Intelligent agent-based data verification method and device, equipment and medium
By automating the processing of bank regulatory reporting data through an intelligent agent collaboration framework, the problem of low efficiency in manual verification has been solved, and efficient and accurate data verification has been achieved.
Patent Information
- Application Number
- CN202511710882.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-20
- Publication Date
- 2026-02-10
AI Technical Summary
In the current technology, banks rely on manual verification of data submitted to regulators, which is inefficient and prone to errors, making it difficult to meet the timeliness requirements of regulatory reporting.
A data verification method based on intelligent agents is adopted. Through a collaborative framework of management intelligent agents, expert intelligent agents, construction intelligent agents and verification intelligent agents, the regulatory reporting relationship table and specification documents are processed automatically, association rules are constructed and verified.
It achieves a high degree of automation in data verification, improving efficiency and accuracy, reducing human intervention, and ensuring the accuracy and adaptability of verification results.
Smart Images
Figure CN121503471A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, and in particular to a data verification method, apparatus, device, and medium based on intelligent agents. Background Technology
[0002] Within the banking system, commercial banks are required to regularly submit substantial amounts of business and financial data to financial regulatory agencies. This data is typically submitted in the form of dozens or even hundreds of strictly defined and structured relational tables. The data entry standards for these relational tables are derived from regulatory reporting requirements documents, which detail the business implications of each table, the definition of each field, the filling specifications, the cross-referencing relationships between data, and the key areas of regulatory focus. To ensure the accuracy and compliance of the submitted data, banks need to internally verify the data to promptly identify errors or potential risks that do not meet regulatory requirements. Simultaneously, regulatory agencies also need to verify the submitted data to assess the quality of bank data, verify its compliance status, and serve as the data foundation for off-site supervision and macro-prudential management.
[0003] Currently, the verification of regulatory data relies heavily on manual checks. This involves business or compliance personnel manually reviewing regulatory documents, comparing, calculating, and verifying data relationships in reports based on their experience and understanding. However, this method is time-consuming when dealing with massive amounts of reports and complex cross-table rules, making it difficult to meet the timeliness requirements of regulatory reporting. Furthermore, manual verification is prone to errors and omissions due to fatigue or negligence, and accuracy is difficult to guarantee.
[0004] Therefore, how to intelligently verify regulatory data and improve the efficiency of data verification has become an urgent problem to be solved. Summary of the Invention
[0005] This invention provides a data verification method, apparatus, device, and medium based on intelligent agents to address the problem of how to intelligently verify regulatory reporting data and improve the efficiency of data verification.
[0006] A data verification method based on intelligent agents is applied to a data verification platform, the data verification platform including a management intelligent agent, an expert intelligent agent, a construction intelligent agent, and a verification intelligent agent, the data verification method comprising: Based on the management agent, obtain the regulatory reporting relationship table and regulatory reporting specification document to be verified, and select at least one target regulatory reporting relationship table from all the regulatory reporting relationship tables to be verified. Send all target regulatory reporting relationship tables to the construction agent, send all regulatory reporting relationship tables to be verified to the verification agent, and initialize the expert agent according to the regulatory reporting specification document to obtain the initialized expert agent; Using the construction agent and the initialization expert agent, rules are constructed based on the table of all target regulatory reporting relationships and the regulatory reporting specification document to obtain association rules, and the association rules are sent to the verification agent; Using the aforementioned verification agent, the verification results are obtained by verifying all regulatory reporting relationship tables to be verified according to the association rules.
[0007] A data verification device based on an intelligent agent is applied to a data verification platform, the data verification platform including a management intelligent agent, an expert intelligent agent, a construction intelligent agent, and a verification intelligent agent, the data verification device comprising: The data acquisition module is used to acquire, based on the management intelligent agent, the regulatory reporting relationship table to be verified and the regulatory reporting specification document, and select at least one target regulatory reporting relationship table from all the regulatory reporting relationship tables to be verified. The distribution module is used to send all target regulatory reporting relationship tables to the construction agent, send all regulatory reporting relationship tables to be verified to the verification agent, and initialize the expert agent according to the regulatory reporting specification document to obtain the initialized expert agent. The rule building module is used to build rules based on the table of all target regulatory reporting relationships and the regulatory reporting specification document using the building agent and the initialization expert agent, obtain associated rules, and send the associated rules to the verification agent; The data verification module is used to verify all regulatory reporting relationship tables to be verified according to the association rules, using the verification intelligent agent, and to obtain the verification results.
[0008] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described agent-based data verification method.
[0009] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described agent-based data verification method.
[0010] The aforementioned agent-based data verification method is applied to a data verification platform, which includes a management agent, an expert agent, a construction agent, and a verification agent. The management agent obtains the regulatory reporting relationship tables and regulatory reporting specification documents to be verified. At least one target regulatory reporting relationship table is selected from all the tables to be verified. All target regulatory reporting relationship tables are sent to the construction agent, and all the regulatory reporting relationship tables to be verified are sent to the verification agent. The expert agent is initialized according to the regulatory reporting specification documents to obtain an initialized expert agent. Using the construction agent and the initialized expert agent, rules are constructed based on all target regulatory reporting relationship tables and the regulatory reporting specification documents to obtain association rules. These association rules are sent to the verification agent. The verification agent then verifies all the regulatory reporting relationship tables to be verified according to the association rules to obtain the verification results.
[0011] By constructing a multi-agent collaboration framework, the automated information transmission and collaborative flow between agents enable the entire data verification process to proceed rapidly, achieving a high degree of automation in the data verification process. This significantly reduces the need for manual intervention and improves the efficiency of data verification. Furthermore, by extracting target regulatory reporting relationships from the regulatory reporting relationship table to be verified and constructing rules, and then verifying the entire regulatory reporting relationship table to be verified based on these targeted rules, the verification rules originate from the data itself and feed back into the data verification, improving the accuracy and adaptability of the verification. Thus, while improving efficiency, it also improves the accuracy and reliability of data verification. Attached Figure Description
[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0013] Figure 1 This is a schematic diagram of an application environment for a data verification method based on an intelligent agent according to an embodiment of the present invention; Figure 2 This is a flowchart of a data verification method based on an intelligent agent according to an embodiment of the present invention; Figure 3 This is another flowchart of a data verification method based on an intelligent agent in one embodiment of the present invention; Figure 4 This is another flowchart of a data verification method based on an intelligent agent in one embodiment of the present invention; Figure 5This is another flowchart of a data verification method based on an intelligent agent in one embodiment of the present invention; Figure 6 This is another flowchart of a data verification method based on an intelligent agent in one embodiment of the present invention; Figure 7 This is another flowchart of a data verification method based on an intelligent agent in one embodiment of the present invention; Figure 8 This is another flowchart of a data verification method based on an intelligent agent in one embodiment of the present invention; Figure 9 This is a schematic diagram of a data verification device based on an intelligent agent according to an embodiment of the present invention; Figure 10 This is a schematic diagram of a computer device according to an embodiment of the present invention. Detailed Implementation
[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0015] The agent-based data verification method provided in this invention can be applied to, for example... Figure 1 The application environment shown. Specifically, this agent-based data verification method is applied in a data verification platform, which includes, for example, […]. Figure 1 The diagram illustrates a client and server that communicate over a network to address the challenge of intelligently verifying regulatory data and improving verification efficiency. The client, also known as the user terminal, is the program that provides local services to the client, corresponding to the server. The client can be installed on, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be a standalone server or a server cluster consisting of multiple servers.
[0016] In one embodiment, such as Figure 2 As shown, a data verification method based on intelligent agents is provided. This data verification method is applied to a data verification platform, which includes a management intelligent agent, an expert intelligent agent, a construction intelligent agent, and a verification intelligent agent. The data verification method includes the following steps: Step S201: Based on the management agent, obtain the regulatory reporting relationship table and regulatory reporting specification document to be verified, and select at least one target regulatory reporting relationship table from all the regulatory reporting relationship tables to be verified.
[0017] Step S202: Send all target regulatory reporting relationship tables to the construction agent, send all regulatory reporting relationship tables to be verified to the verification agent, and initialize the expert agent according to the regulatory reporting specification document to obtain the initialized expert agent.
[0018] In this embodiment, the data verification platform can refer to a software system used to automate data verification tasks. For example, the data verification platform can be an internal compliance system for financial institutions (such as banks) to conduct self-checks and verifications of regulatory reporting data before submission, or it can be a regulatory system for regulatory agencies to conduct batch reviews and quality assessments of received regulatory reporting data from financial institutions. The management agent can refer to an agent responsible for coordinating verification tasks, scheduling other agents, and managing data flow and processes. The expert agent can refer to an agent that can parse specification documents and provide specification queries for other agents. The construction agent can refer to an agent that can extract, summarize, and generate specific verification rules from the regulatory reporting data to be verified. The verification agent can refer to the regulatory agent to be verified according to the established rules. The intelligent agent that verifies the submitted data includes: a regulatory reporting relationship table (which can be a data table with a predetermined structure and business meaning filled out by financial institutions and submitted to regulatory agencies), a regulatory reporting relationship table as structured data input to the data verification platform, a regulatory reporting specification document (which can be a specification document issued by regulatory agencies that clearly stipulates the format, scope, correlation, and business rules of the submitted data), a regulatory reporting specification document as unstructured data input to the data verification platform, a target regulatory reporting relationship table (which can be a relationship table extracted from the regulatory reporting relationship table to be verified for rule construction), and an initialization expert intelligent agent (which can be an intelligent agent that has loaded and parsed the business rules and logic in the regulatory reporting specification document, can answer questions in the regulatory field, and is in a collaborative ready state).
[0019] Specifically, by managing the intelligent agent, the regulatory reporting relationship table to be verified and the corresponding regulatory reporting specification document are obtained. At least one regulatory reporting relationship table is extracted from all the regulatory reporting relationship tables to be verified as the target regulatory reporting relationship table. All target regulatory reporting relationship tables are sent to the construction intelligent agent. All regulatory reporting relationship tables to be verified are sent to the verification intelligent agent. And according to the regulatory reporting specification document, the expert intelligent agent is initialized to obtain the initialized expert intelligent agent.
[0020] Step S203: Using the construction agent and initialization agent, construct rules based on all target regulatory reporting relationship tables and regulatory reporting specification documents to obtain association rules, and send the association rules to the verification agent.
[0021] Step S204: Using the verification agent, verify all regulatory reporting relationship tables to be verified according to the association rules to obtain the verification results.
[0022] In this embodiment, the association rule can refer to the logical judgment conditions used to verify data compliance derived from the analysis of the structure and content of the target regulatory reporting relationship table and the regulatory reporting specification document. The verification result can refer to the result of verifying the regulatory reporting relationship table to be verified based on the association rule.
[0023] Specifically, the system constructs intelligent agents and initializes expert intelligent agents to build rules based on all target regulatory reporting relationship tables and regulatory reporting specification documents, thereby obtaining association rules. These association rules are then sent to the verification intelligent agent, which verifies all regulatory reporting relationship tables to be verified based on the association rules, and obtains the verification results.
[0024] Optionally, the data verification platform also includes a transformation agent for natural language conversion of rules. After obtaining the association rules, the complex association rule form can be converted into a natural language form that is easy for bank personnel and regulators to understand, based on the transformation agent. For example, the check rule is: For corporate clients identified as 'significant related group members', the total contract amount (to be calculated on a consolidated basis) under all effective loan agreements and guarantee agreements and the type of their related group recorded in the "Related Group Information" table should comply with the total credit ratio limit stipulated in the regulatory document 'Article XX'.
[0025] In this embodiment, by constructing a multi-agent collaborative framework, the automated information transmission and collaborative flow between agents enable the entire data verification process to proceed rapidly, achieving a high degree of automation in the data verification process. This significantly reduces the need for manual intervention, improves the efficiency of data verification, and extracts target regulatory reporting relationships from the regulatory reporting relationship table to be verified to construct rules. Then, based on these targeted rules, the entire regulatory reporting relationship table to be verified is verified. This ensures that the verification rules originate from the data itself and feed back into the data verification, improving the accuracy and adaptability of the verification. Thus, while improving efficiency, it also improves the accuracy and reliability of data verification.
[0026] In one embodiment, such as Figure 3 As shown, a data verification method based on an intelligent agent is provided. In step S202 above, the expert intelligent agent is initialized according to the regulatory reporting specification document to obtain the initialized expert intelligent agent, including the following steps: Step S301: Determine the chapters and clauses in the regulatory reporting specification document, and divide the regulatory reporting specification document into blocks according to the chapters and clauses to obtain text blocks.
[0027] Step S302: Using a preset embedding model, construct a vector index corresponding to the regulatory reporting specification document based on each text block and the corresponding chapter and clause to which the text block belongs.
[0028] Step S303: Initialize the expert agent according to the vector index to obtain the initialized expert agent.
[0029] In this embodiment, a text block can refer to a structured text unit with independent business meaning and logical integrity obtained by segmenting regulatory reporting standard documents based on chapters and clauses. The preset embedding model can refer to a pre-trained deep learning model that can convert the semantic information of text blocks and their chapters and clauses into numerical vector representations in a high-dimensional space. The vector index can refer to a structured database that stores the vectors corresponding to all text blocks and their chapters and clauses and supports similarity retrieval. It is the core component for expert intelligent agents to quickly find knowledge.
[0030] Specifically, the chapters and clauses in the regulatory reporting standard document are determined. Based on the chapters and clauses, the regulatory reporting standard document is divided into blocks to obtain text blocks. Through a preset embedding model, a vector index corresponding to the regulatory reporting standard document is constructed based on each text block and the corresponding chapter and clause. Based on the vector index, the expert agent is initialized to obtain the initialized expert agent.
[0031] In this embodiment, unstructured regulatory documents are segmented into text blocks with independent semantics according to chapters and clauses. These blocks are then converted into vector indexes containing the contextual semantics of the chapters and clauses using a preset embedding model. Finally, the initialization of the expert agent is completed based on this index. This systematically transforms complex document knowledge into machine-understandable and searchable semantic vectors, enabling the expert agent to accurately and efficiently understand and invoke specific clauses in the regulatory documents. This provides a solid knowledge foundation for the subsequent automatic rule construction, thereby improving the accuracy and intelligence level of the entire data verification process from the source.
[0032] In one embodiment, such as Figure 4 As shown, a data verification method based on intelligent agents is provided. The intelligent agents include a first intelligent agent for correlation analysis, a second intelligent agent for predicate construction, and a third intelligent agent for rule construction. In step S203 above, the intelligent agents and the initialization expert intelligent agent are used to construct rules based on all target regulatory reporting relationship tables and regulatory reporting specification documents to obtain association rules, including the following steps: Step S401: Using the first constructed intelligent agent and the initialized expert intelligent agent, according to the regulatory reporting specification document, analyze the correlation between all target regulatory reporting relationship tables and between fields in the target regulatory reporting relationship tables, and obtain the correlation analysis results.
[0033] Step S402: Using the second building agent and the initialization expert agent, predicates are constructed based on the regulatory reporting documents and the correlation analysis results to obtain the predicate construction results.
[0034] Step S403: Using the third building agent and the initialization expert agent, rules are constructed based on the regulatory reporting specification documents and predicate construction results to obtain association rules.
[0035] In this embodiment, the first constructing agent can refer to an agent used for correlation analysis, the second constructing agent can refer to an agent used for predicate construction, the third constructing agent can refer to an agent used for rule construction, the correlation analysis result can refer to the result of analyzing the correlation between fields between tables and within tables, and the predicate construction result can refer to the result of predicate construction based on regulatory reporting standard documents and correlation analysis results.
[0036] Specifically, the first construction agent, combined with the initialization expert agent's ability to query knowledge of regulatory reporting standard documents, analyzes the correlation between all target regulatory reporting relationship tables and between fields in the target regulatory reporting relationship tables to obtain correlation analysis results. The second construction agent, combined with the initialization expert agent's ability to query knowledge of regulatory reporting standard documents, constructs predicates based on the correlation analysis results to obtain predicate construction results. The third construction agent, combined with the initialization expert agent's ability to query knowledge of regulatory reporting standard documents, constructs rules based on the predicate construction results to obtain association rules.
[0037] In this embodiment, by constructing a multi-agent collaborative framework, the complex task of rule discovery is decomposed into multiple sub-tasks (such as document analysis, field correlation judgment, predicate construction, rule construction, etc.). The automated information transmission and collaborative flow between agents enable the entire rule discovery process to proceed quickly, improving the efficiency of rule discovery and thus improving the efficiency of data verification.
[0038] In one embodiment, such as Figure 5 As shown, a data verification method based on intelligent agents is provided. In step S301 above, a first intelligent agent is constructed and an expert intelligent agent is initialized. According to the regulatory reporting specification document, the correlation between all target regulatory reporting relationship tables and between fields in the target regulatory reporting relationship tables is analyzed to obtain the correlation analysis results. The method includes the following steps: Step S501: Using the first intelligent agent, determine the fields and corresponding metadata of each target regulatory reporting relationship table. Based on the metadata, analyze the correlation between any two fields in all target regulatory reporting relationship tables to obtain field analysis results. From all field analysis results, determine the two fields corresponding to the field analysis results that meet the preset first condition as candidate field pairs.
[0039] Step S502: Using the initialization expert agent, perform semantic verification on each candidate field pair according to the regulatory reporting specification document, obtain the semantic verification result, and determine the target field pair from all candidate field pairs based on the semantic verification result.
[0040] Step S503: Using the first constructed agent, based on the target field pairs, filter from all target regulatory reporting relationship tables to obtain the isolated regulatory reporting relationship table.
[0041] Step S504: Using the initialization expert agent, analyze the correlation between any two isolated regulatory reporting relationship tables according to the regulatory reporting specification document, obtain the table analysis results, and determine the two isolated regulatory reporting relationship tables corresponding to the table analysis results that meet the preset second condition as the target association tables from all the table analysis results.
[0042] Step S505: Obtain the correlation analysis results for all target field pairs and target relationship tables.
[0043] In this embodiment, the field analysis result can refer to the numerical or grade index used to quantify the potential correlation strength between any two fields in the target regulatory reporting relationship table, analyzing the correlation between them. The preset first condition can refer to a threshold standard set in advance to filter out field combinations with significant correlation. The candidate field pair can refer to the field combination whose field analysis result meets the preset first condition. The semantic verification result can refer to the result of the expert agent verifying the semantic correlation compliance of the candidate field pair according to regulatory norms. The target field pair can refer to the field combination that meets the preset first condition and has a strong correlation in business logic after semantic verification. The isolated regulatory reporting relationship table can be... This refers to regulatory reporting relationship tables that are not included in the target field pair filtering, meaning they lack strong field associations with other tables and require separate table-level correlation analysis. The table analysis result can refer to the correlation analysis between any two isolated regulatory reporting relationship tables, or a numerical or grade indicator used to quantify the potential correlation strength between two isolated regulatory reporting relationship tables. The preset second condition can refer to a pre-set criterion for determining whether two isolated regulatory reporting relationship tables have a valid business relationship, such as sharing key fields or belonging to the same business reporting process. The target relationship table can refer to a pair of isolated regulatory reporting relationship tables whose table analysis results meet the preset second condition and are confirmed to have a strong business relationship.
[0044] Specifically, firstly, the first intelligent agent is used to determine the fields and corresponding metadata (including Chinese names of fields, data types, whether they can be null, etc.) in each target regulatory reporting relationship table. Based on the metadata, the correlation between any two fields in all target regulatory reporting relationship tables is analyzed, and the two fields corresponding to the field analysis results that meet the preset first condition are determined as candidate field pairs (for example, if two fields in different tables have the same or highly similar Chinese names and compatible data types, they are recorded as a set of candidate field pairs). Then, for any candidate field pair, the first constructed agent automatically generates a query for that candidate field pair (e.g., please verify whether 'IOU number' is used as an identifier connecting the 'Loan IOU' and the 'Guarantee Agreement'). Using the initialization expert agent, based on this query, a search is performed in the regulatory reporting specification knowledge. If the returned document fragments clearly indicate their association semantics, then the candidate field pair is confirmed as the target field pair. Finally, based on all target field pairs, from all target regulatory reporting relationship tables, isolated regulatory reporting relationship tables that were not included in the target field pair screening, i.e., lack strong field associations with other tables and require separate table-level association analysis, are identified. Using the initialization expert agent, a broader exploratory query is generated for each isolated regulatory reporting relationship table (e.g., "Please explain which tables the 'Related Group Information' table is usually cross-checked with"). By understanding the business logic in the documents, two isolated regulatory reporting relationship tables with effective business associations are identified as target association tables. All target field pairs and target association tables are the results of the correlation analysis.
[0045] In this embodiment, the above three-stage process collects the association information and related fields between data tables, providing richer data for subsequent predicate construction and rule discovery. This helps to discover rules between multiple tables and, based on expert intelligence, automatically and deeply utilizes the definitions of fields and business logic in unstructured regulatory documents to assist the correlation analysis process. This makes the correlation analysis results closer to actual business needs and regulatory requirements, improving the accuracy and effectiveness of field analysis.
[0046] In one embodiment, such as Figure 6 As shown, a data verification method based on an intelligent agent is provided. In step S302 above, a second constructing intelligent agent and an initializing expert intelligent agent are used to construct predicates based on regulatory reporting documents and correlation analysis results, and to obtain the predicate construction results. The method includes the following steps: Step S601: Using the second building agent, predicates are constructed based on the metadata of all target relation tables, target field pairs, and fields in the corresponding target field pairs to obtain the basic predicates.
[0047] Step S602: Using the second building agent and the initialization expert agent, construct predicates based on the regulatory reporting specification document and the regulatory reporting relationship table of all targets to obtain the enhanced predicates.
[0048] Step S603: Construct the result of all basic predicates and enhanced predicates into predicates.
[0049] In this embodiment, the basic predicate can refer to the predicate constructed based on the target relation table, target field pairs and metadata, and the enhanced predicate can refer to the predicate containing business meaning constructed based on the regulatory reporting specification document and the target regulatory reporting relation table. In this application, the predicate is explicitly defined as an atomic, reusable Structured Query Language (SQL) code snippet.
[0050] Specifically, firstly, the second building agent is used to construct predicates based on all target relation tables, target field pairs, and metadata to obtain basic predicates. For example, based on verified relationships, relational predicates are automatically generated using SQL templates (e.g., "Loan Receipt.Receipt Number = Loan Extension Agreement.Receipt Number"). Based on the enumerated values of fields, a series of filtering predicates are automatically generated (e.g., "Loan Receipt.Loan Status = 'Overdue'"). Then, when business logic cannot be obtained solely from the data structure, the second building agent and the initialization expert agent are used to construct predicates based on regulatory reporting specification documents and all target regulatory reporting relation tables to obtain enhanced predicates. For example, by querying the regulatory specification documents, it is known that "physical teller" in the "Teller Table" corresponds to the "Employee Table," and a composite logical predicate containing an EXISTS subquery can be synthesized to determine the physical teller.
[0051] In this embodiment, by integrating data structure features and regulatory business semantics, a comprehensive predicate library containing basic operations and high-value business logic is constructed, laying a solid foundation for subsequent discovery of complex rules. This helps to discover rules between multiple tables, and based on expert intelligence, it automatically and deeply utilizes the definitions of business logic in unstructured regulatory documents to assist the predicate construction process. This makes the constructed predicates more closely aligned with actual business needs and regulatory requirements, improving the accuracy and effectiveness of predicate construction.
[0052] In one embodiment, such as Figure 7 As shown, a data verification method based on intelligent agents is provided. In step S303 above, a third intelligent agent is used to construct the intelligent agent and an expert intelligent agent is initialized. Rules are constructed based on the regulatory reporting specification document and the predicate construction results to obtain association rules, including the following steps: Step S701: Use the initialization expert agent to determine the business meaning of each predicate in the predicate construction result.
[0053] Step S702: Using a third construction agent, based on the business meaning, determine the conclusion predicate that satisfies the preset third condition from all predicates in the predicate construction results.
[0054] Step S703: For any conclusion predicate, obtain the preconditions corresponding to the conclusion predicate based on the target relation table and target field pair corresponding to the conclusion predicate, and construct the association rule based on the preconditions and the conclusion predicate.
[0055] In this embodiment, the preset third condition can refer to a pre-set conclusion that can serve as a rule, or a condition that the predicate needs to satisfy.
[0056] After constructing a comprehensive predicate library containing basic operations and high-value business logic, the third construction agent and the initialization expert agent combine the predicates to generate association rules. Each rule is formalized into a logical structure of "if X (condition) then Y (conclusion)". Since the predicate is a fragment of SQL code, the rule is exactly a complete SQL statement.
[0057] Specifically, first, an initial expert agent is used to query the business meaning of each predicate. A third-party construction agent is then used to determine whether the predicate is suitable as a conclusion predicate (the conclusion of a rule) based on its business meaning. For example, it might represent a risk event or a compliance requirement, thus obtaining conclusion predicates suitable as rule conclusions. Then, for any conclusion predicate, a breadth-first search is used to find its conditional predicates. Before the search, the search space can be pruned based on the target relation table and target fields associated with the conclusion predicate, retaining only relevant predicates. Subsequently, the number of combined predicates is increased layer by layer, systematically exploring all conditional combinations from simple to complex to obtain the corresponding association rules.
[0058] In this embodiment, by intelligently combining predicates with clear business meanings into a "condition-conclusion" logical structure, semantically clear, executable, and regulatory-intent-aligned association rules are constructed. Furthermore, based on expert intelligence agents, the definition of business logic in unstructured regulatory documents is automatically and deeply utilized to assist the rule construction process, enabling the constructed rules to be closer to actual business needs and regulatory requirements, thereby improving the accuracy and effectiveness of rule construction.
[0059] In one embodiment, such as Figure 8 As shown, a data verification method based on an intelligent agent is provided. The construction of the intelligent agent also includes a fourth intelligent agent for rule validity verification. In step S703 above: for any conclusion predicate, based on the target relation table and target field pair corresponding to the conclusion predicate, the preconditions corresponding to the conclusion predicate are obtained. Based on the preconditions and the conclusion predicate, an association rule is constructed, including the following steps: Step S801: For any conclusion predicate, use the third-party construction agent to obtain the condition predicate corresponding to the conclusion predicate based on the target relation table and target field pair corresponding to the conclusion predicate.
[0060] Step S802: Using the fourth construction agent and the initialization expert agent, the validity of each conditional predicate is verified according to the regulatory reporting specification document, and the validity verification result is obtained.
[0061] Step S802: Using the third intelligent agent, based on the validity verification results, determine the preconditions corresponding to the conclusion predicates from all conditional predicates, and construct association rules based on the preconditions and conclusion predicates.
[0062] In this embodiment, the fourth constructing agent can refer to an agent used to verify the validity of business logic for the rule preconditions.
[0063] Specifically, for any conclusion predicate, a third construction agent is used to find the corresponding conditional predicate through breadth-first search. For any conditional predicate, a fourth construction agent is used to generate a validity query statement regarding the conditional predicate and the conclusion predicate (e.g., verify whether "loan status is overdue" and "should be recorded as non-performing loan" have a logical business relationship). The expert agent is initialized to search in the regulatory reporting norms based on the query statement. If the returned document fragments clearly indicate a business relationship, then the premise predicate is confirmed as a premise condition of the conclusion predicate. The third construction agent is then used to construct the association rule based on the conclusion predicate and its premise condition. If the returned document fragments clearly indicate no business relationship, then the invalid search path is pruned immediately to terminate it in advance.
[0064] Optionally, pruning can also be performed during the search process based on data support: if the coverage of the current condition combination X in the real data is lower than a preset threshold (e.g., 1%), pruning is performed immediately because subsequent expansion will result in even lower coverage. This judgment comes from an internally built database system, with data from several regulatory reporting relationship tables input by users during data preparation. Rules are executed in the database system in the form of SQL queries, thereby outputting the rule coverage.
[0065] Optionally, for the complete association rule formed after pruning, the system can also perform a final validity verification, which must simultaneously meet the following conditions: 1) Data-level confidence verification: The confidence level of the rule on real data must be higher than a preset threshold (e.g., 90%). This judgment comes from the internally established database system, and the operation is the same as described above. 2) Business-level scenario compliance verification: The overall business logic of the rule must not contradict the description in the regulatory reporting specification document. This judgment comes from the cooperation of the initialization expert agent and the fourth construction agent, and the operation is the same as described above.
[0066] In this embodiment, by introducing a fourth intelligent agent and a multi-dimensional pruning verification mechanism, dual filtering of business logic validity and data statistical significance is performed simultaneously during the rule generation process, which improves the efficiency of rule discovery. Furthermore, the use of an expert intelligent agent ensures that each precondition complies with regulatory standards, guaranteeing the business rationality of the rules from the source. Through quantitative evaluation of data support and confidence, redundant or accidental rules with insufficient statistical significance are effectively screened out, improving the accuracy and effectiveness of rule construction. This enables efficient and reliable automated data verification in complex business scenarios.
[0067] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.
[0068] In one embodiment, an agent-based data verification device is provided. This device is applied to a data verification platform, which includes a management agent, an expert agent, a construction agent, and a verification agent. This agent-based data verification device corresponds one-to-one with the agent-based data verification method described in the above embodiments. Figure 9 As shown, the agent-based data verification device includes a data acquisition module 91, a distribution module 92, a rule construction module 93, and a data verification module 94. Detailed descriptions of each functional module are as follows: The data acquisition module 91 is used to acquire, based on the management intelligent agent, the regulatory reporting relationship table to be verified and the regulatory reporting specification document, and select at least one target regulatory reporting relationship table from all the regulatory reporting relationship tables to be verified. The distribution module 92 is used to send all target regulatory reporting relationship tables to the construction agent, send all regulatory reporting relationship tables to be verified to the verification agent, and initialize the expert agent according to the regulatory reporting specification document to obtain the initialized expert agent. Rule building module 93 is used to build rules based on the table of all target regulatory reporting relationships and the regulatory reporting specification document using the building agent and the initialization expert agent, obtain associated rules, and send the associated rules to the verification agent; The data verification module 94 is used to use the verification agent to verify all regulatory reporting relationship tables to be verified according to the association rules, and obtain the verification results.
[0069] Optionally, the above-mentioned distribution module 92 includes: The segmentation unit is used to determine the chapters and clauses in the regulatory reporting specification document, and to segment the regulatory reporting specification document into blocks according to the chapters and clauses to obtain text blocks; The index building unit is used to construct a vector index corresponding to the regulatory reporting specification document based on each text block and the chapter and clause to which the corresponding text block belongs, using a preset embedding model; An initialization unit is used to initialize the expert agent according to the vector index to obtain the initialized expert agent.
[0070] Optionally, the rule construction module 93 mentioned above includes: The field analysis unit is used to analyze the correlation between all target regulatory reporting relationship tables and between fields in the target regulatory reporting relationship tables, based on the first constructed intelligent agent and the initialized expert intelligent agent, according to the regulatory reporting specification document, and obtain the correlation analysis results; The predicate construction unit is used to construct predicates based on the regulatory reporting specification document and the correlation analysis results using the second construction agent and the initialization expert agent, and to obtain the predicate construction result; The rule building unit is used to build rules based on the regulatory reporting specification document and the predicate construction results using the third building agent and the initialization expert agent to obtain the association rules.
[0071] Optionally, the above field analysis unit includes: The field pair determination subunit is used to determine the fields and corresponding metadata of each target regulatory reporting relationship table using the first constructed intelligent agent, analyze the correlation between any two fields in all target regulatory reporting relationship tables based on the metadata, obtain field analysis results, and determine the two fields corresponding to the field analysis results that meet the preset first condition as candidate field pairs from all field analysis results. The semantic verification subunit is used to use the initialization expert agent to perform semantic verification on each candidate field pair according to the regulatory reporting specification document, obtain the semantic verification result, and determine the target field pair from all candidate field pairs based on the semantic verification result. The table filtering subunit is used to filter out the isolated regulatory reporting relationship table from all target regulatory reporting relationship tables based on the target field pairs using the first constructed intelligent agent. The association table determination subunit is used to use the initialization expert agent to analyze the correlation between any two isolated regulatory reporting relationship tables according to the regulatory reporting specification document, obtain the table analysis results, and determine the two isolated regulatory reporting relationship tables corresponding to the table analysis results that meet the preset second condition as the target association tables from all the table analysis results. The first sub-unit is used to generate the correlation analysis results for all target field pairs and target relationship tables.
[0072] Optionally, the above predicate construction unit includes: The basic construction subunit is used to construct predicates based on the metadata of all target relation tables, target field pairs and corresponding fields in the target field pairs using the second construction agent, so as to obtain basic predicates; An enhanced construction subunit is used to construct predicates based on the regulatory reporting specification document and the table of all target regulatory reporting relationships, using the second construction agent and the initialization expert agent, to obtain enhanced predicates; The second forming subunit is used to construct the result of all basic predicates and enhanced predicates.
[0073] Optionally, the above rule construction unit includes: The meaning determination subunit is used to determine the business meaning of each predicate in the predicate construction result using the initialization expert agent; The conclusion determination subunit is used to use the third construction agent to determine the conclusion predicate that satisfies the preset third condition from all predicates in the predicate construction result according to the business meaning; The condition determination subunit is used to obtain the preconditions corresponding to any conclusion predicate based on the target relation table and target field pair corresponding to the conclusion predicate, and to construct the association rule based on the preconditions and the conclusion predicate.
[0074] Optionally, the above conditions determine the sub-unit, including: The search subunit is used to, for any conclusion predicate, use the third constructing agent to obtain the condition predicate corresponding to the conclusion predicate based on the target relation table and target field pair corresponding to the conclusion predicate; The validity verification subunit is used to perform validity verification on each conditional predicate according to the regulatory reporting specification document using the fourth construction agent and the initialization expert agent, and obtain the validity verification result. The combined subunit is used to construct the association rule using the third agent, based on the validity verification result, to determine the preconditions corresponding to the conclusion predicate from all conditional predicates, and to construct the association rule based on the preconditions and the conclusion predicate.
[0075] For specific limitations regarding the agent-based data verification device, please refer to the limitations of the agent-based data verification method above, which will not be repeated here. Each module in the aforementioned agent-based data verification device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.
[0076] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 10 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores regulatory reporting relationship tables and regulatory reporting specification documents to be verified. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements an agent-based data verification method.
[0077] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the agent-based data verification method described in the above embodiments, for example... Figure 2 As shown in S201-S204, or Figures 3 to 8 As shown, to avoid repetition, it will not be described again here. Alternatively, when the processor executes a computer program, it implements the functions of each module / unit in this embodiment of the data management device, for example, Figure 9 The functions of the data acquisition module 91, distribution module 92, rule construction module 93, and data verification module 94 shown are not described again here to avoid duplication.
[0078] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When executed by a processor, the computer program implements the agent-based data verification method described in the above embodiments, for example... Figure 2 As shown in S201-S204, or Figures 3 to 8 As shown, to avoid repetition, it will not be described again here. Alternatively, when the processor executes a computer program, it implements the functions of each module / unit in this embodiment of the data management device, for example, Figure 9 The functions of the data acquisition module 91, distribution module 92, rule construction module 93, and data verification module 94 shown are not described again here to avoid duplication.
[0079] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0080] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.
[0081] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.
Claims
1. A data verification method based on intelligent agents, characterized in that, The data verification method is applied to a data verification platform, which includes a management agent, an expert agent, a construction agent, and a verification agent. The data verification method includes: Based on the management agent, obtain the regulatory reporting relationship table and regulatory reporting specification document to be verified, and select at least one target regulatory reporting relationship table from all the regulatory reporting relationship tables to be verified. Send all target regulatory reporting relationship tables to the construction agent, send all regulatory reporting relationship tables to be verified to the verification agent, and initialize the expert agent according to the regulatory reporting specification document to obtain the initialized expert agent; Using the construction agent and the initialization expert agent, rules are constructed based on the table of all target regulatory reporting relationships and the regulatory reporting specification document to obtain association rules, and the association rules are sent to the verification agent; Using the aforementioned verification agent, the verification results are obtained by verifying all regulatory reporting relationship tables to be verified according to the association rules.
2. The data verification method based on intelligent agents according to claim 1, characterized in that, The initialization of the expert agent based on the regulatory reporting specification document to obtain the initialized expert agent includes: Identify the chapters and clauses in the regulatory reporting specification document, and divide the regulatory reporting specification document into blocks according to the chapters and clauses to obtain text blocks; By using a pre-defined embedding model, a vector index corresponding to the regulatory reporting specification document is constructed based on each text block and the corresponding chapter and clause to which the text block belongs. The expert agent is initialized according to the vector index to obtain the initialized expert agent.
3. The data verification method based on intelligent agents according to claim 1, characterized in that, The construction agent includes a first construction agent for correlation analysis, a second construction agent for predicate construction, and a third construction agent for rule construction. Using the construction agents and the initialization expert agent, rules are constructed based on the table of all target regulatory reporting relationships and the regulatory reporting specification document to obtain association rules, including: Using the first constructed intelligent agent and the initialized expert intelligent agent, the correlation between all target regulatory reporting relationship tables and between fields in the target regulatory reporting relationship tables is analyzed according to the regulatory reporting specification document, and the correlation analysis results are obtained. Using the second construction agent and the initialization expert agent, predicates are constructed based on the regulatory reporting specification document and the correlation analysis results to obtain the predicate construction results; Using the third building agent and the initialization expert agent, rules are constructed based on the regulatory reporting specification document and the predicate construction results to obtain the association rules.
4. The data verification method based on intelligent agents according to claim 3, characterized in that, The first constructed intelligent agent and the initialized expert intelligent agent are used to analyze the correlation between all target regulatory reporting relationship tables and between fields in the target regulatory reporting relationship tables, according to the regulatory reporting specification document, to obtain the correlation analysis results, including: Using the first constructed intelligent agent, determine the fields and corresponding metadata of each target regulatory reporting relationship table. Based on the metadata, analyze the correlation between any two fields in all target regulatory reporting relationship tables to obtain field analysis results. From all field analysis results, determine the two fields corresponding to the field analysis results that meet the preset first condition as candidate field pairs. Using the initialization expert agent, based on the regulatory reporting specification document, perform association semantic verification on each candidate field pair to obtain semantic verification results, and determine the target field pair from all candidate field pairs based on the semantic verification results; Using the first constructed intelligent agent, based on the target field pair, the isolated regulatory reporting relationship table is obtained by filtering from all target regulatory reporting relationship tables; Using the initialization expert agent, the correlation between any two isolated regulatory reporting relationship tables is analyzed according to the regulatory reporting specification document to obtain the table analysis results. From all the table analysis results, the two isolated regulatory reporting relationship tables corresponding to the table analysis results that meet the preset second condition are determined as the target association tables. The correlation analysis results are obtained by setting all target field pairs and target relationship tables.
5. The data verification method based on intelligent agents according to claim 4, characterized in that, The second construction agent and the initialization expert agent are used to construct predicates based on the regulatory reporting specification document and the correlation analysis results, resulting in predicate construction results, including: Using the second intelligent agent, predicates are constructed based on the metadata of all target relation tables, target field pairs, and fields in the corresponding target field pairs to obtain basic predicates; Using the second constructing agent and the initializing expert agent, predicates are constructed based on the regulatory reporting specification document and the table of all target regulatory reporting relationships to obtain enhanced predicates; Construct the result of all basic predicates and enhanced predicates into the predicates.
6. The data verification method based on intelligent agents according to claim 5, characterized in that, The third construction agent and the initialization expert agent are used to construct rules based on the regulatory reporting specification document and the predicate construction results to obtain the association rules, including: The initialization expert agent is used to determine the business meaning of each predicate in the predicate construction result; Using the third intelligent agent, based on the business meaning, determine the conclusion predicate that satisfies the preset third condition from all predicates in the predicate construction result; For any conclusion predicate, the preconditions corresponding to the conclusion predicate are obtained based on the target relation table and target field pair corresponding to the conclusion predicate. The association rule is then constructed based on the preconditions and the conclusion predicate.
7. The data verification method based on intelligent agents according to claim 6, characterized in that, The construction agent further includes a fourth construction agent for rule validity verification. For any conclusion predicate, based on the target relation table and target field pair corresponding to the conclusion predicate, the preconditions corresponding to the conclusion predicate are obtained. Based on the preconditions and the conclusion predicate, the association rule is constructed, including: For any conclusion predicate, the third constructing agent is used to obtain the condition predicate corresponding to the conclusion predicate based on the target relation table and target field pair corresponding to the conclusion predicate; Using the fourth construction agent and the initialization expert agent, the validity of each conditional predicate is verified according to the regulatory reporting specification document, and the validity verification result is obtained. Using the third intelligent agent, based on the validity verification result, the preconditions corresponding to the conclusion predicate are determined from all conditional predicates, and the association rule is constructed based on the preconditions and the conclusion predicate.
8. A data verification device based on an intelligent agent, characterized in that, The data verification device is applied to a data verification platform, which includes a management agent, an expert agent, a construction agent, and a verification agent. The data verification device includes: The data acquisition module is used to acquire, based on the management intelligent agent, the regulatory reporting relationship table to be verified and the regulatory reporting specification document, and select at least one target regulatory reporting relationship table from all the regulatory reporting relationship tables to be verified. The distribution module is used to send all target regulatory reporting relationship tables to the construction agent, send all regulatory reporting relationship tables to be verified to the verification agent, and initialize the expert agent according to the regulatory reporting specification document to obtain the initialized expert agent. The rule building module is used to build rules based on the table of all target regulatory reporting relationships and the regulatory reporting specification document using the building agent and the initialization expert agent, obtain associated rules, and send the associated rules to the verification agent; The data verification module is used to verify all regulatory reporting relationship tables to be verified according to the association rules, using the verification agent, and to obtain the verification results.
9. A computer device, characterized in that, The computer device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the agent-based data verification method as described in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the agent-based data verification method as described in any one of claims 1 to 7.