Data auditing method and device, equipment, medium and program product

Abstract

The invention provides a data auditing method which can be applied to the field of artificial intelligence and the technical field of financial science and technology. The data auditing method comprises the following steps: in response to input to-be-audited data, determining a to-be-audited object and case-related content corresponding to the to-be-audited object; and calling the legal document related to the to-be-audited object. And determining a prompt text according to the case-related content and the legal document. And inputting the prompt text to a pre-trained large language model, and outputting a prediction result representing the risk level of the to-be-audited object. And according to the prediction result, automatically auditing the to-be-audited data of which the risk level is the specific risk level. The invention further provides a data auditing device and equipment, a storage medium and a program product.

Description

Technical Field

[0001] The present disclosure relates to the fields of artificial intelligence and financial technology, and specifically to intelligent data review, and more specifically to a data review method, device, equipment, medium and program product. Background Art

[0002] In recent years, as the requirements for anti-money laundering work have continued to increase, the system has added about 100,000 new transactions per day to external investigations, covering emergency stop payments, freezing and deductions by law enforcement agencies, and the overall situation is relatively complicated.

[0003] At present, the industry's review of judicial investigation and freezing documents of suspected money laundering customers is mainly conducted manually by classification posts. Classification posts need to identify various situations when handling external joint investigations and accurately push enhanced due diligence. Wrong judgments and audits will leave hidden risks of money laundering or cause additional classification workload for subsequent due diligence posts, business review posts, classification posts, and classification approval posts, affecting overall efficiency. Summary of the invention

[0004] In view of the above problems, embodiments of the present disclosure provide a data audit method, apparatus, device, medium, and program product for improving data audit efficiency.

[0005] According to the first aspect of the present disclosure, a data audit method is provided, comprising: in response to inputting data to be audited, determining an object to be audited and the case-related content corresponding to the object to be audited. Calling legal documents related to the object to be audited. Determining prompt text based on the case-related content and the legal documents. Inputting the prompt text into a pre-trained large language model, and outputting a prediction result representing the risk level of the object to be audited. Based on the prediction result, automatically auditing the data to be audited with a risk level of a specific risk level.

[0006] According to an embodiment of the present disclosure, the prompt text is determined based on the case content and the legal document, including any of the following: comparing the case content and the legal document to determine the case type of the object to be reviewed. Determining specific keywords in the case content, the specific keywords representing the suspension of payment. Comparing the case content and the legal document to determine the consistency of the case account of the object to be reviewed.

[0007] According to an embodiment of the present disclosure, a prompt text is input into a pre-trained large language model, and a prediction result representing the risk level of the object to be reviewed is output, including any of the following: when the case involved is a civil case, the risk level of the data to be reviewed is determined to be a specific risk level. When the content involved has specific keywords, the risk level of the data to be reviewed is determined to be a specific risk level. When the account number involved in the object to be reviewed is inconsistent with the account number in the legal document, the risk level of the data to be reviewed is determined to be a specific risk level.

[0008] According to an embodiment of the present disclosure, the data audit method further includes: determining that the reason for automatically auditing the data to be audited is any one of the following: the case type is a civil case, the case has a specific keyword, and the account involved is inconsistent. The reason for the automatic audit is displayed on the audit page.

[0009] According to an embodiment of the present disclosure, the data to be reviewed includes any one or more of the description documents, approval forms and notices issued by law enforcement agencies related to property freezing.

[0010] According to an embodiment of the present disclosure, the data to be reviewed is multimodal data, which includes documents, pictures, audio and video. In response to inputting the data to be reviewed, determining the object to be reviewed and the content involved in the case corresponding to the object to be reviewed includes at least one of the following: extracting information from the multimodal data using a pre-trained large language model to determine the object to be reviewed and the content involved in the case. Recognizing the image through OCR to determine the object to be reviewed and the content involved in the case.

[0011] According to an embodiment of the present disclosure, a data audit method is applied to a financial institution, a pre-trained large language model is fine-tuned based on a first training sample of the financial institution, and the method further includes: determining a second training sample based on the data to be audited and the prediction result. The pre-trained large language model is fine-tuned using the second training sample.

[0012] The second aspect of the present disclosure provides a data review device, including: a first determination module, for determining the object to be reviewed and the case-related content corresponding to the object to be reviewed in response to input of the data to be reviewed. A calling module, for calling the legal document related to the object to be reviewed. A second determination module, for determining the prompt text according to the case-related content and the legal document. A prediction module, for inputting the prompt text into a pre-trained large language model, and outputting a prediction result representing the risk level of the object to be reviewed. An audit module, for automatically auditing the data to be reviewed with a risk level of a specific risk level according to the prediction result.

[0013] A third aspect of the present disclosure provides an electronic device, comprising: one or more processors; a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the data audit method.

[0014] The fourth aspect of the present disclosure further provides a computer-readable storage medium on which a computer program or instruction is stored. When the computer program or instruction is executed by a processor, the steps of the data audit method are implemented.

[0015] The fifth aspect of the present disclosure also provides a computer program product, including a computer program or instructions, which implement the steps of the above-mentioned data audit method when the above-mentioned computer program or instructions are executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0017] Figure 1 The application scenario diagram of the data audit method, apparatus, device, medium and program product according to the embodiment of the present disclosure is schematically shown;

[0018] Figure 2 A flowchart of a data audit method according to an embodiment of the present disclosure is schematically shown;

[0019] Figure 3 A flowchart of a method for determining a prompt text according to an embodiment of the present disclosure is schematically shown;

[0020] Figure 4 A flowchart of a method for determining a prompt text according to an embodiment of the present disclosure is schematically shown;

[0021] Figure 5 A flowchart of a data audit method according to another embodiment of the present disclosure is schematically shown;

[0022] Figure 6 A flowchart of a method for determining an object to be audited according to an embodiment of the present disclosure is schematically shown;

[0023] Figure 7 A flowchart of a data audit method according to another embodiment of the present disclosure is schematically shown;

[0024] Figure 8 A structural block diagram of a data audit device according to an embodiment of the present disclosure is schematically shown; and

[0025] Fig. 9 A block diagram of an electronic device suitable for implementing a data audit method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0026] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0027] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0028] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0029] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0030] The embodiments of the present disclosure provide a data audit method, device, equipment, medium and program product, which can be used in the field of financial technology or other fields. It should be noted that the data audit method, device, equipment, medium and program product of the present disclosure can be used in the field of financial technology, and can also be used in any field other than the field of financial technology, and the application field of the data audit method, device, equipment, medium and program product of the present disclosure is not limited.

[0031] In the technical solution of the present disclosure, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0032] In the scenario of using personal information for automated decision-making, the methods, devices, and systems provided by the embodiments of the present disclosure provide users with corresponding operation portals for users to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered. The expression "automated decision-making" here refers to the activity of automatically analyzing and evaluating a person's behavioral habits, interests and hobbies, or economic, health, credit status, etc. through computer programs, and making decisions. The expression "expert decision-making" here refers to the activity of making decisions by people who specialize in a certain field, have specialized experience, knowledge and skills, and have reached a certain level of professionalism.

[0033] An embodiment of the present disclosure provides a data audit method, including: in response to input of data to be audited, determining the object to be audited and the case-related content corresponding to the object to be audited. Calling legal documents related to the object to be audited. Determining prompt text based on the case-related content and the legal documents. Inputting the prompt text into a pre-trained large language model, and outputting a prediction result representing the risk level of the object to be audited. Based on the prediction result, automatically auditing the data to be audited with a risk level of a specific risk level. This method determines the prompt text based on the case-related information of the object to be audited and the relevant legal documents called from a specific platform (such as Rong'an e-mail, etc.). Then, the pre-trained large language model is used to generate the risk level of the object to be audited based on the prompt text, and low-risk data is automatically processed, which greatly improves the audit efficiency of legal-related risk data.

[0034] Figure 1 The application scenario diagram of the data audit method, apparatus, device, medium and program product according to the embodiments of the present disclosure is schematically shown.

[0035] like Figure 1 As shown, the application scenario 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0036] The user can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only for example).

[0037] The first terminal device 101, the second terminal device 102, and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0038] The server 105 may be a server that provides various services, such as a background management server (only as an example) that provides support for websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0039] It should be noted that the data audit method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the data audit device provided in the embodiment of the present disclosure can generally be set in the server 105. The data audit method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the data audit device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0040] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0041] The following will be based on Figure 1 The scene described by Figure 2~Figure 7 The data audit method of the embodiment of the present disclosure is described in detail.

[0042] Figure 2 The flowchart of the data audit method according to the embodiment of the present disclosure is schematically shown.

[0043] According to the embodiments of the present disclosure, Figure 2 As shown, the data audit method of this embodiment includes, for example, operations S210 to S250, and the method can be executed by a computer program on corresponding computer hardware.

[0044] In operation S210, in response to input of the data to be reviewed, an object to be reviewed and the case-related content corresponding to the object to be reviewed are determined.

[0045] In some embodiments, the method can be used, for example, in a scenario where a financial institution conducts an anti-money laundering audit on an account involved in a case. In this scenario, the data to be audited is, for example, a customer's account transaction record, the object to be audited is the account holder, and the content involved is an abnormal pattern or high-risk transaction in the transaction record that may involve money laundering behavior.

[0046] For example, the compliance department of a financial institution receives a batch of account transaction records as data to be reviewed. The system automatically identifies each account holder in this batch of data as a subject to be reviewed.

[0047] For another example, the data to be reviewed can also include external investigation data. The system analyzes transaction records and investigation data through a preset algorithm to identify abnormal transaction patterns, such as frequent large-value transfers, cross-border transactions, transactions with high-risk regions or individuals, and suspicious activities or related persons mentioned in the investigation data as the content of the case.

[0048] It is understandable that the data to be reviewed may also be external coordination data, which is used to enable financial institutions to quickly review external coordination instruction tasks.

[0049] In operation S220, the legal documents related to the object to be reviewed are retrieved.

[0050] For example, the system accesses the internal database of a financial institution or an external legal database or a law enforcement agency database to retrieve legal documents related to the account holder, such as court judgments, execution notices, freezing orders, etc., especially legal documents related to execution information, to obtain the judicial records of the account holder.

[0051] In operation S230, a prompt text is determined according to the case content and the legal document.

[0052] For example, the system combines the content involved in the case (such as abnormal transaction patterns, suspicious activities in the cooperative investigation data) with the execution information in the legal document (such as the reason why the account was frozen, the legal case involved, etc.) to generate one or more prompt texts that outline the risk points that the account holder may be involved in.

[0053] In operation S240 , the prompt text is input into a pre-trained large language model, and a prediction result representing the risk level of the object to be reviewed is output.

[0054] For example, the system inputs the generated prompt text into a pre-trained large language model, which has been trained with a large amount of text data and can understand and analyze complex information and predict the risk level.

[0055] The large language model outputs a prediction result based on the prompt text, which represents the risk level of the account holder in numerical or categorical form, such as low risk, medium risk, high risk, etc.

[0056] For example, in the external cooperative investigation task scenario of enhanced due diligence, automatic pre-examination of judicial investigation and freezing based on big models can be achieved.

[0057] Identify the image content of external legal documents and extract the document number and investigation subject. Check the frozen content at the counter using the image of the corresponding transaction code. And connect with Rongan e-letter to obtain the relevant judgment document information of the customer.

[0058] The judicial investigation and freezing data of anti-money laundering customers are scored, and information is extracted and semantically analyzed using a large model. The results are divided into three categories: A, B, and C. A is certain, B is uncertain, and C is directly filtered. A and B enter the manual review stage to identify civil and criminal judicial documents. Civil, protective stop payment, and non-customer documents related data can be directly automatically reviewed, and uncertain and the rest can be pushed to manual review.

[0059] Large model judgment logic:

[0060] 1. Whether it involves a criminal case;

[0061] 2. Check the type of frozen buckle;

[0062] 3. Check the channels for freezing the buckles;

[0063] 4. Whether the content of the document clearly states the type of crime involved or the reason for freezing and detaining;

[0064] 5. Whether the name of the criminal suspect is clear.

[0065] If the large model can query information from other channels, you can add:

[0066] 6. The current status of the customer's accounts;

[0067] 7. Whether the client has document information and a summary of the document content on the Judgment Documents Network;

[0068] 8. Whether the customer is included in the recognized disciplinary list for buying and selling bank cards or accounts.

[0069] According to the results of the large model, external investigations are divided into three categories: Category A: meets 1, 7 and 8 at the same time. Category B: meets 1. Category C: other situations.

[0070] In operation S250, based on the prediction result, the to-be-audited data having a risk level of a specific risk level is automatically audited.

[0071] For example, the system may mark high-risk account transaction records as requiring further manual review based on preset rules, while low-risk or medium-risk records may go through an automatic review process or take other appropriate measures.

[0072] For data to be reviewed that is determined to be of low risk, the system automatically performs the review process, such as checking the compliance of transaction records, verifying the identity information of account holders, etc. If the automatic review passes, no human intervention is required; if the automatic review finds any abnormalities or high-risk signs, the system will automatically upgrade the review level and transfer it to manual further review.

[0073] For example, the system can also record the results of the automatic audit, including the accounts and transaction records that have passed the audit, as well as any accounts that require further manual review. At the same time, the system generates an audit report for review by the compliance department of the financial institution and relevant regulatory authorities.

[0074] By combining multiple data sources, abnormal patterns of account transaction records, legal documents of account holders (especially execution information), and the predictive capabilities of large language models, this embodiment enables financial institutions to efficiently and accurately identify potential high-risk accounts and take necessary compliance measures. This method not only improves audit efficiency, but also enhances the anti-money laundering capabilities of financial institutions, helping to maintain the stability and security of the financial system. The method of the disclosed embodiment can be used to complete the standardized processing of non-criminal external investigations. The production and use of this large model is expected to automatically identify approximately 100,000 external investigation businesses per day, and improve the processing efficiency of external investigation and audit business by front-line personnel by 60%. "Early identification, early warning, early exposure, and early disposal" of various risks.

[0075] In operation S210, user information is obtained.

[0076] In an embodiment of the present disclosure, before obtaining the user's information, the user's consent or authorization may be obtained. For example, before operation S210, a request to obtain the user's information may be issued to the user. When the user agrees or authorizes that the user's information may be obtained, the operation S210 is performed.

[0077] In operation S240, the user information is used to perform subsequent action execution processing to implement terminal operations or information acquisition.

[0078] In the embodiments of the present disclosure, a corresponding operation entry may be provided for the user to choose to agree or reject the automated decision result. That is, before the user information is processed for subsequent action execution, the user's instruction to agree or reject the processing input through the corresponding operation entry may be obtained. If the user agrees to perform the processing, the user information is processed for subsequent action execution, that is, step S240 is executed. If the user refuses to perform the processing, the expert decision process is entered.

[0079] Figure 3A flowchart of a method for determining a prompt text according to an embodiment of the present disclosure is schematically shown. Figure 2 In addition to the operations S210 to S250 described above, the operations S331 to S333 are also included. For the sake of brevity, the description of operations S210 to S250 is omitted here, and the subsequent related method embodiments are similar and will not be repeated.

[0080] According to the embodiments of the present disclosure, Figure 3 As shown, for example, by performing any one of operations S331 to S333, the prompt text is determined according to the case content and the legal document.

[0081] In operation S331, the case content and the legal document are compared to determine the case type of the object to be reviewed.

[0082] For example, the system compares abnormal patterns in transaction records (such as frequent large-value transfers) with the criminal behaviors described in legal documents (such as money laundering, fraud, etc.) to determine the type of case that the account holder may be involved in, such as "criminal case", "civil case", etc., as part of the prompt text.

[0083] In operation S332, specific keywords in the content involved in the case are determined, and the specific keywords represent suspension of payment.

[0084] For example, the system searches for specific keywords in transaction records or investigation data, such as "suspended payment", "account frozen", "funds intercepted", etc. These keywords usually indicate that the account holder may be involved in high-risk transactions or has been taken measures by law enforcement agencies. The system uses these keywords and their context as part of the prompt text.

[0085] In operation S333, the content involved in the case is compared with the legal document to determine the consistency of the account involved in the case of the object to be reviewed.

[0086] For example, the system compares the account information in the transaction record with the account number mentioned in the legal document to determine their consistency. If the account numbers are consistent, the system generates a prompt text containing the account information, such as "Account XXX is suspected of money laundering activities, which is consistent with the account number in the legal document."

[0087] This embodiment describes in detail how to determine the prompt text based on the case content and legal documents, including comparing the case content and legal documents to determine the type of case, searching for specific keywords to characterize the suspension of payment, and comparing the consistency of account information. These methods provide accurate and comprehensive input information for subsequent large language model predictions, which helps to improve the accuracy and efficiency of anti-money laundering audits.

[0088] Figure 4A flowchart of a method for determining a prompt text according to an embodiment of the present disclosure is schematically shown.

[0089] According to the embodiments of the present disclosure, Figure 4 As shown, for example, through any one of operations S441 to S443, the prompt text is input into the pre-trained large language model, and a prediction result representing the risk level of the object to be reviewed is output.

[0090] In operation S441, when the case type is a civil case, the risk level of the data to be reviewed is determined to be a specific risk level.

[0091] For example, if the prompt text clearly states that the case type is a civil case (such as a contract dispute, debt dispute, etc.), the system believes that such cases usually do not involve financial crimes such as money laundering, and therefore determines the risk level of the data to be reviewed as "low risk" (the specific level can be set according to the internal regulations of the financial institution) and marks it for automatic review.

[0092] In operation S442, when the content involved in the case has a specific keyword, the risk level of the data to be reviewed is determined to be a specific risk level.

[0093] For example, if the prompt text contains specific keywords that indicate that the account holder may be involved in low-risk behavior, such as "protective stop payment", "normal transaction", "compliant operation", etc., the system will also determine that the risk level of the data to be reviewed is a low risk level and mark it for automatic review.

[0094] In operation S443, when the account number involved in the object to be reviewed is inconsistent with the account number in the legal document, the risk level of the data to be reviewed is determined to be a specific risk level.

[0095] Although this situation usually indicates that further investigation is needed, if the system can confirm that the inconsistency is due to reasonable reasons such as information entry errors, account changes, and there are no other high-risk indicators, the system can prudently determine that the risk level of the data to be reviewed is low risk and mark it for automatic review. However, automatic review in this case should be conducted with caution and may require additional verification steps.

[0096] For example, when the account involved in the case is inconsistent with the account to be executed in the legal document, it is considered that one of the account involved in the case and the account in the legal document is not the customer's own account, and thus the audit can be automatically performed.

[0097] This embodiment describes in detail how to output the prediction results of the risk level of the object to be audited through a large language model based on the content of the prompt text, and pays special attention to determining the risk level of the data to be audited as a low risk level that can be automatically audited in certain circumstances. By comprehensively applying these methods, financial institutions can improve audit efficiency and reduce the workload of manual audits while ensuring compliance.

[0098] Figure 5 The flowchart of a data audit method according to another embodiment of the present disclosure is schematically shown.

[0099] According to the embodiments of the present disclosure, Figure 5 As shown, the data audit method, for example, also includes operations S510 to S520.

[0100] In operation S510, it is determined that the reason for automatically reviewing the data to be reviewed is any one of the following: the case type is a civil case, has a specific keyword, and the account number involved is inconsistent.

[0101] In operation S520, the reason for the automatic review is displayed on the review page.

[0102] In some embodiments, the data audit method further details how to determine the reasons for automatically auditing the audited data, and displays these reasons on the audit page. Through this step, the compliance department of the financial institution can more clearly understand the decision basis of the automatic audit, and improve the transparency and traceability of the audit process.

[0103] If the predicted result is a low risk level, the system further determines the reasons for the automatic review. These reasons include: the case is a civil case, there are specific keywords (indicating low-risk behavior), the account involved is inconsistent with the account in the legal document but is judged to be reasonable by the system (such as information entry errors, account changes, etc.). The system records these reasons as a decision basis for automatic review.

[0104] At the same time, the system is equipped with an audit page, which not only contains detailed information on the data to be audited and the risk level prediction results, but also clearly displays the reasons for the automatic audit. For example, if the reason for the automatic audit is "the case involved is a civil case", the page will clearly indicate: "This account transaction record has been automatically audited because the case involved is a civil case and does not involve high-risk behaviors such as money laundering." This display method helps compliance department personnel quickly understand the basis for audit decisions and improves the transparency of the audit process.

[0105] For example, you can add the query condition [Large Model Identification Type] on the review page, which displays all by default and can filter by category A, B, and C. Add [Large Model Identification Type] to the query results and display it after the case type field.

[0106] For example, the system can also record the results of all audit processes, including automatic and manual audits. At the same time, the system generates audit reports that list in detail the audit status, risk level, automatic audit reason (if applicable), and any actions that need to be taken for each audited data. These reports can be reviewed by the compliance department of financial institutions and relevant regulatory authorities.

[0107] This embodiment describes in detail how to determine the reasons for automatic review of the data to be reviewed in the data review method, and display these reasons on the review page. Through this step, the compliance department of the financial institution can more clearly understand the decision basis of the automatic review, and improve the transparency and traceability of the review process. At the same time, this method also helps to reduce the workload of manual review and improve the efficiency of review.

[0108] According to an embodiment of the present disclosure, the data to be reviewed includes any one or more of the description documents, approval forms and notices issued by law enforcement agencies related to property freezing.

[0109] In some embodiments, a fast and automated data review method is proposed for low-risk external investigation data such as civil cases, protective stop payments, and inconsistent investigation accounts. Through this method, financial institutions can efficiently process these low-risk data, reduce the workload of manual review, and ensure the accuracy and compliance of the review.

[0110] Financial institutions receive data from law enforcement agencies on asset freeze-related explanatory documents, approval forms and notices awaiting review. These data may involve civil cases, protective stop payments or inconsistencies in account numbers.

[0111] For example, the system can also automatically pre-process the received data for review, including data cleaning, format conversion, and key information extraction. This step is to ensure the accuracy and readability of the data in preparation for subsequent data analysis.

[0112] The system uses a large model to assess the risk level of the data to be reviewed. In this embodiment, the system pays special attention to data related to low-risk situations such as civil cases, protective stop payments, and inconsistent accounts. If the data meets these low-risk characteristics, it will be marked as a low-risk level.

[0113] For data to be reviewed that is marked as low risk, the system automatically performs the review process. This process may include steps such as data comparison, information verification, and compliance checks. The results of the system's automatic review will be directly generated and stored without manual intervention.

[0114] It should be noted that although most low-risk data can be processed through automated review, in some cases the system may find anomalies or information that requires further verification. At this time, the system will automatically trigger the manual review process and transfer the relevant data to the compliance department for detailed review. The results of the manual review will be updated in the system and replace the previous automated review results.

[0115] For example, the system can also generate audit reports regularly, detailing the processing status of all data to be audited, including data that has passed the automated audit, data that has been manually audited, and any anomalies or actions that need to be taken. These reports can be reviewed by financial institution management, compliance departments, and law enforcement agencies to ensure transparency and compliance of the audit process.

[0116] For another example, financial institutions can also regularly evaluate the effectiveness and accuracy of automated review processes and optimize and improve them according to actual needs. For example, they can adjust risk assessment rules, update algorithm models, or add new data types to improve the efficiency and accuracy of automated reviews.

[0117] This embodiment introduces an automated review process to achieve rapid processing of low-risk external investigation data involving civil cases, protective stop payments, and inconsistent investigation accounts. This method not only improves the efficiency of the review and reduces the manual workload, but also ensures the accuracy and compliance of the review.

[0118] Figure 6 The flowchart of the method for determining an object to be reviewed according to an embodiment of the present disclosure is schematically shown.

[0119] According to an embodiment of the present disclosure, the data to be reviewed is multimodal data, which includes documents, pictures, audio and video, etc. Figure 6 As shown, for example, through at least one of operations S611 to S612, in response to input of the data to be reviewed, the object to be reviewed and the case-related content corresponding to the object to be reviewed are determined.

[0120] In operation S611, a pre-trained large language model is used to extract information from multimodal data to determine the object to be reviewed and the content involved.

[0121] In some embodiments, a large language model with multimodal data processing capabilities can be directly used to process the data to be reviewed.

[0122] For example, the data to be audited can be initially formatted and cleaned to ensure smooth processing of subsequent steps.

[0123] Then, the pre-trained large language model is used to deeply analyze the document content. The model understands the contextual information in the document and identifies key entities (such as names of people, places, and organizations) and events (such as transactions, disputes, etc.). Based on the identified key entities and events, the subject to be reviewed (such as the individuals, organizations, or events involved) and the related content of the case are determined.

[0124] In operation S612, the image is recognized through OCR to determine the object to be reviewed and the content involved.

[0125] In some embodiments, the image may be first subjected to OCR processing to convert the text information in the image into an editable text format, and the converted text may be further analyzed using natural language processing technology to identify key information in the image (such as license plate number, text description, etc.).

[0126] Based on the identified key information, determine the objects to be reviewed in the image and the content involved.

[0127] For example, you can also analyze video data frame by frame to extract key frames and key information in the video. Use computer vision technology to identify objects, people, and scenes in the video, as well as their behaviors and interactions. Combine natural language processing technology (such as speech recognition) to analyze the audio content in the video and extract key information from the speech. Video data can be, for example, a counter transaction video.

[0128] Based on the results of the video analysis, determine the objects to be reviewed in the video (such as specific people, behaviors or events) and their content involved in the case.

[0129] In some embodiments, the results of large language models, OCR and video analysis can also be summarized and integrated. Through cross-validation and logical reasoning, the accuracy and completeness of the subject to be reviewed and the content involved can be further confirmed.

[0130] Output the final objects to be reviewed and the content involved for use in subsequent review processes.

[0131] This embodiment combines a large language model, OCR, and video analysis to determine the objects to be reviewed and the content involved in the case. It has high efficiency and accuracy when processing multimodal data, and provides strong support for subsequent review processes.

[0132] Figure 7 The flowchart of a data audit method according to another embodiment of the present disclosure is schematically shown.

[0133] According to an embodiment of the present disclosure, the data audit method is applied to a financial institution, and the pre-trained large language model is fine-tuned based on the first training sample of the financial institution, such as Figure 7As shown, the data audit method, for example, also includes operations S710 to S720.

[0134] In operation S710, a second training sample is determined according to the data to be reviewed and the prediction result.

[0135] In operation S720, the pre-trained large language model is fine-tuned using the second training sample.

[0136] In some embodiments, when a financial institution processes a large amount of business data, it needs to ensure the accuracy and integrity of the data. First, data is collected from multiple reliable data sources, such as internal business systems, external credit reporting agencies, government departments, etc.

[0137] Then, the collected data is preliminarily checked and cleaned to remove erroneous data, duplicate data and incomplete data, and obtain a first training sample. The first training sample includes known correct data and erroneous data, as well as corresponding audit results.

[0138] The pre-trained large language model is fine-tuned using the first training samples of the financial institution to adapt it to the data audit needs of the financial institution.

[0139] At the same time, for each audit task, the data to be audited is input into the fine-tuned large language model to obtain the prediction results. The prediction results include whether the data is accurate, complete, and whether there are potential errors.

[0140] By comparing the data to be reviewed with the predicted results, samples with incorrect or uncertain predictions are screened out. These samples are manually reviewed to determine their authenticity and accuracy.

[0141] The manually reviewed samples are used as the second training samples to further fine-tune the large language model.

[0142] Using the second training sample to fine-tune the pre-trained large language model again can improve the accuracy and generalization ability of the model.

[0143] During the fine-tuning process, you can adjust the model's parameters, optimize the algorithm, etc. to better meet the data audit needs of financial institutions. Verify the fine-tuned model to ensure its accuracy and reliability in practical applications. If the model is found to have problems such as overfitting or underfitting, make adjustments and optimizations in a timely manner.

[0144] Through the method of this embodiment, financial institutions can process large amounts of business data more efficiently and ensure the accuracy and completeness of the data. At the same time, with the continuous increase of the second training samples and the continuous fine-tuning of the model, the accuracy and generalization ability of the large language model will continue to improve, providing more powerful support for the data audit work of financial institutions.

[0145] Based on the above data audit method, the present disclosure also provides a data audit device. Figure 8 The data audit device is described in detail.

[0146] Figure 8 The structural block diagram of the data audit device according to the embodiment of the present disclosure is schematically shown.

[0147] like Figure 8 As shown, the data audit device 800 of this embodiment includes, for example: a first determination module 810 , a calling module 820 , a second determination module 830 , a prediction module 840 and an audit module 850 .

[0148] The first determination module 810 is used to determine the object to be reviewed and the case-related content corresponding to the object to be reviewed in response to the input of the data to be reviewed. In one embodiment, the first determination module 810 can be used to perform the operation S210 described above, which will not be repeated here.

[0149] The calling module 820 is used to call the legal documents related to the object to be reviewed. In one embodiment, the calling module 820 can be used to perform the operation S220 described above, which will not be repeated here.

[0150] The second determination module 830 is used to determine the prompt text according to the case content and the legal document. In one embodiment, the second determination module 830 can be used to perform the operation S230 described above, which will not be described in detail here.

[0151] The prediction module 840 is used to input the prompt text into the pre-trained large language model and output a prediction result representing the risk level of the object to be reviewed. In one embodiment, the prediction module 840 can be used to perform the operation S240 described above, which will not be repeated here.

[0152] The review module 850 is used to automatically review the to-be-reviewed data with a risk level of a specific risk level according to the prediction result. In one embodiment, the review module 850 can be used to perform the operation S250 described above, which will not be described in detail here.

[0153] According to an embodiment of the present disclosure, any multiple modules of the first determination module 810, the call module 820, the second determination module 830, the prediction module 840 and the audit module 850 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the first determination module 810, the call module 820, the second determination module 830, the prediction module 840 and the audit module 850 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in a suitable combination of any of them. Alternatively, at least one of the first determination module 810, the calling module 820, the second determination module 830, the prediction module 840 and the audit module 850 may be at least partially implemented as a computer program module, which may perform a corresponding function when executed.

[0154] According to an embodiment of the present disclosure, the second determination module 830 includes any one of a first comparison submodule, a first determination submodule, and a second comparison submodule.

[0155] The first comparison submodule is used to compare the case content and the legal documents to determine the case type of the object to be reviewed.

[0156] The first determination submodule is used to determine specific keywords in the content involved in the case, and the specific keywords represent the suspension of payment.

[0157] The second comparison submodule is used to compare the content involved in the case with the legal documents to determine the consistency of the account involved in the case of the object to be reviewed.

[0158] According to an embodiment of the present disclosure, the prediction module 840 includes any one of a second determination submodule, a third determination submodule, and a fourth determination submodule.

[0159] The second determination submodule is used to determine that the risk level of the data to be reviewed is a specific risk level when the case type is a civil case.

[0160] The third determination submodule is used to determine that the risk level of the data to be reviewed is a specific risk level when the content involved in the case has specific keywords.

[0161] The fourth determination submodule is used to determine that the risk level of the data to be reviewed is a specific risk level when the account number involved in the case of the object to be reviewed is inconsistent with the account number in the legal document.

[0162] According to an embodiment of the present disclosure, the data audit device 800 further includes, for example: a third determination module and a display module.

[0163] The third determination module is used to determine that the reason for automatically reviewing the data to be reviewed is any one of the following: the case type is a civil case, has specific keywords, and the account number involved is inconsistent.

[0164] The display module is used to display the reasons for automatic review on the review page.

[0165] According to an embodiment of the present disclosure, the data to be reviewed is multimodal data, and the multimodal data includes documents, pictures and videos. The first determination module 810 includes an extraction submodule and an identification submodule.

[0166] The extraction submodule is used to extract information from multimodal data using a pre-trained large language model to determine the objects to be reviewed and the content involved.

[0167] The recognition submodule is used to recognize images through OCR to determine the objects to be reviewed and the content involved.

[0168] According to an embodiment of the present disclosure, the data audit method is applied to a financial institution, and the pre-trained large language model is fine-tuned based on the first training sample of the financial institution. The data audit device, for example, also includes: a fourth determination module and a fine-tuning module.

[0169] The fourth determination module is used to determine the second training sample according to the data to be reviewed and the prediction result.

[0170] The fine-tuning module is used to fine-tune the pre-trained large language model using the second training sample.

[0171] Fig. 9 A block diagram of an electronic device suitable for implementing a data audit method according to an embodiment of the present disclosure is schematically shown.

[0172] like Fig. 9As shown, the electronic device 900 according to an embodiment of the present disclosure includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage part 908 to a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include an onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0173] In RAM 903, various programs and data required for the operation of electronic device 900 are stored. Processor 901, ROM 902 and RAM 903 are connected to each other via bus 904. Processor 901 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 902 and / or RAM 903. It should be noted that the program can also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0174] According to an embodiment of the present disclosure, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to the bus 904. The electronic device 900 may further include one or more of the following components connected to the input / output (I / O) interface 905: an input portion 906 including a keyboard, a mouse, etc.; an output portion 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 908 including a hard disk, etc.; and a communication portion 909 including a network interface card such as a LAN card, a modem, etc. The communication portion 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 910 as needed, so that a computer program read therefrom is installed into the storage portion 908 as needed.

[0175] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the data audit method according to the embodiment of the present disclosure is implemented.

[0176] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: 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), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 902 and / or RAM 903 described above and / or one or more memories other than ROM 902 and RAM 903.

[0177] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the data audit method provided by the embodiment of the present disclosure.

[0178] The above functions defined in the system / device of the embodiment of the present disclosure are performed when the computer program is executed by the processor 901. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0179] In one embodiment, the computer program may be based on a tangible storage medium such as an optical storage device, a magnetic storage device, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of a signal on a network medium, and downloaded and installed through the communication part 909, and / or installed from a removable medium 911. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0180] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, the above functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0181] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0182] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0183] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0184] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A data audit method, characterized in that: The method comprises: In response to input of the data to be reviewed, determining the object to be reviewed and the case-related content corresponding to the object to be reviewed; Retrieve legal documents related to the subject to be reviewed; Determine the prompt text according to the case content and the legal document; Input the prompt text into a pre-trained large language model, and output a prediction result representing the risk level of the object to be reviewed; According to the prediction result, the to-be-audited data having a risk level of a specific risk level is automatically audited.

2. The method according to claim 1, characterized in that The step of determining the prompt text according to the case content and the legal document includes any of the following: Compare the case content and the legal document to determine the case type of the subject to be reviewed; Determine specific keywords in the content involved in the case, wherein the specific keywords represent suspension of payment; Compare the content involved in the case with the legal document to determine the consistency of the account involved in the case of the subject to be reviewed.

3. The method according to claim 2, characterized in that The inputting of the prompt text into a pre-trained large language model and outputting a prediction result representing the risk level of the object to be reviewed include any of the following: In the case where the case type is a civil case, determining the risk level of the data to be reviewed to be the specific risk level; In the case where the case-related content has the specific keyword, determining the risk level of the data to be reviewed to be the specific risk level; In the case where the account number involved in the object to be reviewed is inconsistent with the account number in the legal document, the risk level of the data to be reviewed is determined to be the specific risk level.

4. The method according to claim 3, characterized in that The method further comprises: Determining that the reason for automatically reviewing the data to be reviewed is any one of the following: the case type is a civil case, the specific keyword is present, and the account number involved is inconsistent; The reason for the automatic review is displayed on the review page.

5. The method according to claim 1, characterized in that: The data to be reviewed includes any one or more of the explanatory documents, approval forms and notices related to property freezing issued by the law enforcement agency.

6. The method according to claim 5, characterized in that The data to be reviewed is multimodal data, and the multimodal data includes documents, pictures, audio and video. The determination of the object to be reviewed and the case-related content corresponding to the object to be reviewed in response to the input of the data to be reviewed includes at least one of the following: Using the pre-trained large language model to extract information from the multimodal data to determine the object to be reviewed and the content involved in the case; The image is identified through OCR to determine the object to be reviewed and the content involved in the case.

7. The method according to claim 1, characterized in that The data audit method is applied to a financial institution, the pre-trained large language model is fine-tuned based on a first training sample of the financial institution, and the method further includes: Determining a second training sample according to the data to be reviewed and the prediction result; The pre-trained large language model is fine-tuned using the second training sample.

8. A data audit device, characterized in that: The device comprises: A first determination module, configured to determine an object to be reviewed and the case-related content corresponding to the object to be reviewed in response to input of the data to be reviewed; A calling module, used to call the legal documents related to the object to be reviewed; A second determination module is used to determine a prompt text according to the case content and the legal document; A prediction module, used for inputting the prompt text into a pre-trained large language model and outputting a prediction result representing the risk level of the object to be reviewed; The audit module is used to automatically audit the data to be audited with a risk level of a specific risk level according to the prediction result.

9. An electronic device, comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the data audit method according to any one of claims 1 to 7.

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

11. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the data audit method according to any one of claims 1 to 7 are implemented.

Images