Supervision system knowledge graph construction method and device based on multistage entity model

Abstract

The invention discloses a supervision system knowledge graph construction method and device based on a multi-level entity model, and relates to the field of financial supervision science and technology, and the method comprises the steps: carrying out the modeling analysis of supervision submission system file data, so as to design a knowledge graph model containing multi-dimensional entities and relation attributes; triad data is extracted from the supervision submission system file data based on a knowledge graph model and stored in a data lake warehouse, the triad data comprises entity nodes, entity attributes and relationships among entities, and the relationships among the entities comprise an affiliation relationship and a co-affiliation relationship; and through a knowledge mapping module, converting the triple data in the data lake warehouse into node data and edge data of a graph database, reasoning a cross-institution entity association relationship through a semantic rule, importing the node data, the edge data and the cross-institution entity association relationship into the graph database, and generating a visual knowledge graph network. The data analysis efficiency can be improved, and the data consistency requirement under strict supervision can be met.

Description

Technical Field

[0001] This application relates to the field of financial regulatory technology, and in particular to a method and apparatus for constructing a regulatory system knowledge graph based on a multi-level entity model. Background Technology

[0002] With the continuous development of knowledge graph technology, its application in information retrieval and analysis across various scenarios has become increasingly widespread, providing new ideas and methods for information integration and utilization. However, in the field of financial regulatory technology, the application of knowledge graph structures to the characteristic data of regulatory reporting systems remains significantly insufficient.

[0003] Currently, the main method for analyzing regulatory reporting system characteristic data is through offline review and analysis of regulatory reporting system documents. However, this offline document management model results in the fragmented storage of regulatory reporting system documents and significant obstacles to cross-system correlation analysis, severely impacting the efficiency of data analysts and failing to meet the demands for data integration and efficient analysis under increasingly stringent regulatory requirements. Summary of the Invention

[0004] In view of this, this application provides a method and apparatus for constructing a regulatory system knowledge graph based on a multi-level entity model, which can realize the structured storage and cross-system correlation analysis of regulatory reporting system document data, improve data analysis efficiency and meet the data consistency requirements under strict supervision.

[0005] According to the first aspect of this application, a method for constructing a regulatory system knowledge graph based on a multi-level entity model is provided, including: Modeling and analysis of regulatory reporting system documents are conducted to design a knowledge graph model that includes multi-dimensional entity and relational attributes; Based on the knowledge graph model, triple data is extracted from the regulatory reporting system document data and stored in a data lake warehouse. The triple data includes entity nodes, entity attributes and relationships between entities. The relationships between entities include attribution relationships and same-family relationships. The knowledge mapping module converts the triple data in the data lake warehouse into node data and edge data in the graph database. It then uses semantic rules to infer cross-institutional entity relationships and imports the node data, edge data, and cross-institutional entity relationships into the graph database to generate a visualized knowledge graph network.

[0006] According to a second aspect of this application, a device for constructing a regulatory system knowledge graph based on a multi-level entity model is provided, comprising: The analysis module is used to model and analyze regulatory reporting system documents to design a knowledge graph model that includes multi-dimensional entity and relational attributes. The extraction module is used to extract triple data from the regulatory reporting system document data based on the knowledge graph model and store it in the data lake warehouse. The triple data includes entity nodes, entity attributes and relationships between entities. The relationships between entities include attribution relationships and same-family relationships. The generation module is used to convert the triple data in the data lake warehouse into node data and edge data of the graph database through the knowledge mapping module, and to infer cross-institutional entity relationships through semantic rules, and import the node data, edge data and cross-institutional entity relationships into the graph database to generate a visualized knowledge graph network.

[0007] According to a third aspect of this application, a storage medium is provided on which a computer program is stored, which, when executed by a processor, implements the above-described method for constructing a regulatory knowledge graph based on a multi-level entity model.

[0008] According to a fourth aspect of this application, an electronic device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described method for constructing a regulatory knowledge graph based on a multi-level entity model.

[0009] By employing the aforementioned technical solutions, the regulatory system knowledge graph construction method and apparatus based on a multi-level entity model provided in this application, through modeling and analysis of regulatory reporting system document data, designs a knowledge graph model containing multi-dimensional entity and relational attributes. Based on the knowledge graph model, it extracts triple data containing entity nodes, attributes, and attribution and affiliation relationships from the regulatory reporting system document data and stores it in a data lake warehouse. Then, through a knowledge mapping module, it converts this into node and edge data of a graph database. Simultaneously, it infers cross-system entity association relationships and imports them into the graph database to generate a visualized knowledge graph network. This application integrates scattered offline regulatory reporting system document data into structured triple data through structured modeling and storage of knowledge graphs. It utilizes the association expression capabilities of graph databases to achieve semantic association reasoning between cross-system entities. The visualized knowledge graph network not only solves the problem of scattered data storage but also replaces traditional manual offline retrieval through automated association analysis, significantly improving data analysis efficiency and meeting the needs for data integration and efficient analysis under increasingly stringent regulatory requirements.

[0010] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0011] Figure 1The illustration shows a flowchart of a method for constructing a regulatory system knowledge graph based on a multi-level entity model, as provided in an embodiment of this application. Figure 2 A flowchart illustrating a method for constructing a regulatory system knowledge graph based on a multi-level entity model, according to another embodiment of this application, is shown. Figure 3 This illustration shows a schematic diagram of a regulatory system knowledge graph construction device based on a multi-level entity model, as provided in an embodiment of this application. Figure 4 This paper illustrates a schematic diagram of another regulatory system knowledge graph construction device based on a multi-level entity model provided in an embodiment of this application. Detailed Implementation

[0012] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0013] Currently, the main method for analyzing regulatory reporting system characteristic data is through offline review and analysis of regulatory reporting system documents. However, this offline document management model results in the fragmented storage of regulatory reporting system documents and significant obstacles to cross-system correlation analysis, severely impacting the efficiency of data analysts and failing to meet the demands for data integration and efficient analysis under increasingly stringent regulatory requirements.

[0014] To address the aforementioned technical problems, embodiments of the present invention provide a method for constructing a regulatory system knowledge graph based on a multi-level entity model, such as... Figure 1 As shown, the method includes: Step 110: Model and analyze the regulatory reporting system documents to design a knowledge graph model that includes multi-dimensional entity and relational attributes.

[0015] Among them, regulatory reporting system documents data refers to institutional data related to regulatory reporting, usually various system documents, data specifications, and other information required by regulatory agencies; modeling analysis refers to analyzing and abstracting data, establishing corresponding models, and clarifying the structure, relationships, and attributes of the data; multi-dimensional entities refer to entities defined from multiple perspectives and levels, such as regulatory bodies at different levels, reporting systems, data tables, etc.; relationship attributes refer to the associations between entities and the attributes that these relationships have, such as the hierarchical relationship between entities, the direction of association, and other characteristics; knowledge graph models are models that use graph structures to represent knowledge. Through the combination of entities, relationships, and attributes, a knowledge network model with semantic associations is constructed. It is an abstract design framework for defining entity types, relationship rules, and attribute dimensions.

[0016] In this embodiment of the disclosure, by modeling and analyzing data related to the regulatory reporting system, a knowledge graph model can be designed. This model includes entities with multiple dimensions and relational attributes between entities. Specifically, the data in the regulatory reporting system is broken down into entities at different levels (such as regulatory agencies, reporting systems, interface tables, fields, etc.) according to certain rules, the relationships between these entities are defined (such as issuance, inclusion, etc.), and each entity is assigned corresponding attributes (such as name, business scope, etc.), ultimately forming a structured knowledge graph model.

[0017] This modeling and analysis allows regulatory reporting documents to be organized in a structured way, transforming previously scattered data into a logically connected whole. This facilitates subsequent data extraction, storage, and analysis, laying the foundation for building a visualized knowledge graph network and improving data management and usage efficiency.

[0018] Step 120: Extract triple data from regulatory reporting system document data based on knowledge graph model and store it in data lake warehouse. Triple data includes entity nodes, entity attributes and relationships between entities. Relationships between entities include attribution relationship and same-family relationship.

[0019] In this embodiment of the disclosure, the knowledge graph model has predefined entity types, relationship rules, and attribute dimensions (such as entity types like regulatory agencies and reporting systems, affiliation / compatibility relationship types, and entity attribute fields). When processing regulatory reporting system document data, the knowledge graph model acts as a "semantic template," guiding the system to identify entity nodes that conform to the preset types from unstructured / semi-structured data, and extracting the relationships between entities and corresponding entity attributes according to the rules. Thus, based on the designed knowledge graph model, triple data consisting of "entity nodes, entity attributes, and relationships between entities" can be extracted from the regulatory reporting system document data and stored in a data lake warehouse.

[0020] Among them, entity nodes are key objects in the data, representing specific things in the regulatory reporting system (such as regulatory agencies and reporting systems); entity attributes are the characteristics or properties of these objects (such as entity name, business scope, reporting frequency, etc.); the relationships between entities include two types: affiliation relationship (reflecting vertical association logic, such as hierarchical affiliation) and same affiliation relationship (parallel association between entities of the same category or level, reflecting horizontal association logic, such as different systems belonging to the same regulatory agency); the data lake warehouse is a centralized warehouse used to store massive amounts of raw data, supporting the unified management and storage of structured and unstructured data.

[0021] By extracting and storing triplet data, unstructured regulatory data is transformed into structured knowledge units, facilitating subsequent data integration, querying, and semantic analysis. The data lake warehouse storage method enables centralized management of massive amounts of regulatory data, providing a standardized data foundation for building knowledge graph networks and cross-regulatory correlation analysis.

[0022] Step 130: Through the knowledge mapping module, the triple data in the data lake warehouse is converted into node data and edge data in the graph database. Then, the cross-institutional entity relationship is inferred through semantic rules. The node data, edge data and cross-institutional entity relationship are imported into the graph database to generate a visualized knowledge graph network.

[0023] The knowledge mapping module is a functional module used to convert structured data (such as triple data) into a graph database format according to rules, and is used to define the mapping rules for data transformation. A graph database is a database that stores data in a "node-edge" graph structure, suitable for handling complex relationships, such as Neo4j, and is better at storing and querying semantic networks than traditional databases. Node data is the basic unit in the graph database, representing entities in the knowledge graph (such as regulatory agencies, reporting systems), and each node contains entity attributes (such as name, type). Edge data is the directed line segment connecting nodes, representing the relationship between entities (such as issuing, containing), and the edge contains the relationship type and pointing rule (such as pointing from a regulatory agency to a reporting system). Semantic rules are rules used to infer semantic relationships between entities (such as name similarity matching, attribute consistency judgment), used to identify synonymous or homologous entities in cross-system data. Cross-system entity relationships are the relationships between entities in different regulatory reporting systems (such as indicators with the same name in different systems), generated through semantic rule reasoning, breaking down data silos between systems. The knowledge graph network is a visual graph structure instance composed of nodes, edges, and attributes, which is an intuitive representation of regulatory reporting system knowledge.

[0024] In this embodiment of the disclosure, the "entity node-entity attribute-entity relationship" triple data stored in the data lake warehouse can be converted into nodes (representing entities) and edges (representing relationships) that can be recognized by the graph database through the knowledge mapping module. At the same time, semantic rules are used to reason about the associations between entities in different reporting systems (such as matching entities with the same name). Then, these nodes, edges, and cross-system associations are imported into the graph database, and finally a visualized knowledge graph network is generated. This process essentially transforms structured triple data into graph structure data and improves the semantic connections of the graph through association reasoning.

[0025] Transforming abstract triplet data into an intuitive graph structure enables the visualization of regulatory reporting system documents, facilitating users' intuitive understanding of the hierarchy and relationships between data. The graph database storage method supports efficient semantic querying and association analysis, improving the retrieval efficiency of cross-system data. By supplementing cross-system relationships through semantic rule reasoning, the obstacles to association analysis caused by traditional data dispersion can be solved, providing structured knowledge network support for regulatory compliance analysis.

[0026] In summary, the regulatory system knowledge graph construction method based on a multi-level entity model provided by this invention, through modeling and analysis of regulatory reporting system document data, designs a knowledge graph model containing multi-dimensional entities and relational attributes. Based on the knowledge graph model, it extracts triple data containing entity nodes, attributes, and affiliation and related relationships from the regulatory reporting system document data and stores it in a data lake warehouse. Then, through a knowledge mapping module, it converts this data into node and edge data of a graph database. Simultaneously, it infers cross-system entity relationships and imports them into the graph database to generate a visualized knowledge graph network. This application integrates scattered offline regulatory reporting system document data into structured triple data through structured modeling and storage of knowledge graphs. It utilizes the relational expression capabilities of graph databases to achieve semantic relational reasoning between cross-system entities. The visualized knowledge graph network not only solves the problem of scattered data storage but also replaces traditional manual offline retrieval with automated relational analysis, significantly improving data analysis efficiency and meeting the needs for data integration and efficient analysis under increasingly stringent regulatory requirements.

[0027] Furthermore, as a refinement and extension of the specific implementation methods of the above embodiments, and to fully illustrate the implementation methods of this embodiment, this embodiment also provides another method for constructing a regulatory system knowledge graph based on a multi-level entity model, such as... Figure 2 As shown, the method includes: Step 210: Model and analyze the regulatory reporting system document data to design a knowledge graph model that includes multi-dimensional entity and relational attributes.

[0028] In this embodiment of the disclosure, when modeling and analyzing regulatory reporting system document data, the data can first be divided into detailed level (with fields as the smallest unit) and summary report level (including indicator system) according to the collection specifications. Then, the entity hierarchy is split according to rules: the main entity is the regulatory agency and the competent authority, and the child nodes include reporting system, interface table, and fields / indicators. The summary report level indicators are further split into second-level / third-level indicators. A globally unique UID (such as "PBOC_Reg_001") is generated using the entity's Chinese name as the primary key, and entity attributes (such as "reporting frequency" in the interface table) are associated. Attribution relationships (such as issued system) and peer relationships (such as system association with the same institution) are defined, ultimately forming a knowledge graph model containing entities, attributes, and relationships.

[0029] Accordingly, the specific implementation steps may include: dividing the regulatory reporting system document data into detailed-level reporting systems and summary report-level reporting systems according to the data collection specifications; splitting the detailed-level reporting systems and summary report-level reporting systems into entity hierarchies according to entity splitting rules to obtain entity types and multi-level entity hierarchies. The entity splitting rules are as follows: the main entity nodes are regulatory agencies and competent responsible departments, and the child nodes include reporting systems, interface tables, and fields / indicators. Among them, the indicator sub-nodes in the summary report-level reporting systems are split into secondary and tertiary indicators; generating a globally unique UID as the primary key using the entity's Chinese name as the matching item, and associating the entity attributes of each entity node in the multi-level entity hierarchy with the primary key; defining the types and relationship pointing rules of affiliation and related relationships for multiple entity nodes in the multi-level entity hierarchy; and generating a knowledge graph model containing multi-dimensional entity and relationship attributes based on each entity node, entity attribute, and relationship pointing rules in the multi-level entity hierarchy.

[0030] Among them, the detailed reporting system is based on data with fields as the smallest unit (such as "customer name" and "loan balance"), which is the lowest level of reporting data; the summary report-level reporting system refers to the indicator system formed by summarizing detailed data (such as "capital adequacy ratio" and "non-performing loan ratio"), which usually contains multi-level calculation relationships; entity-level splitting refers to the process of splitting data into different levels of entities according to business logic (such as regulatory agencies → systems → tables → fields).

[0031] By dividing regulatory reporting system documents into categories according to standards and splitting them into entity hierarchies, generating globally unique UIDs associated with entity attributes, defining relationship types and pointing rules, the resulting knowledge graph model can achieve the structuring and standardization of regulatory data. This solves the problems of data dispersion and difficulty in correlation analysis under the traditional offline management model, enabling scattered regulatory system data to form a logically connected whole through multi-level entity hierarchies and semantic relationships. At the same time, the globally unique UID can ensure the unique identification of entities, laying the foundation for cross-system data integration and efficient retrieval.

[0032] Step 220: Based on the entity types and multi-level entity hierarchy defined in the knowledge graph model, extract entity nodes from the regulatory reporting system document data and generate globally unique UIDs, while associating them with the entity attributes preset by the knowledge graph model.

[0033] In this embodiment of the disclosure, entity nodes can be extracted from regulatory reporting system document data based on predefined entity types (such as regulatory agencies, reporting systems, interface tables, fields / indicators, etc.) and multi-level entity hierarchies (regulatory agencies → reporting systems → interface tables → fields / indicators → secondary indicators → tertiary indicators) in the knowledge graph model. The specific process is as follows: OCR technology is used to parse unstructured documents such as PDF / Word documents, converting the text content into editable data. Then, entity nodes are extracted by matching the entity types defined in the knowledge graph model using pre-set regular expression templates (such as matching rules containing keywords like "table," "indicator," and "regulatory agency"). Next, according to the model's multi-level entity hierarchy rules, the extracted entities are categorized into corresponding levels. For example, the "non-performing loan ratio" in the summary report is classified as a primary indicator, and its calculation elements "subprime loan balance" and "doubtful loan balance" are classified as secondary and tertiary indicators, respectively. Then, based on the entity's Chinese name, a globally unique UID (such as "CBRC_Indicator_001") is generated according to the model encoding rules. Finally, the entity attributes preset by the model are extracted from the document (such as "reporting frequency" in the interface table and "business scope" in the field), and the association is established with the entity node through the UID.

[0034] Accordingly, the specific steps of the implementation plan may include: using OCR technology combined with regular expression templates to extract entity nodes from the regulatory reporting system document data according to the entity types defined by the knowledge graph model; classifying the extracted entity nodes hierarchically according to the multi-level entity hierarchy of the knowledge graph model, wherein the indicator entity nodes in the summary report-level reporting system are divided into first-level indicators, second-level indicators, and third-level indicators; generating a globally unique UID for each entity node according to the coding rules of the knowledge graph model, using the Chinese name of the entity as the matching item; extracting the entity attributes preset by the knowledge graph model from the regulatory reporting system document data, and associating the entity attributes with the globally unique UID.

[0035] In the steps of this embodiment, the structured extraction of regulatory data can be achieved through automation technology, which can significantly improve the efficiency and accuracy of entity recognition. Specifically, OCR technology can solve the problem of text recognition in unstructured documents, regular expression templates can ensure the standardization of entity extraction, multi-level hierarchical division can avoid entity classification confusion, globally unique UID can eliminate the risk of confusion between entities with the same name, and attribute association can achieve standardized data storage.

[0036] Step 230: Construct the association relationships between entity nodes according to the attribution relationship, same-attribution relationship and relationship pointing rules defined in the knowledge graph model.

[0037] In this embodiment of the disclosure, directed semantic connections can be established between extracted entity nodes according to the hierarchical relationships (such as issuing systems, included fields, etc.), same-system relationships (parallel associations between systems of the same organization), and relationship pointing rules (such as the issuing relationship pointing from the regulatory agency to the reporting system) predefined in the knowledge graph model. For example, the "regulatory agency" entity and the "reporting system" entity can be connected through the "issuing" relationship edge, and the "interface table" entity and the "field / indicator" entity can be connected through the "include" relationship edge, forming an association network that conforms to the regulatory business logic.

[0038] By constructing affiliation and shared relationships between entities, scattered regulatory data entities can be transformed into a directed and interconnected knowledge network. This makes the hierarchical logic and cross-institutional relationships of regulatory systems explicit, supporting automated semantic queries and cross-institutional analysis.

[0039] Step 240: Generate triplet data consisting of entity nodes, entity attributes, and relationships. After validating the triplet data, store it in the data lake warehouse.

[0040] In this embodiment of the disclosure, the extracted entity nodes (such as regulatory agencies and reporting systems), entity attributes (such as system names and reporting requirements), and inter-entity relationships (such as issued systems and included fields) can be combined into triplet data of "entity-attribute-relationship". After performing integrity checks (such as checking whether the entity is missing a UID or whether the relationship points to the correct one) and consistency checks (such as whether there are conflicts between cross-system entity attributes), the triplet data is stored in a data lake warehouse to form a standardized regulatory data set.

[0041] By generating triplet data consisting of entity nodes, entity attributes, and relationships, and then storing it in a data lake warehouse after verification, the originally scattered and unstructured regulatory reporting documents can be transformed into standardized and structured data units. The verification mechanism ensures data integrity, consistency, and accuracy, eliminating data errors and conflicts. The data lake warehouse enables unified storage and management of massive amounts of data, providing a reliable data foundation for subsequent knowledge graph construction, in-depth cross-regulatory data correlation analysis, and visualization, significantly improving the utilization efficiency and management level of regulatory data, and effectively meeting increasingly stringent regulatory requirements.

[0042] Step 250: Through the knowledge mapping module, the triple data in the data lake warehouse is converted into node data and edge data in the graph database. Then, through semantic rule reasoning, cross-institutional entity relationships are inferred, and the node data, edge data, and cross-institutional entity relationships are imported into the graph database to generate a visualized knowledge graph network.

[0043] For the embodiments of this disclosure, step 250 may specifically include the following steps: Step 250-1: Using the mapping rules defined in the knowledge mapping module, convert the entity nodes in the triplet data into node data in the graph database, and convert the entity attributes into the attribute fields of the nodes.

[0044] In this embodiment of the disclosure, the triple data in the data lake warehouse can be formatted using the pre-defined conversion rules of the knowledge mapping module. Specifically, the entity nodes in the triples can be converted into node data recognizable by the graph database, giving them a unique identifier in the graph structure; at the same time, the entity attributes are converted into attribute fields of the corresponding nodes and stored in the nodes in key-value pairs, so that the abstract triple data is adapted to the storage and query format of the graph database.

[0045] Through the transformation of the knowledge mapping module, seamless integration between triplet data and graph databases can be achieved, resolving data format incompatibility issues. Unified data transformation rules ensure standardized storage of entity nodes and attributes, facilitating efficient subsequent retrieval and management. The structured storage of node data and attribute fields makes the display and analysis of regulatory data more intuitive; for example, users can quickly obtain all attribute information of a regulatory agency, significantly improving data query efficiency and utilization value, and providing fundamental support for the construction of knowledge graph networks.

[0046] Step 250-2: Based on the attribution relationship, kinship relationship and relationship pointing rule defined in the knowledge graph model, the association relationship between entity nodes in the triplet data is converted into edge data in the graph database. The relationship type corresponds to the edge label, and the relationship pointing rule corresponds to the edge direction, thus obtaining the association edge between entity nodes.

[0047] In this embodiment of the disclosure, the relationships between entity nodes in the triplet data can be transformed based on the hierarchical relationships (such as issuing regulations, containing fields), the same-affiliation relationships (related to regulations of the same institution), and the relationship pointing rules (the direction of the relationship, such as the issuing relationship pointing from the regulatory agency to the reporting system) defined in the knowledge graph model. Specifically, the relationship type (such as "issuing" or "containing") is mapped to the label of an edge in the graph database, and the relationship pointing rules are transformed into the direction of the edge, thereby constructing directed edges between nodes in the graph database. For example, the triplet relationship "Bank A - Issuing - Financial Basic Data Specification" is transformed into an edge in the graph database pointing from the "Bank A" node to the "Financial Basic Data Specification" node, with the edge label "Issuing," clearly reflecting the hierarchical relationship between the two.

[0048] By transforming the relationships in triplet data into edge data in a graph database, the complex relationship network between entities in regulatory reporting documents can be clearly presented, making the originally implicit hierarchy and relationships intuitive and visible. The edge structure of the graph database supports efficient relational queries and reasoning. For example, it can quickly retrieve all regulations and their related indicators issued by a regulatory agency, significantly improving query efficiency compared to traditional data storage methods. At the same time, standardized relation transformation rules ensure the accuracy and consistency of the relation network, providing reliable data structure support for applications such as knowledge graph-based deep analysis and compliance checks.

[0049] Step 250-3: Using preset semantic rules, perform semantic matching and association reasoning on entity nodes in different reporting systems to generate cross-system entity association relationships.

[0050] In this embodiment of the disclosure, pre-defined semantic rules (such as name similarity calculation and attribute consistency judgment) can be used to perform semantic-level matching and association deduction of entity nodes in different regulatory reporting systems: First, the similarity of the Chinese names of entities (such as "non-performing loan rate" and "non-performing loan ratio") is calculated using the edit distance algorithm. When the similarity exceeds the threshold, it is used as a candidate association; then, entity attributes (such as the calculation formula of the indicator and business scope) are matched. If the attribute matching degree meets the standard, it is determined to be a synonym entity; finally, the global unique UID is used to verify whether they are the same source entity, and finally, cross-system entity association relationship is generated (such as the association of the "capital adequacy ratio" indicator in different systems).

[0051] Accordingly, the specific steps of the implementation plan may include: calculating the first similarity of the Chinese names of entities in different reporting systems using an edit distance algorithm; when the first similarity is greater than or equal to a first preset threshold, it is determined as a candidate entity and a first cross-system association relationship is generated; for the candidate entities in the first cross-system association relationship, the entity attribute matching degree of the same entity in different reporting systems is calculated; when the entity attribute matching degree is greater than or equal to a second preset threshold, it is determined as a synonym entity and an association is established to obtain a second cross-system association relationship, which includes an entity attribute matching degree parameter; for the synonym entities confirmed in the second cross-system association relationship, cross-system matching is performed based on the globally unique UID of the entity node; if they have the same globally unique UID, they are finally determined as homologous entities and a third cross-system association relationship with a unique identifier is generated; for the indicator entities in the summary report-level reporting system, after completing the above matching of the entity's Chinese name, entity attributes, and UID, the business association between indicators in different reporting systems is inferred by matching the calculation rules and composition fields of secondary and tertiary indicators, and a fourth cross-system association relationship containing indicator level information is generated.

[0052] The first similarity is the numerical value of the similarity of the Chinese names of entities obtained based on the edit distance algorithm, with a value range of 0-100%; the first preset threshold is the judgment threshold for name similarity (e.g., 80%), and entities exceeding this value become candidate related objects; the entity attribute matching degree is the comparison of the consistency of attributes of the same entity in different systems (e.g., the matching degree of the "calculation formula" and "reporting frequency" of the indicator), with a value range of 0-100%; the second preset threshold is the judgment threshold for attribute matching (e.g., 85%), which is used to filter entities with similar names but large differences in attributes (e.g., if the attribute difference between "non-performing loan rate" and "non-performing asset rate" exceeds the threshold, they are excluded).

[0053] By employing multi-dimensional hierarchical verification mechanisms such as edit distance algorithms, attribute matching, and UID verification, accurate association across institutional entities can be achieved, effectively avoiding misassociations caused by similar names and significantly improving the accuracy of data association. Deep matching of the calculation rules and constituent fields of summary report-level indicators can uncover implicit business relationships between indicators in different systems, deepening the understanding of regulatory data. The entire automated process can replace manual comparison, significantly improving the efficiency of cross-institutional association construction, supporting the rapid processing of massive amounts of regulatory data, and providing strong support for the global integration and in-depth analysis of regulatory data.

[0054] Step 250-4: After standardizing the obtained node data, edge data, and cross-institutional entity relationships according to the import format requirements of the graph database, import them into the graph database to construct a visualized knowledge graph network of the regulatory reporting system.

[0055] In this embodiment of the disclosure, the graph database node data and edge data obtained in the previous processing, as well as the cross-institutional entity relationships generated through semantic matching, can be standardized according to the import format requirements supported by the graph database (such as Neo4j). This includes operations such as data format conversion, missing value completion, and unified relationship labels. After standardization, the data is imported into the graph database, and the system automatically constructs a visualized knowledge graph network based on the nodes, edges, and relationships. This graphically displays the entity hierarchy, relationship network, and cross-institutional relationships in the regulatory reporting system.

[0056] By standardizing and importing node data, edge data, and cross-institutional relationships into a graph database, a visualized knowledge graph network can be constructed, enabling the transformation of regulatory reporting documents from scattered storage to a structured and intuitive presentation. The efficient storage and query capabilities of the graph database significantly shorten the response time for complex correlation analyses (such as quickly retrieving all regulations and their related indicators issued by a regulatory agency). The visualized graph network makes the hierarchical relationships and cross-institutional connections of regulatory data readily apparent, reducing the difficulty for business personnel in understanding regulatory rules. Standardization ensures data consistency and compatibility, providing a solid data foundation for subsequent applications such as knowledge graph-based intelligent question answering and compliance checks, effectively improving the utilization efficiency and value of regulatory data.

[0057] In specific application scenarios, one possible implementation is to build a front-end display interface (such as a webpage or client) to provide users with an operation entry point. The front-end display interface is equipped with a fuzzy search interface for the knowledge graph network of regulatory reporting systems. After the user enters keywords (such as "non-performing loans") into the interface, the system uses an edit distance algorithm to calculate the similarity between the keywords and preset matching terms (including system names, entity names, attribute values, etc., such as "non-performing loan rate" and "non-performing loan balance"). When the similarity exceeds a third preset threshold (such as 70%), the search is automatically triggered, and detailed information of the entity matching the keywords is displayed on the visualization page, including the entity's level (such as "Loan Risk Classification Management Measures → Indicator Layer"), the type of association (such as issuance, inclusion), complete attribute values ​​(such as calculation caliber, reporting frequency), and the reference position of the entity in the original regulatory document (such as Article 3 of Chapter 5 of a certain system document).

[0058] Accordingly, the implementation steps may further include: constructing a front-end display interface, providing a fuzzy search interface for knowledge graph networks on the front-end display interface; receiving search keywords input by the user based on the fuzzy search interface, and calculating the second similarity between the search keywords and preset matching words using an edit distance algorithm; triggering a search when the second similarity is greater than or equal to a third preset threshold; and displaying the entity details of the search results on a visualization page, including the entity's level, relationship type, complete attribute values, and the source document citation information of the entity in the regulatory reporting system, wherein the preset matching words are any one of the following: the Chinese name of the main node entity, the Chinese name of the child node entity, or the entity attribute.

[0059] The second similarity is the similarity value between the search keywords and the preset matching words calculated based on the edit distance algorithm, with a value range of 0-100%; the third preset threshold is the similarity judgment threshold (e.g., 70%), keywords exceeding this value trigger the search, avoiding overly strict or lenient matching; the preset matching words are key information extracted from the knowledge graph in advance, including the name of the regulatory system, entity name, attribute value, etc., as the benchmark data for search matching; the source document citation information records the specific location of the entity in the original regulatory reporting system document (e.g., chapter, clause number), which is convenient for tracing and reference.

[0060] In summary, the technical solution of this application, through modeling and analysis of regulatory reporting system document data, designs a knowledge graph model containing multi-dimensional entity and relational attributes. Based on the knowledge graph model, it extracts and stores triple data containing entity nodes, attributes, and attribution and affiliation relationships. Then, it converts the data into node and edge data of the graph database through the knowledge mapping module. At the same time, it infers cross-system entity association relationships and imports them into the graph database to generate a visualized knowledge graph network. In this process, data is divided according to collection specifications and entity hierarchical splitting is performed. The resulting multi-level entity hierarchy and globally unique UID encoding enable structured data storage and unique identification, solving the data dispersion problem caused by offline file management. By defining attribution and affiliation relationships and relationship pointing rules, and combining semantic rules to reason about cross-institutional entity relationships, entities scattered across different institutions can establish semantic relationships through nodes and edges in the graph database, replacing the traditional manual offline cross-institutional relationship analysis method. The final generated visualized knowledge graph network can not only quickly locate entity details and relationships through fuzzy retrieval, but also realize business relationship reasoning based on indicator hierarchy information and calculation rules, significantly improving data analysis efficiency and meeting the needs for data integration and efficient analysis under increasingly stringent regulatory requirements.

[0061] Furthermore, as Figure 1 and Figure 2 The specific implementation of the method shown in this embodiment provides a device for constructing a regulatory system knowledge graph based on a multi-level entity model, such as... Figure 3 As shown, the device includes: an analysis module 31, an extraction module 32, and a generation module 33.

[0062] Analysis module 31 can be used to model and analyze regulatory reporting system document data to design a knowledge graph model containing multi-dimensional entity and relation attributes; Extraction module 32 can be used to extract triple data from regulatory reporting system document data based on knowledge graph model and store it in data lake warehouse. The triple data includes entity nodes, entity attributes and relationships between entities. The relationships between entities include attribution relationship and same-family relationship. The generation module 33 can be used to convert triple data in the data lake warehouse into node data and edge data in the graph database through the knowledge mapping module, and to import the node data, edge data and cross-institutional entity relationships into the graph database through semantic rule reasoning to generate a visualized knowledge graph network.

[0063] In some embodiments of this application, the analysis module 31 can be specifically used to divide regulatory reporting system document data into detailed-level reporting systems and summary report-level reporting systems according to the data collection specifications; to perform entity-level splitting on the detailed-level reporting systems and summary report-level reporting systems according to entity splitting rules, thereby obtaining entity types and multi-level entity hierarchies. The entity splitting rules are as follows: the main entity nodes are regulatory agencies and competent responsible departments, and the child nodes include reporting systems, interface tables, and fields / indicators. Among them, the indicator child nodes in the summary report-level reporting systems are split into secondary indicators and tertiary indicators; a globally unique UID is generated as the primary key using the entity's Chinese name as the matching item, and the entity attributes of each entity node in the multi-level entity hierarchy are associated with the primary key; for multiple entity nodes in the multi-level entity hierarchy, the types of attribution relationships and same-attribution relationships and relationship pointing rules are defined; based on each entity node, entity attribute, and relationship pointing rules in the multi-level entity hierarchy, a knowledge graph model containing multi-dimensional entity and relationship attributes is generated.

[0064] In some embodiments of this application, the extraction module 32 can be specifically used to extract entity nodes from regulatory reporting system document data and generate globally unique UIDs based on the entity types and multi-level entity hierarchies defined by the knowledge graph model, while associating them with entity attributes preset by the knowledge graph model; constructing association relationships between entity nodes according to the attribution, affiliation, and relationship pointing rules defined by the knowledge graph model; generating triple data consisting of entity nodes, entity attributes, and association relationships; and storing the triple data in a data lake warehouse after verification processing.

[0065] In some embodiments of this application, when extracting entity nodes and generating globally unique UIDs from regulatory reporting system document data based on entity types and multi-level entity hierarchies defined by the knowledge graph model, and simultaneously associating them with entity attributes preset by the knowledge graph model, the extraction module 32 can specifically be used to extract entity nodes from regulatory reporting system document data according to entity types defined by the knowledge graph model using OCR technology combined with regular expression templates; classify the extracted entity nodes according to the multi-level entity hierarchy of the knowledge graph model, wherein the indicator entity nodes in the summary report-level reporting system are divided into first-level indicators, second-level indicators, and third-level indicators; generate globally unique UIDs for each entity node using the Chinese name of the entity as the matching item according to the coding rules of the knowledge graph model; extract the entity attributes preset by the knowledge graph model from the regulatory reporting system document data, and associate the entity attributes with the globally unique UIDs.

[0066] In some embodiments of this application, the generation module 33 can be specifically used to convert entity nodes in triplet data into node data in a graph database and entity attributes into attribute fields of nodes using mapping rules defined by the knowledge mapping module; based on the attribution, membership, and relationship pointing rules defined by the knowledge graph model, the association relationships between entity nodes in triplet data are converted into edge data in a graph database, wherein the relationship type corresponds to the edge label and the relationship pointing rule corresponds to the edge direction, thus obtaining the association edges between entity nodes; using preset semantic rules, semantic matching and association reasoning are performed on entity nodes in different reporting systems to generate cross-system entity association relationships; after standardizing the converted node data, edge data, and cross-system entity association relationships according to the import format requirements of the graph database, the data is imported into the graph database to construct a visualized knowledge graph network of regulatory reporting systems.

[0067] In some embodiments of this application, when using preset semantic rules to perform semantic matching and association reasoning on entity nodes in different reporting systems to generate cross-system entity association relationships, the generation module 33 can specifically be used to calculate the first similarity of the Chinese names of entities in different reporting systems using an edit distance algorithm. When the first similarity is greater than or equal to a first preset threshold, it is determined as a candidate entity and a first cross-system association relationship is generated. For candidate entities in the first cross-system association relationship, the entity attribute matching degree of the same entity in different reporting systems is calculated. When the entity attribute matching degree is greater than or equal to a second preset threshold, it is determined as a synonym entity and an association is established, thus obtaining the first... The second cross-system association relationship includes entity attribute matching parameters. For synonymous entities confirmed in the second cross-system association relationship, cross-system matching is performed based on the globally unique UID of the entity node. If they have the same globally unique UID, they are ultimately determined to be entities from the same source and a third cross-system association relationship with a unique identifier is generated. For indicator entities in the summary report-level reporting system, after completing the above matching of the entity's Chinese name, entity attributes, and UID, the business association between indicators in different reporting systems is inferred by matching the calculation rules and composition fields of secondary and tertiary indicators, and a fourth cross-system association relationship containing indicator hierarchy information is generated.

[0068] In some embodiments of this application, such as Figure 4 As shown, the device also includes: a construction module 34 and a display module 35; Module 34 can be used to build a front-end display interface, which provides a fuzzy search interface for knowledge graph networks. The display module 35 can be used to receive search keywords input by users based on the fuzzy search interface, and use the edit distance algorithm to calculate the second similarity between the search keywords and the preset matching words. When the second similarity is greater than or equal to the third preset threshold, the search is triggered, and the entity details of the search results are displayed on the visualization page, including the entity's level, relationship type, complete attribute value, and the source document citation information of the entity in the regulatory reporting system. The preset matching words are any one of the following: the Chinese name of the main node entity, the Chinese name of the child node entity, and the entity attribute.

[0069] It should be noted that other corresponding descriptions of the functional units involved in the regulatory system knowledge graph construction device based on a multi-level entity model provided in this embodiment can be found in the following references. Figure 1 and Figure 2 The corresponding descriptions in [the document] will not be repeated here.

[0070] Based on the above, Figure 1 and Figure 2Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method. Figure 1 and Figure 2 The method for constructing a regulatory system knowledge graph based on a multi-level entity model is shown.

[0071] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause an electronic device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0072] Based on the above, Figure 1 and Figure 2 The method shown, and Figure 3 and Figure 4 To achieve the above objectives, the present application also provides an electronic device, specifically a personal computer, tablet computer, server, or other network device, as shown in the virtual device embodiment. This device includes a storage medium and a processor; the storage medium stores a computer program; the processor executes the computer program to achieve the above-described objectives. Figure 1 and Figure 2 The method for constructing a regulatory system knowledge graph based on a multi-level entity model is shown.

[0073] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0074] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0075] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0076] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware.

[0077] This invention, through modeling and analysis of regulatory reporting system document data, designs a knowledge graph model containing multi-dimensional entity and relational attributes. Based on the knowledge graph model, it extracts and stores triples containing entity nodes, attributes, and attribution and affiliation relationships. These are then converted into node and edge data in a graph database via a knowledge mapping module. Simultaneously, it infers cross-system entity relationships and imports them into the graph database to generate a visualized knowledge graph network. In this process, data is divided according to collection specifications and entity hierarchical splitting is performed. The resulting multi-level entity hierarchy and globally unique UID encoding enable structured data storage and unique identification, solving the data dispersion problem caused by offline document management. By defining attribution and affiliation relationships and relationship pointing rules, and combining semantic rules to infer cross-system entity relationships, entities scattered across different systems can establish semantic connections through the nodes and edges of the graph database, replacing the traditional manual offline cross-system relationship analysis method. The final generated visualized knowledge graph network can not only quickly locate entity details and relationships through fuzzy retrieval but also achieve business relationship inference based on indicator hierarchy information and calculation rules, significantly improving data analysis efficiency and meeting the needs for data integration and efficient analysis under increasingly stringent regulatory requirements.

[0078] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0079] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A method for constructing a regulatory system knowledge graph based on a multi-level entity model, characterized in that, include: Modeling and analysis of regulatory reporting system documents are conducted to design a knowledge graph model that includes multi-dimensional entity and relational attributes; Based on the knowledge graph model, triple data is extracted from the regulatory reporting system document data and stored in a data lake warehouse. The triple data includes entity nodes, entity attributes and relationships between entities. The relationships between entities include attribution relationships and same-family relationships. The knowledge mapping module converts the triple data in the data lake warehouse into node data and edge data in the graph database. It then uses semantic rules to infer cross-institutional entity relationships and imports the node data, edge data, and cross-institutional entity relationships into the graph database to generate a visualized knowledge graph network.

2. The method according to claim 1, characterized in that, The aforementioned modeling and analysis of regulatory reporting system document data, in order to design a knowledge graph model containing multi-dimensional entity and relational attributes, includes: The regulatory reporting system documents are divided into detailed-level reporting systems and summary report-level reporting systems according to the data collection specifications. The detailed reporting system and the summary report reporting system are split into entity levels according to the entity splitting rules to obtain entity types and multi-level entity levels. The entity splitting rules are as follows: the main entity node is the regulatory agency and the competent department, and the sub-nodes include reporting system, interface table, and fields / indicators. Among them, the indicator sub-nodes in the summary report reporting system are split into secondary indicators and tertiary indicators. A globally unique UID is generated using the entity's Chinese name as the matching item and used as the primary key. The entity attributes of each entity node in the multi-level entity hierarchy are associated with the primary key. For multiple entity nodes in the multi-level entity hierarchy, define the types of attribution and same-attribution relationships and the relationship pointing rules; Based on each entity node in the multi-level entity hierarchy, the entity attributes, and the relationship pointing rules, a knowledge graph model containing multi-dimensional entity and relationship attributes is generated.

3. The method according to claim 1, characterized in that, The step of extracting triplet data from the regulatory reporting system document data based on the knowledge graph model and storing it in a data lake warehouse includes: Based on the entity types and multi-level entity hierarchy defined in the knowledge graph model, entity nodes are extracted from the regulatory reporting system document data and globally unique UIDs are generated, while also associating them with the entity attributes preset by the knowledge graph model. Based on the attribution, membership, and relationship pointing rules defined in the knowledge graph model, construct the association relationships between the entity nodes; Generate triplet data consisting of the entity node, the entity attribute, and the association relationship, and store the triplet data in the data lake warehouse after verification processing.

4. The method according to claim 3, characterized in that, Based on the entity types and multi-level entity hierarchies defined in the knowledge graph model, entity nodes are extracted from the regulatory reporting system document data and globally unique UIDs are generated. Simultaneously, the entity attributes preset by the knowledge graph model are associated, including: Using OCR technology combined with regular expression templates, entity nodes are extracted from the regulatory reporting system document data according to the entity types defined in the knowledge graph model. According to the multi-level entity hierarchy of the knowledge graph model, the extracted entity nodes are divided into levels, wherein the indicator entity nodes in the summary report-level reporting system are divided into first-level indicators, second-level indicators and third-level indicators. According to the encoding rules of the knowledge graph model, a globally unique UID is generated for each entity node, using the entity's Chinese name as the matching item. Extract the entity attributes preset by the knowledge graph model from the regulatory reporting system document data, and associate the entity attributes with the globally unique UID.

5. The method according to claim 1, characterized in that, The process involves using a knowledge mapping module to convert the triple data in the data lake warehouse into node data and edge data in a graph database. Semantic rules are then used to infer cross-institutional entity relationships. The node data, edge data, and cross-institutional entity relationships are then imported into the graph database to generate a visualized knowledge graph network. This includes: Using the mapping rules defined by the knowledge mapping module, the entity nodes in the triplet data are converted into node data in the graph database, and the entity attributes are converted into attribute fields of the nodes. Based on the attribution relationship, kinship relationship and relationship pointing rule defined in the knowledge graph model, the association relationship between entity nodes in the triplet data is converted into edge data of the graph database, wherein the relationship type corresponds to the edge label, and the relationship pointing rule corresponds to the edge direction, thus obtaining the association edge between entity nodes; Using pre-defined semantic rules, semantic matching and association reasoning are performed on entity nodes in different reporting systems to generate cross-system entity association relationships; The obtained node data, edge data, and cross-institutional entity relationships are standardized according to the import format requirements of graph databases and then imported into a graph database to construct a visualized knowledge graph network of the regulatory reporting system.

6. The method according to claim 5, characterized in that, The process involves using preset semantic rules to perform semantic matching and association reasoning on entity nodes in different reporting systems to generate cross-system entity association relationships, including: The first similarity of the Chinese names of entities in different reporting systems is calculated using the edit distance algorithm. When the first similarity is greater than or equal to the first preset threshold, it is determined as a candidate entity and the first cross-system association relationship is generated. For the candidate entities in the first cross-system association relationship, calculate the entity attribute matching degree of the same entity in different reporting systems. When the entity attribute matching degree is greater than or equal to the second preset threshold, it is determined to be a synonym entity and an association is established to obtain the second cross-system association relationship. The second cross-system association relationship includes the entity attribute matching degree parameter. For the synonymous entities already confirmed in the second cross-system association relationship, cross-system matching is performed based on the globally unique UID of the entity node. If they have the same globally unique UID, they are finally determined to be the same source entity and a third cross-system association relationship with a unique identifier is generated. For indicator entities in the summary report-level reporting system, after completing the matching of the Chinese name, entity attributes and UID of the above entities, the business relationship between indicators in different reporting systems is inferred by matching the calculation rules and component fields of secondary and tertiary indicators, and a fourth cross-system relationship containing indicator hierarchical information is generated.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: Construct a front-end display interface, and provide a fuzzy search interface for the knowledge graph network in the front-end display interface; Based on the fuzzy search interface, the system receives search keywords input by the user and uses an edit distance algorithm to calculate the second similarity between the search keywords and preset matching words. When the second similarity is greater than or equal to a third preset threshold, a search is triggered, and the entity details of the search results are displayed on the visualization page, including the entity's level, relationship type, complete attribute values, and the source document citation information of the entity in the regulatory reporting system. The preset matching words are any one of the following: the Chinese name of the main node entity, the Chinese name of the child node entity, and the entity attribute.

8. A device for constructing a regulatory system knowledge graph based on a multi-level entity model, characterized in that, include: The analysis module is used to model and analyze regulatory reporting system documents to design a knowledge graph model that includes multi-dimensional entity and relational attributes. The extraction module is used to extract triple data from the regulatory reporting system document data based on the knowledge graph model and store it in the data lake warehouse. The triple data includes entity nodes, entity attributes and relationships between entities. The relationships between entities include attribution relationships and same-family relationships. The generation module is used to convert the triple data in the data lake warehouse into node data and edge data of the graph database through the knowledge mapping module, and to infer cross-institutional entity relationships through semantic rules, and import the node data, edge data and cross-institutional entity relationships into the graph database to generate a visualized knowledge graph network.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

10. An electronic device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

Images