questions can be converted into token semantic vectors corresponding to the financial audit questions through the embedding coding layer so that semantic operations and comparisons can be performed in the vector space.

[0118] In some optional embodiments, before word segmentation is performed on the financial audit questions, named entity recognition (NER) may be performed on the financial audit questions to obtain target identification, and then the target identification is converted into a Json format to obtain preprocessed financial audit questions, and then word segmentation and embedding coding and other operations are performed on the preprocessed financial audit questions.

[0119] S430, performing word segmentation on multiple text blocks in the financial audit document to obtain a word segmentation result corresponding to each text block; performing embedding coding on the word segmentation result corresponding to each text block to obtain a word segmentation semantic vector corresponding to each text block.

[0120] Among them, a financial audit document may include multiple text blocks. In other words, each text block is a part of the financial audit document. The word segmentation result corresponding to the text block refers to the basic processing unit obtained by decomposing the text block. The word segmentation semantic vector corresponding to the text block is a vector representation of the semantic meaning of the word segmentation result corresponding to the text block.

[0121] For example, the text blocks can be segmented through the tokenization layer to obtain the segmentation results corresponding to the text blocks so that the large language model can understand and process them, and then the segmentation results corresponding to the text blocks can be converted into the segmentation semantic vectors corresponding to the text blocks through the embedding coding layer so as to perform semantic operations and comparisons in the vector space.

[0122] S440. Determine the cosine similarity between the word segmentation semantic vector corresponding to each text block and the word segmentation semantic vector corresponding to the financial audit question; sort the multiple text blocks in the financial audit document based on the cosine similarity between the word segmentation semantic vector corresponding to each text block and the word segmentation semantic vector corresponding to the financial audit question; and screen and obtain multiple text blocks similar to the financial audit question based on the document block sorting results.

[0123] The text blocks similar to the financial audit issues refer to the top m text blocks ranked by cosine similarity with the financial audit issues, and m is a custom parameter value, which is not specifically limited here.

[0124] Exemplarily, the calculation formula of cosine similarity is as follows:

[0125]

[0126] Among them, q′ represents the preprocessed financial audit problem, D i represents the i-th text block in the financial audit document, C(q′,D i ) represents cosine similarity. Further, based on the cosine similarity between the word segmentation semantic vector corresponding to each text segmentation block and the word segmentation semantic vector corresponding to the financial audit question, multiple text segments in the financial audit document are sorted from large to small, and the first m text segments in the document segmentation sorting result are taken as text segments similar to the financial audit question.

[0127] S450. Contextually sort the multiple text blocks similar to the financial audit question to obtain associated text corresponding to the financial audit question, wherein the associated text is text with contextual association, and generate a document query result corresponding to the financial audit question based on the associated text corresponding to the financial audit question.

[0128] Since the multiple text blocks similar to financial audit issues that are screened out are static vectors and do not change with the context, the document query results generated by them are inaccurate. In order to consider the context information between texts, the embodiment of the present invention contextually sorts the multiple text blocks similar to financial audit issues, obtains texts with context associations, and then generates document query results through the texts with context associations, thereby improving the accuracy of the document query results.

[0129] Optionally, contextual sorting is performed on multiple text blocks similar to the financial audit question to obtain associated text corresponding to the financial audit question, including: determining the Jaccard similarity between multiple text blocks similar to the financial audit question and the financial audit question; sorting multiple text blocks similar to the financial audit question based on the Jaccard similarity between the multiple text blocks similar to the financial audit question and the financial audit question; and using the re-sorted multiple text blocks similar to the financial audit question as associated text corresponding to the financial audit question.

[0130] Exemplarily, the calculation formula for determining the associated text is:

[0131]

[0132] Among them, q′ represents the preprocessed financial audit problem, d i represents the i-th text block similar to the financial audit question, Represents the Jaccard similarity between the ith text block similar to the financial audit question and the preprocessed financial audit question. Indicates based on The size of d i Sort from largest to smallest. d′ represents the associated text.

[0133] Furthermore, the formula for determining the document query result is:

[0134] Answer = LLM(d′);

[0135] Among them, LLM(·) represents the large language model, and Answer represents the document query result.

[0136] The technical solution of the embodiment of the present invention is to segment financial audit issues to obtain segmentation results corresponding to the financial audit issues, embed code the segmentation results corresponding to the financial audit issues to obtain segmentation semantic vectors corresponding to the financial audit issues, segment multiple text blocks in the financial audit document to obtain segmentation results corresponding to each text block; embed code the segmentation results corresponding to each text block to obtain segmentation semantic vectors corresponding to each text block; determine the cosine similarity between the segmentation semantic vector corresponding to each text block and the segmentation semantic vector corresponding to the financial audit issue; sort multiple text blocks in the financial audit document based on the cosine similarity between the segmentation semantic vector corresponding to each text block and the segmentation semantic vector corresponding to the financial audit issue; based on the document segmentation results, screen and obtain multiple text blocks similar to the financial audit issue; context sort multiple text blocks similar to the financial audit issue to obtain associated texts corresponding to the financial audit issue, wherein the associated text is text with context association. The above technical solution realizes text with context association through word segmentation, embedded coding, cosine similarity sorting and screening, and context sorting, effectively improving the reliability and effectiveness of the associated text.

[0137] Embodiment 5

[0138] Figure 5 This is a schematic diagram of the structure of a financial audit device provided by Embodiment 5 of the present invention. Figure 5 As shown, the device comprises:

[0139] A financial audit question type determination module 510 is used to determine the task type corresponding to the financial audit question in response to the financial audit question input by the user;

[0140] The structured query language query result determination module 520 is used to determine the structured query language query statement corresponding to the financial audit question when the task type corresponding to the financial audit question is a structured query language query task, execute the structured query language query statement corresponding to the financial audit question, and obtain the structured query language query result corresponding to the financial audit question.

[0141] The document query result determination module 530 is used to search the financial audit question in the financial audit document when the task type corresponding to the financial audit question is a text understanding task, obtain the associated text corresponding to the financial audit question, and generate a document query result corresponding to the financial audit question based on the associated text corresponding to the financial audit question.

[0142] The technical solution of the embodiment of the present invention determines the task type corresponding to the financial audit question in response to the financial audit question input by the user. When the task type corresponding to the financial audit question is a structured query language query task, the structured query language query statement corresponding to the financial audit question is determined, and the structured query language query statement corresponding to the financial audit question is executed to obtain the structured query language query result corresponding to the financial audit question; when the task type corresponding to the financial audit question is a text understanding task, the financial audit question is searched in the financial audit document to obtain the associated text corresponding to the financial audit question, and the document query result corresponding to the financial audit question is generated based on the associated text corresponding to the financial audit question. The above technical solution realizes the rapid acquisition of the query results of the financial audit question by classifying the financial audit question input by the user and performing SQL query or document query, thereby effectively improving the efficiency of financial audit.

[0143] In some optional implementations, the financial audit question type determination module 510 may be specifically used to:

[0144] Constructing a query target based on the historical question and answer information and the financial audit question to determine a prompt word;

[0145] Inputting the query target determination prompt word into the large language model to obtain the query target corresponding to the financial audit question;

[0146] The query targets corresponding to the financial auditing problems are classified to obtain the task types corresponding to the financial auditing problems.

[0147] In some optional implementations, the structured query language query result determination module 520 includes:

[0148] A financial audit question sample acquisition unit, used to acquire multiple financial audit question samples and a structured query language query statement corresponding to each financial audit question sample;

[0149] A Jaccard similarity determination unit, used to determine the Jaccard similarity between each financial audit question sample and the financial audit question;

[0150] A financial audit question sample sorting unit, configured to sort the plurality of financial audit question samples based on the Jaccard similarity between each financial audit question sample and the financial audit question;

[0151] A similar financial audit question sample screening unit is used to screen and obtain financial audit question samples similar to the financial audit question and structured query language query statements corresponding to the financial audit question samples similar to the financial audit question based on the sample sorting results;

[0152] A structured query language conversion prompt word construction unit, used to construct a structured query language conversion prompt word based on the financial audit question, the financial audit question sample similar to the financial audit question, and the structured query language query sentence corresponding to the financial audit question sample similar to the financial audit question;

[0153] The structured query language query statement prediction unit is used to input the structured query language conversion prompt word into the large language model to obtain the structured query language query statement corresponding to the financial audit question.

[0154] In some optional implementations, the structured query language query result determination module 520 may further include:

[0155] A plurality of candidate query result acquisition units, used for executing the structured query language query statement corresponding to the financial audit question to obtain a plurality of candidate query results;

[0156] An identical candidate query result processing unit, configured to determine, if an identical candidate query result exists among the plurality of candidate query results, the identical candidate query result as a structured query language query result corresponding to the financial audit question;

[0157] The different candidate query result processing unit is used to select from the multiple candidate query results based on Jaccard similarity if there is no identical candidate query result among the multiple candidate query results, to obtain a structured query language query result corresponding to the financial audit question.

[0158] In some optional implementations, the document query result determination module 530 includes:

[0159] The financial audit question word segmentation and embedding coding unit is used to segment the financial audit question to obtain the word segmentation result corresponding to the financial audit question; embedding code is performed on the word segmentation result corresponding to the financial audit question to obtain the word segmentation semantic vector corresponding to the financial audit question;

[0160] The financial audit document word segmentation and embedding coding unit is used to segment multiple text blocks in the financial audit document to obtain the word segmentation results corresponding to each text block; embedding coding is performed on the word segmentation results corresponding to each text block to obtain the word segmentation semantic vector corresponding to each text block;

[0161] The text block screening unit is used to determine the cosine similarity between the word segmentation semantic vector corresponding to each text block and the word segmentation semantic vector corresponding to the financial audit question; sort the multiple text blocks in the financial audit document based on the cosine similarity between the word segmentation semantic vector corresponding to each text block and the word segmentation semantic vector corresponding to the financial audit question; based on the document block sorting result, screen and obtain multiple text blocks similar to the financial audit question;

[0162] The text block context sorting unit is used to contextually sort the multiple text blocks similar to the financial audit question to obtain the associated text corresponding to the financial audit question, wherein the associated text is text with contextual association.

[0163] In some optional implementations, the text block context sorting unit may also be used to:

[0164] Determine the Jaccard similarity between the plurality of text blocks similar to the financial audit question and the financial audit question;

[0165] sorting the multiple text blocks similar to the financial audit question based on the Jaccard similarities between the multiple text blocks similar to the financial audit question and the financial audit question;

[0166] The reordered multiple text blocks similar to the financial audit question are used as associated texts corresponding to the financial audit question.

[0167] In some optional implementations, the document query result determination module 530 includes:

[0168] An audit document retrieval prompt word construction unit, used to construct an audit document retrieval prompt word based on the financial audit question and the associated text corresponding to the financial audit question;

[0169] The document query result prediction unit is used to input the audit document retrieval prompt word into the large language model to obtain the document query result corresponding to the financial audit question.

[0170] The financial auditing device provided in the embodiment of the present invention can execute the financial auditing method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0171] Embodiment 6

[0172] Figure 6A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0173] like Figure 6 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. The I / O interface 15 is also connected to the bus 14.

[0174] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0175] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a financial audit method, which includes:

[0176] In response to a financial audit question input by a user, determining a task type corresponding to the financial audit question;

[0177] In the case where the task type corresponding to the financial audit question is a structured query language query task, determining a structured query language query statement corresponding to the financial audit question, executing the structured query language query statement corresponding to the financial audit question, and obtaining a structured query language query result corresponding to the financial audit question;

[0178] When the task type corresponding to the financial audit question is a text understanding task, the financial audit question is searched in the financial audit document to obtain the associated text corresponding to the financial audit question, and a document query result corresponding to the financial audit question is generated based on the associated text corresponding to the financial audit question.

[0179] In some embodiments, the financial audit method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the financial audit method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform the financial audit method in any other appropriate manner (e.g., by means of firmware).

[0180] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0181] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0182] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0183] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0184] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0185] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0186] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0187] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A financial audit method, characterized in that: include: In response to a financial audit question input by a user, determining a task type corresponding to the financial audit question; In the case where the task type corresponding to the financial audit question is a structured query language query task, determining a structured query language query statement corresponding to the financial audit question, executing the structured query language query statement corresponding to the financial audit question, and obtaining a structured query language query result corresponding to the financial audit question; When the task type corresponding to the financial audit question is a text understanding task, the financial audit question is searched in the financial audit document to obtain the associated text corresponding to the financial audit question, and a document query result corresponding to the financial audit question is generated based on the associated text corresponding to the financial audit question.

2. The method according to claim 1, characterized in that The determining of the task type corresponding to the financial audit problem includes: Constructing a query target based on the historical question and answer information and the financial audit question to determine the prompt words; Inputting the query target determination prompt word into the large language model to obtain the query target corresponding to the financial audit question; The query targets corresponding to the financial auditing problems are classified to obtain the task types corresponding to the financial auditing problems.

3. The method according to claim 1, characterized in that The step of determining the structured query language query statement corresponding to the financial audit question includes: Obtain multiple financial audit question samples and a structured query language query statement corresponding to each financial audit question sample; Determine the Jaccard similarity of each sample financial audit question to the financial audit question; sorting the plurality of financial audit question samples based on the Jaccard similarity between each financial audit question sample and the financial audit question; Based on the sample sorting results, financial audit problem samples similar to the financial audit problem and structured query language query statements corresponding to the financial audit problem samples similar to the financial audit problem are screened; Constructing structured query language conversion prompt words based on the financial audit problem, the financial audit problem samples similar to the financial audit problem, and the structured query language query statements corresponding to the financial audit problem samples similar to the financial audit problem; The structured query language conversion prompt word is input into the large language model to obtain a structured query language query statement corresponding to the financial audit question.

4. The method according to claim 1, characterized in that: The executing the structured query language query statement corresponding to the financial audit question to obtain the structured query language query result corresponding to the financial audit question includes: Execute a structured query language query statement corresponding to the financial audit question to obtain multiple candidate query results; If there is an identical candidate query result among the multiple candidate query results, determining the identical candidate query result as the structured query language query result corresponding to the financial audit question; If there is no identical candidate query result among the multiple candidate query results, a query result of the structured query language corresponding to the financial audit question is obtained by selecting from the multiple candidate query results based on the Jaccard similarity.

5. The method according to claim 1, characterized in that The step of searching the financial audit question in the financial audit document to obtain the associated text corresponding to the financial audit question includes: Performing word segmentation on the financial audit question to obtain a word segmentation result corresponding to the financial audit question; performing embedding coding on the word segmentation result corresponding to the financial audit question to obtain a word segmentation semantic vector corresponding to the financial audit question; Segment multiple text blocks in the financial audit document to obtain a segmentation result corresponding to each text block; embed code the segmentation result corresponding to each text block to obtain a segmentation semantic vector corresponding to each text block; Determine the cosine similarity between the word segmentation semantic vector corresponding to each text block and the word segmentation semantic vector corresponding to the financial audit question; Sorting multiple text blocks in the financial audit document based on the cosine similarity between the word segmentation semantic vector corresponding to each text block and the word segmentation semantic vector corresponding to the financial audit question; Based on the document block sorting results, multiple text blocks similar to the financial audit problem are screened and obtained; The multiple text blocks similar to the financial audit question are context-sorted to obtain associated texts corresponding to the financial audit question, wherein the associated texts are texts with contextual associations.

6. The method according to claim 5, characterized in that The context sorting of the multiple text blocks similar to the financial audit question to obtain the associated text corresponding to the financial audit question includes: Determine the Jaccard similarity between the plurality of text blocks similar to the financial audit question and the financial audit question; sorting the multiple text blocks similar to the financial audit question based on the Jaccard similarities between the multiple text blocks similar to the financial audit question and the financial audit question; The reordered multiple text blocks similar to the financial audit question are used as associated texts corresponding to the financial audit question.

7. The method according to claim 1, characterized in that The generating of the document query result corresponding to the financial audit question based on the associated text corresponding to the financial audit question includes: Constructing audit document retrieval prompt words based on the financial audit question and the associated text corresponding to the financial audit question; The audit document retrieval prompt word is input into the large language model to obtain the document query result corresponding to the financial audit question.

8. A financial auditing device, characterized in that: include: A financial audit question type determination module, used to determine the task type corresponding to the financial audit question in response to the financial audit question input by the user; A structured query language query result determination module is used to determine the structured query language query statement corresponding to the financial audit question, execute the structured query language query statement corresponding to the financial audit question, and obtain the structured query language query result corresponding to the financial audit question when the task type corresponding to the financial audit question is a structured query language query task; The document query result determination module is used to search the financial audit question in the financial audit document when the task type corresponding to the financial audit question is a text understanding task, obtain the associated text corresponding to the financial audit question, and generate a document query result corresponding to the financial audit question based on the associated text corresponding to the financial audit question.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the financial audit method described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the financial audit method according to any one of claims 1 to 7 when executed.

Images