Digital credit service management method and platform based on multi-source data

Abstract

The application provides a digital credit service management method and platform based on multi-source data, and relates to the technical field of service management.The method comprises the following steps: performing basic authorization verification on a credit query request to construct a standardized event message, extracting multi-modal heterogeneous original data from a multi-source data lake for vectorization coding, performing entity association mining according to a query context, constructing an asynchronous query task based on an entity association network, taking a subject credit state feature vector as a center node feature, dynamically weighting and fusing credit feature vectors of associated entities, and outputting an interpretable credit report.The technical problem of the prior art that the data source is single in credit evaluation, and there is a lack of risk analysis of associated entities, resulting in one-sided evaluation results and low accuracy and reliability is solved.The technical effect of comprehensively evaluating the credit of a subject through multi-source data, deeply mining the risk of associated entities, improving the accuracy and reliability of credit evaluation, and further improving the efficiency of credit service management is achieved.

Description

Technical Field

[0001] This invention relates to the field of service management technology, specifically to a digital credit service management method and platform based on multi-source data. Background Technology

[0002] Traditional credit assessment methods primarily rely on an entity's static financial data and historical credit records, such as loan statements, repayment records, or credit rating data from financial institutions. These data dimensions are limited, lacking a comprehensive analysis of business activities, asset status, and external events, making it difficult to fully and accurately depict an entity's creditworthiness. Furthermore, in today's complex economic environment, businesses or individuals often have close ties with multiple related entities, such as equity investments, cross-appointment of personnel, fund transfers, or transaction chains. These relationships can have a direct or potential impact on an entity's creditworthiness. However, current technologies do not adequately mine information from related entities during the credit assessment process, leading to incomplete risk identification and significant biases in the assessment results.

[0003] Existing technologies suffer from technical problems such as relying on a single data source and lacking risk analysis of related entities in credit assessments, leading to biased, inaccurate, and unreliable assessment results. Summary of the Invention

[0004] The purpose of this application is to provide a digital credit service management method and platform based on multi-source data, in order to solve the technical problems of existing technologies, such as the single data source in credit assessment and the lack of risk analysis of related entities, which leads to one-sided assessment results and low accuracy and reliability.

[0005] In view of the above problems, this application provides a digital credit service management method and platform based on multi-source data.

[0006] The first aspect of this application provides a digital credit service management method based on multi-source data. This method includes: after a query gateway performs basic authorization verification on a credit query request received by a digital credit service platform, constructing a standardized event message based on the credit query request, wherein the standardized event message includes a query object identifier and a query context; extracting multimodal heterogeneous raw data from a multi-source data lake based on the query object identifier; aggregating the multimodal heterogeneous raw data and performing vectorized encoding to construct a subject credit status feature vector; using the query object identifier as the query starting point, performing entity association mining based on the query context to construct an entity association network; constructing an asynchronous query task based on the entity association network to extract full-dimensional time-series credit features from the multi-source data lake and constructing an associated entity credit feature vector; and using the subject credit status feature vector as the central node feature, performing context-aware dynamic weighted fusion on the associated entity credit feature vector to output an interpretable credit report.

[0007] Optionally, starting with the query object identifier, related transaction queries are performed based on the query context to construct an explicit related entity topology; the transaction behavior patterns of the target query object corresponding to the query object identifier are traced back to mine and expand the implicit relationship of the explicit related entity topology to construct an implicit related entity topology; the explicit and implicit related entity topologies are used as the entity association network to construct an asynchronous query task, extract full-dimensional time-series credit features from the multi-source data lake, and obtain the related entity credit feature vector including explicit and implicit related credit state vectors.

[0008] Optionally, using the subject credit status feature vector as the central node feature, and based on the explicit and implicit related entity topologies, the explicit related credit status vector and the implicit related credit status vector are dynamically weighted and fused to output the interpretable credit report.

[0009] Optionally, by parsing the credit query request, request verification elements are extracted, wherein the request verification elements include the requesting financial institution code, the unique identifier of the query subject, and the query scenario purpose; the authorization rule service is invoked to traverse the basic authorization verification of the request verification elements; if the verification passes, the query context is constructed based on the requesting financial institution code and the query scenario purpose, wherein the query context includes the querying party identifier, query request metadata, and audit trail information; the unique identifier of the query subject is used as the query object identifier, and combined with the query context, the standardized event message is constructed.

[0010] Optionally, subject data is extracted from the multi-source data lake based on the query object identifier to obtain the multimodal heterogeneous raw data, wherein the multimodal heterogeneous raw data includes basic information, business information, financial information, property information, positive information, negative information, and credit rating information; a feature engineering pipeline is applied to the multimodal heterogeneous raw data using a credit feature rule base to obtain multi-dimensional credit features, wherein the feature engineering pipeline covers classification feature encoding, numerical index standardization, existence feature labeling, and time-series index derivation processing; the multi-dimensional credit features are sequentially concatenated to generate the subject credit status feature vector.

[0011] Optionally, the initiator identifier and associated user information are determined based on the requesting financial institution code to form the querying party identifier; business association mining is performed based on the purpose of the query scenario to obtain business purpose information and construct the query request metadata; a unique tracking serial number is generated as the audit trail information, wherein the querying party identifier, query request metadata and audit trail information constitute the query context.

[0012] Optionally, the association mining instructions are transformed based on the business purpose information of the query request metadata, wherein the association mining instructions include the mining target depth, mining target breadth, and association type preference; business registration information is retrieved based on the query party identifier to obtain explicit legal association information, wherein the explicit legal association information includes a list of entities associated with equity control chains and a list of entities associated with cross-appointment of personnel; taking the query object identifier as the query starting point, the entity associations of the explicit legal association information are integrated according to the association mining instructions to construct the explicit association entity topology.

[0013] Optionally, after extracting multiple transaction entities from the explicit related entity topology, they are mapped and added to the target query object to construct multiple transaction entities to be analyzed; multiple corporate transaction records of the multiple transaction entities to be analyzed are retrieved; multiple implicit related entities are filtered from the multiple corporate transaction records based on preset transaction frequency thresholds and transaction amount thresholds; transaction frequency and transaction amount are quantified into edge weight attributes to construct the topological association relationship between the multiple implicit related entities and the target query object, thereby generating the implicit related entity topology.

[0014] Optionally, the explicit and implicit related entity topologies are fused to obtain the transaction related entity topology; using the subject credit status feature vector as the central node feature, and based on the transaction related entity topology, the explicit and implicit related credit status vectors are dynamically weighted and fused according to business scenario context awareness, and the interpretable credit report is output.

[0015] A second aspect of this application provides a digital credit service management platform based on multi-source data. The platform includes: an event message construction module, used to construct a standardized event message based on the credit query request received by the digital credit service platform after the query gateway performs basic authorization verification on the request. The standardized event message includes a query object identifier and a query context. A data extraction module is used to extract multimodal heterogeneous raw data from a multi-source data lake based on the query object identifier. A feature vector construction module is used to aggregate the multimodal heterogeneous raw data and perform vectorized encoding to construct a subject credit status feature vector. A correlation mining module is used to perform entity correlation mining based on the query object identifier and the query context to construct an entity correlation network. A feature extraction module is used to construct an asynchronous query task based on the entity correlation network, extract full-dimensional time-series credit features from the multi-source data lake, and construct a related entity credit feature vector. A credit report output module is used to perform context-aware dynamic weighted fusion of the related entity credit feature vector, using the subject credit status feature vector as the central node feature, to output an interpretable credit report.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: After performing basic authorization verification on credit query requests received by the digital credit service platform through the query gateway, a standardized event message is constructed based on the credit query request. This standardized event message includes a query object identifier and a query context. Multimodal heterogeneous raw data is extracted from a multi-source data lake based on the query object identifier. The multimodal heterogeneous raw data is aggregated and vectorized to construct a subject credit status feature vector. Using the query object identifier as the query starting point, entity association mining is performed based on the query context to construct an entity association network. An asynchronous query task is constructed based on the entity association network to extract full-dimensional time-series credit features from the multi-source data lake, constructing a related entity credit feature vector. Using the subject credit status feature vector as the central node feature, the related entity credit feature vector is dynamically weighted and fused based on context awareness to output an interpretable credit report. This achieves the technical effect of comprehensively assessing subject credit through multi-source data, deeply mining related entity risks, improving the accuracy and reliability of credit assessment, and thus improving the efficiency of credit service management.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the digital credit service management method based on multi-source data provided in this application.

[0020] Figure 2 A schematic diagram of the structure of the digital credit service management platform based on multi-source data provided in this application.

[0021] Figure labeling: Event message construction module 11, data extraction module 12, feature vector construction module 13, association mining module 14, feature extraction module 15, credit report output module 16. Detailed Implementation

[0022] This application provides a digital credit service management method and platform based on multi-source data, addressing the technical problems of existing technologies where credit assessments suffer from single data sources and lack of risk analysis of related entities, leading to biased, inaccurate, and unreliable assessment results. It achieves the technical effect of comprehensively assessing the creditworthiness of entities through multi-source data, deeply mining the risks of related entities, improving the accuracy and reliability of credit assessments, and ultimately enhancing the efficiency of credit service management.

[0023] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0024] Example 1, as Figure 1 As shown, this application provides a digital credit service management method based on multi-source data, which includes: After the query gateway performs basic authorization verification on the credit query request received by the digital credit service platform, it constructs a standardized event message based on the credit query request. The standardized event message includes the query object identifier and the query context.

[0025] Furthermore, after the query gateway performs basic authorization verification on the credit query request received by the digital credit service platform, a standardized event message is constructed based on the credit query request. The method includes: parsing the credit query request to extract request verification elements, wherein the request verification elements include the requesting financial institution code, the unique identifier of the query subject, and the query scenario purpose; calling the authorization rule service to traverse the basic authorization verification of the request verification elements; if the verification passes, constructing the query context based on the requesting financial institution code and the query scenario purpose, wherein the query context includes the querying party identifier, query request metadata, and audit trail information; using the unique identifier of the query subject as the query object identifier, and combining it with the query context, constructing the standardized event message.

[0026] Specifically, in a digital credit service platform, when the query gateway receives a credit query request from a financial institution or other authorized user, it uses data parsing technology to disassemble and analyze the credit query request data packet to extract request verification elements. These elements include the requesting financial institution code, the unique identifier of the query subject, and the purpose of the query. For example, for credit query requests transmitted via HTTP / HTTPS, the network message is decapsulated using protocol parsing technology, and the request header and request body are parsed separately. The request header extracts basic fields such as authentication tokens and institution identifiers, while the request body carries specific business parameters. For data in the request body, typically in JSON or XML format, the corresponding data deserialization and parsing engine is invoked. For example, a JSON Schema or XML Schema parser is used to hierarchically decompose the data fields, expanding the nested structure into a set of key-value pairs. Combined with a predefined request field mapping rule table, the parsed fields are semantically matched and located. For example, the requesting financial institution code, the unique identifier of the query subject, and the purpose of the query are identified through field paths or field tags. The requesting financial institution code identifies the institution initiating the query, such as ABC Bank 001, ensuring that each query can be traced back to a specific institution. The unique identifier of the querying entity is the unique ID of the corresponding queried object, such as an individual's ID card number or a company's unified social credit code. The query scenario refers to the specific business scenario of this query, such as loan approval or business cooperation assessment, and is used to provide contextual constraints for targeted authorization and data processing in subsequent credit analysis. For example, the received credit query request data packet is a string in a specific format. Through data parsing algorithms, the requesting financial institution code is extracted as BANK_001, the unique identifier of the querying entity is ID_123456789, and the query scenario is loan approval.

[0027] The authorization rule service is invoked to iterate and verify the extracted verification elements. This service contains authorization conditions and restrictions imposed by different financial institutions on different query subjects under various query scenarios. It is a strategy-based authorization judgment component built on a rule engine. The construction process of the authorization rule service includes rule analysis and parametric modeling of regulatory requirements, business specifications, and institutional authorization agreements. Authorization conditions are layered and decomposed according to the dimensions of financial institution, query scenario, and query subject type. The executable query permissions, query subject scope, and usage restrictions of different financial institutions under different business scenarios are organized into structured authorization rules. Each authorization rule includes at least a rule triggering condition and a rule execution result. The rule triggering condition is used to limit constraints such as the requesting financial institution code, query scenario purpose, query subject type, and time window. The rule execution result is used to define whether the query is allowed, the range of allowed data, and additional restrictions. The authorization rules are configured into the rule engine in the form of rule expressions or decision tables, enabling the rule engine to perform item-by-item matching and traversal judgment on the extracted request verification elements at runtime.

[0028] The rules engine matches and checks the corresponding rule conditions one by one based on the extracted request verification elements. For example, for the requesting financial institution code BANK_001, if the authorization rule stipulates that it only has query permissions when the query scenario is loan approval and the query subject is a company, the engine will iterate through all relevant authorization rules to determine whether the current query request meets the authorization conditions. If the verification fails, the query gateway will reject the query request and return the corresponding error message to the requesting party, ensuring that only legitimate and compliant queries can continue. After the basic authorization verification passes, a query context is constructed based on the requesting financial institution code and the purpose of the query scenario. This query context provides a global description of the query, including: the querying party identifier, query request metadata, and audit trail information. The querying party identifier is composed of the initiator identifier determined by the requesting financial institution code and associated user information, used to identify the initiator and its associated user information, such as bank branch number and operator ID. The query request metadata includes business purpose information, request time, priority, etc. Business purpose information is constructed by mining the business associations of the query scenario to guide subsequent data mining and credit analysis processes. The audit trail information records audit-related information for the query request, generating a unique serial number, such as CQ202601210001, for full-process tracking and auditing, enabling accurate location and tracking of the query request during subsequent auditing and tracing. The unique identifier of the querying entity is used as the query object identifier, combined with the constructed query context to generate a standardized event message. This standardized event message is a unified data structure used internally by the digital credit service platform to process credit query requests, integrating the query object identifier and query context to provide a standardized input for subsequent credit service processes.

[0029] Basic authorization verification ensures that only legitimate and compliant query requests can enter the subsequent process, safeguarding the security and privacy of credit data, preventing unauthorized query behavior, and building standardized event messages to provide a unified and standardized input for subsequent operations such as extracting data from multi-source data lakes and constructing credit feature vectors, enabling the entire credit service process to proceed more efficiently and orderly.

[0030] Furthermore, the query context is constructed based on the requesting financial institution code and the purpose of the query scenario. The method includes: determining the initiator identifier and associated user information based on the requesting financial institution code to form the querying party identifier; performing business association mining based on the purpose of the query scenario to obtain business purpose information and constructing the query request metadata; and generating a unique tracking serial number as the audit trail information, wherein the querying party identifier, the query request metadata, and the audit trail information constitute the query context.

[0031] Specifically, the requesting financial institution's code is used as an index to perform a matching query in the institution master data management database. This identifies the initiator of the query and its associated user information, generating a query identifier. The query identifier comprehensively represents the financial institution that initiated the query and its specific operators. The initiator identifier clarifies the entity that initiated the query, while the associated user information refers to the information of the operators related to that financial institution. The institution master data management database stores detailed information about each financial institution, including the initiator identifier and associated user information. For example, a relational database, such as MySQL or Oracle, can be used to store financial institution information. Indexing techniques can be used to create an index on the financial institution code field, allowing for quick and accurate retrieval of the required information from the institution master data management database using SQL queries.

[0032] For each query scenario, a business knowledge mapping and rule matching mechanism is invoked to mine business associations, identify the actual business purpose information corresponding to the query, and construct query request metadata based on this information, providing clear guidance for credit data querying and processing. For example, if the query scenario is for loan approval, business rules determine that the query may be for obtaining information such as the borrower's credit status and repayment ability, thus determining the business purpose as assessing the borrower's loan risk. Specific business purpose information can be obtained through predefined business rules and knowledge graph technology, analyzing the potential connections between the query scenario and various business purposes, and mining business associations. Business rules can be stored and managed in the form of configuration files or database tables for easy updating and adjustment. Knowledge graph technology can be built using graph databases such as Neo4j, representing the relationships between businesses through nodes and edges. Based on the query scenario, the knowledge graph is traversed and matched to find the most suitable business purpose information.

[0033] Meanwhile, to ensure the traceability of the query process, a unique tracking serial number is generated as the audit trail information. The generation algorithm can use timestamps, random numbers, and system identification codes, combined through a hash algorithm to generate a fixed-length unique string as the tracking serial number. Then, the query party identifier, query request metadata, and unique tracking serial number are combined to form a query context. For example, when ABC Bank 001 initiates a credit inquiry to a company in a loan approval scenario, its query context can include the query party identifier of ABC Bank 001-Credit Department-Operator A12, loan approval, query request metadata valid for 30 days, and the audit trail serial number CTX202501210001. By constructing a query context, the background and purpose of each query can be clearly understood, ensuring that query operations comply with business specifications and regulatory requirements.

[0034] Multimodal heterogeneous raw data is extracted from the multi-source data lake based on the query object identifier.

[0035] The multimodal heterogeneous raw data are aggregated and vectorized to construct a subject credit status feature vector.

[0036] Furthermore, the multimodal heterogeneous raw data is aggregated and vectorized to construct a subject credit status feature vector. The method includes: extracting subject data from the multi-source data lake based on the query object identifier to obtain the multimodal heterogeneous raw data, wherein the multimodal heterogeneous raw data includes basic information, business information, financial information, property information, positive information, negative information, and credit rating information; applying a credit feature rule base to perform feature engineering pipeline processing on the multimodal heterogeneous raw data to obtain multi-dimensional credit features, wherein the feature engineering pipeline covers classification feature encoding, numerical index standardization, existence feature labeling, and time series index derivation processing; and sequentially concatenating the multi-dimensional credit features to generate the subject credit status feature vector.

[0037] Specifically, a multi-source data lake refers to a data management system that unifies the access and storage of data from different business systems, regulatory interfaces, and third-party data sources. It uses the query object identifier as a unified index key, and locates data in the metadata directory service of the multi-source data lake based on this identifier. Through a unified identifier mapping mechanism, the query object identifier is mapped to the subject identifier used within each data source, such as the corresponding key value of an enterprise's unified social credit code in the industrial and commercial database, transaction database, and credit reporting database. The data access interface layer of the data lake, such as standard interfaces like JDBC and ODBC, initiates parallel query requests for different data partitions and data themes, thereby retrieving the subject data, including basic information data, operational information data, financial statement data, property registration data, positive credit event data, and negative risk event data. For structured data, a columnar storage-based conditional query method is used for efficient filtering; for semi-structured or text-based data, tag indexing or full-text indexing technology is used for correlated retrieval. Ultimately, the results returned from different data sources are integrated to form multimodal heterogeneous raw data. Multimodal heterogeneous data refers to a multidimensional data set related to the query object identifier, originating from different data sources and possessing different formats and structures. This includes basic information, operational information, financial information, asset information, positive information, negative information, and credit rating information. Basic information describes the entity's identity attributes; operational and financial information reflects the entity's sustainable operating capacity; asset information reflects the level of asset protection; positive and negative information reflects historical credit behavior; and credit rating information provides external or historical credit judgment references. For example, if the query object identifier is 9131000xxxxxx1234, its business registration information, financial indicators for the past three years, corporate account transaction summaries, and litigation or penalty records can be synchronously obtained from the multi-source data lake. By acquiring comprehensive, authentic, and cross-dimensional multimodal heterogeneous raw data, the bias in credit assessment caused by a single data source can be avoided, improving the completeness and accuracy of digital credit services.

[0038] The feature engineering pipeline processes the multimodal heterogeneous raw data by invoking a pre-built credit feature rule base. This credit feature rule base contains credit feature transformation rules and methods to guide the transformation of different types of data into computable features. Following the transformation rules and methods in the credit feature rule base, the feature engineering pipeline sequentially performs operations such as categorical feature encoding, numerical index standardization, existence feature labeling, and time-series index derivation on the multimodal heterogeneous raw data to obtain multidimensional credit features. Specifically, categorical feature encoding converts categorical features in the multimodal heterogeneous raw data, such as gender and enterprise industry category in basic information, into a numerical form that can be processed by a computer. A common encoding method is one-to-many encoding, which creates a binary column for each category. For example, for gender, male is encoded as [1,0] and female as [0,1]. Numerical index standardization refers to standardizing numerical indicators in the multimodal heterogeneous raw data, such as income and assets. For example, Z-score standardization can be used to convert them into data with the same scale and distribution to eliminate dimensional differences. Simultaneously, existence feature markers are generated for event-based information such as whether there are lawsuits or administrative penalties. Samples with this information are marked as 1, and those without are marked as 0, which can concisely represent the existence or non-existence of information. Furthermore, for data with time-series characteristics, such as quarterly financial statements of enterprises and monthly consumption records of individuals, time-series indicator derivation processing is performed to generate dynamic features such as growth rate and volatility, reflecting the changing trends and stability of the data.

[0039] Multi-dimensional credit features are sequentially concatenated to form a subject credit status feature vector of uniform length and semantic consistency. The concatenation order can be determined according to business needs. For example, basic information-related features can be concatenated first, and then features such as business information and financial information can be concatenated in sequence. The subject credit status feature vector contains the credit information of the queried object in various aspects, and can comprehensively reflect the subject's credit status.

[0040] By transforming multimodal heterogeneous raw data into structured subject credit status feature vectors, not only are the problems of diverse raw data formats and difficulty in direct analysis solved, but the data can also comprehensively and holistically reflect the credit information of the queried object in various aspects, improving the accuracy and objectivity of credit assessment, and thus improving the efficiency and reliability of digital credit service management.

[0041] Using the query object identifier as the starting point, entity association mining is performed based on the query context to construct an entity association network.

[0042] An asynchronous query task is constructed based on the entity association network, and full-dimensional time-series credit features are extracted from the multi-source data lake to construct a credit feature vector of associated entities.

[0043] Furthermore, the method also includes: using the query object identifier as the query starting point, performing related transaction queries based on the query context to construct an explicit related entity topology; tracing back the transaction behavior pattern of the target query object corresponding to the query object identifier, performing implicit relationship mining and expansion on the explicit related entity topology to construct an implicit related entity topology; using the explicit and implicit related entity topologies as the entity association network to construct an asynchronous query task, extracting full-dimensional time-series credit features from the multi-source data lake, and obtaining the related entity credit feature vector including explicit and implicit related credit state vectors.

[0044] Specifically, the query object identifier is used as the starting point for the query. Entity association mining is performed based on the query context, including the query party identifier, query request metadata, and audit trail information, to construct an entity association network. This network includes explicit and implicit entity association topologies. The explicit entity association topology, reflecting clearly identifiable relationships under legal or institutional constraints, is constructed by querying and integrating legal information such as the query object's direct business registration, equity control chains, and cross-appointment of personnel. This topology is built with the query object identifier as the central node, each associated entity as an associated node, and the associated relationships as edges. Simultaneously, the transaction behavior patterns of the target query object corresponding to the query object identifier are traced back. Based on this backtracking mining, potential indirect associated entities are identified by analyzing transaction frequency, transaction amount thresholds, and behavioral pattern similarities, thus constructing an implicit entity association topology.

[0045] The entity association network is based on explicit and implicit entity association topologies. Asynchronous query tasks are generated through parallel graph traversal and batch processing. Specifically, nodes in the entity association network represent the query object and its associated entities, while edges represent explicit or implicit associations. A graph traversal algorithm, such as Breadth-First Search (BFS), is used to sequentially visit each node in the entity association network, extracting the multi-source data access requests required by each node during the traversal. Each data access request is encapsulated into a batch processing task that can be executed in parallel, generating asynchronous query tasks. Asynchronous query tasks are used to improve query efficiency and avoid system blocking caused by synchronous queries of large amounts of data. Message queues, such as RabbitMQ and Kafka, can be used to schedule and manage asynchronous query tasks. Query tasks are encapsulated as messages and sent to the message queue. Consumer processes retrieve messages from the queue and execute query operations, achieving parallel processing of query tasks. Each query task can target one or more related entities, extracting relevant full-dimensional time-series credit features from a multi-source data lake. These full-dimensional time-series credit features include data from different time points, such as basic information, financial information, transaction information, and credit rating information. The specific extraction process is consistent with the data processing method for the entity's credit status feature vector, and will not be elaborated here. For explicit and implicit related entities, full-dimensional time-series credit features are extracted separately. These extracted features are then organized to construct explicit and implicit related credit status vectors, which are ultimately combined into a complete related entity credit feature vector. The organization of the extracted full-dimensional time-series credit features includes standardization or normalization of numerical features and encoding of categorical features to ensure the quality and consistency of the feature vector.

[0046] By constructing an entity association network, we can comprehensively and deeply mine the related entities of the query object, including not only directly visible explicit related entities but also implicit related entities, thereby gaining a more comprehensive understanding of the credit status of the query object. Based on the entity association network, we can construct asynchronous query tasks and extract full-dimensional time-series credit features, enabling efficient and accurate acquisition of credit information of related entities. This improves the accuracy and reliability of credit assessment and effectively prevents credit risk.

[0047] Furthermore, using the query object identifier as the query starting point, related transaction queries are performed based on the query context to construct an explicit related entity topology. The method includes: transforming a related mining instruction based on the business purpose information of the query request metadata, wherein the related mining instruction includes mining target depth, mining target breadth, and related type preference; retrieving business registration information based on the query party identifier to obtain explicit legal related information, wherein the explicit legal related information includes a list of entities related to equity control chains and a list of entities related to cross-appointment of personnel; and integrating the entity associations of the explicit legal related information based on the query object identifier as the query starting point and the related mining instruction to construct the explicit related entity topology.

[0048] Specifically, query request metadata is extracted from the query context, and business purpose information is parsed into specific operation parameters through rule mapping and template matching technology. For example, in the business-mining instruction mapping table predefined by expert experience, loan approval is mapped to a mining depth of 3, a mining breadth of 10 nodes per layer, and an association type preference of equity control chain first, followed by personnel cross-appointment. During the parsing process, based on keyword matching for business objectives, semantic tag analysis, and contextual constraints, the required hierarchical range for mining, the limit on the number of nodes at each level, and the key types of relationships to focus on are determined. This generates association mining instructions that can be directly executed by graph analysis and asynchronous query tasks. These instructions map business semantics into operable parameters, including mining target depth, mining target breadth, and association type preference. Mining target depth refers to the retrieval depth of upward or downward association levels; mining target breadth refers to the maximum number of association nodes to be retrieved at each level; and association type preference is used to specify the priority association types to mine, such as equity associations, transaction associations, and personnel associations. Different business objectives have different preferences for association types. For example, equity associations are a key focus when assessing corporate control risks, while transaction associations have a higher priority when analyzing the stability of business cooperation.

[0049] The system calls the business registration information interface, authorizes access using the queryer's identifier, and retrieves explicit legal association information of the target entity from the business registration database. This explicit legal association information includes a list of entities associated with equity control chains and a list of entities associated with cross-appointment relationships. The equity control chain list reflects the equity control relationships between enterprises, while the cross-appointment relationship list reflects the employment status of key personnel across multiple enterprises. After obtaining the explicit legal association information, based on the depth, breadth, and association type preferences set in the association mining instructions, it filters out eligible associated entities from the explicit legal association information. Centered on the query object identifier, it connects and integrates the entity associations obtained from the explicit legal association information according to association type and hierarchical relationship, mapping equity and personnel relationships from different sources into a unified topological representation, forming an explicit association entity topology. This explicit association entity topology is a directed graph structure. For example, equity control relationships are used as edges to connect the query object to its associated entities on its equity control chain, and cross-appointment relationships are used as edges to connect related cross-appointment relationships. Specifically, a graph database, such as Neo4j, is used to store and represent the explicit association entity topology.

[0050] For example, the query object is identified as Technology Co., Ltd., and the business purpose of the requested metadata is to conduct risk assessment in loan approval. Based on this business purpose, the query is transformed into an association mining instruction: the mining target depth is three levels, the mining target breadth covers relevant industries, and the association type preference is equity association and personnel cross-appointment association. By calling the business registration information interface, the explicit legal association information of Technology Co., Ltd. is obtained: the equity control chain related entity list includes directly controlled A Software Co., Ltd. (70% shareholding) and A Hardware Co., Ltd. (65% shareholding), A Software Co., Ltd. indirectly controls A R&D Center (55% shareholding), and the personnel cross-appointment related entity list shows that the general manager of Technology Co., Ltd. also serves as a director of A Consulting Co., Ltd. Using the graph database Neo4j, with Technology Co., Ltd. as the query starting point, entity associations are integrated according to the association mining instruction. A Software Co., Ltd., A Hardware Co., Ltd., and A R&D Center are connected to Technology Co., Ltd. with equity control relationships as edges, and A Consulting Co., Ltd. is connected to Technology Co., Ltd. with personnel cross-appointment relationships as edges, constructing an explicit related entity topology.

[0051] By precisely converting association mining instructions, we ensured that the mining direction closely aligned with business objectives, thereby improving the targeting and effectiveness of association entity mining. Obtaining explicit legal association information can directly and accurately reflect the legal relationship between the queried object and related entities, which helps to more comprehensively and deeply assess the credit status of the queried object and improve the accuracy and reliability of credit service management.

[0052] Furthermore, by tracing back the transaction behavior patterns of the target query object corresponding to the query object identifier, implicit relationship mining and expansion are performed on the explicit related entity topology to construct an implicit related entity topology. The method includes: extracting multiple transaction entities from the explicit related entity topology and mapping them to the target query object to construct multiple transaction entities to be analyzed; retrieving multiple corporate transaction records of the multiple transaction entities to be analyzed; filtering multiple implicit related entities from the multiple corporate transaction records based on preset transaction frequency thresholds and transaction amount thresholds; quantifying transaction frequency and transaction amount into edge weight attributes to construct the topological association relationship between the multiple implicit related entities and the target query object, thereby generating the implicit related entity topology.

[0053] Specifically, graph query languages ​​such as Cypher are used to extract multiple transaction entities from the explicit related entity topology. This topology is constructed based on explicit relationships such as equity and personnel. The transaction entities within this topology are enterprises or institutions that have direct or indirect connections with the target query object. These transaction entities are mapped and added to the target query object, constructing multiple transaction entities to be analyzed. For example, if the explicit related entity topology shows that target query object A has an equity relationship with company B, and company C has a cross-appointment relationship with company B, then companies A, B, and C will all be considered as transaction entities to be analyzed. Third-party payment / clearing system interfaces are then invoked, and through security authentication and authorization mechanisms, the corporate transaction records between the target query object and these multiple transaction entities to be analyzed are obtained, including direct transaction records and related third-party transaction records, to capture indirect fund flow relationships. The system calculates the frequency and amount of transactions between each transaction entity and the target query object, and compares these values ​​with preset transaction frequency and transaction amount thresholds. This allows for the filtering of multiple implicitly related entities from various corporate transaction records. The transaction frequency threshold refers to the number of transactions occurring within a certain period, while the transaction amount threshold refers to the total amount of a single transaction or the total transaction amount within a certain period. For example, if the transaction frequency threshold is set to more than 5 transactions per month, and the transaction amount threshold is set to more than 100,000 yuan per transaction, and it is found that Company E and the target query object Company A have transacted 8 times in one month, with each transaction exceeding 100,000 yuan, then Company E is filtered as an implicitly related entity.

[0054] By utilizing graph databases, transaction frequency and transaction amount are quantified as edge weight attributes, serving as strength indicators of topological edges between nodes. This connects implicitly related entities with the target query object, generating an implicitly related entity topology. The quantification of transaction frequency and transaction amount as edge weight attributes can employ linear normalization or weighted combination formulas. For example, edge weight W can be defined... ij =a×(f ij / f max )+b×(A ij / A max), where f ij For the number of transactions between target query object i and implicitly related entity j, A ij f represents the cumulative amount of the corresponding transactions. max and A max , , represent the maximum number of transactions and transaction amount in the current network, respectively, used for normalization. 'a' and 'b' are adjustable weight coefficients, summing to 1, used to control the relative importance of transaction frequency and transaction amount in the edge weights, and can be dynamically set according to actual needs. The edge weight reflects the tightness of implicit associations; the higher the transaction frequency and the larger the transaction amount, the greater the edge weight.

[0055] By constructing an implicitly related entity topology, the relationship network of the target query object is made more comprehensive, thereby enabling a more comprehensive and accurate assessment of the target query object's credit status, operational stability, and potential risks, and further improving the accuracy and reliability of interpretable credit reports.

[0056] Using the subject's credit status feature vector as the central node feature, the associated entity's credit feature vector is dynamically weighted and fused based on context awareness to output an interpretable credit report.

[0057] Furthermore, using the subject credit status feature vector as the central node feature, and based on the explicit and implicit related entity topologies, the explicit and implicit related credit status vectors are dynamically weighted and fused to output the interpretable credit report.

[0058] Furthermore, using the subject credit status feature vector as the central node feature, the related entity credit feature vectors are dynamically weighted and fused based on context awareness to output an interpretable credit report. The method includes: fusing the explicit related entity topology and the implicit related entity topology to obtain the transaction related entity topology; using the subject credit status feature vector as the central node feature, and based on the transaction related entity topology, dynamically weighting and fusing the explicit related credit status vector and the implicit related credit status vector based on business scenario context awareness to output the interpretable credit report.

[0059] Specifically, a graph merging algorithm is used to fuse explicit and implicit related entity topologies to form a transaction-related entity topology. This transaction-related entity topology more comprehensively includes all related entities and their relationships surrounding the main enterprise. Specifically, the explicit and implicit related entity topologies are represented as graphs G and G, respectively. s =(V s E s ) and G h =(V h E hIn this graph, V represents the set of nodes, such as businesses or individuals, and E represents the set of edges, including relationships and additional weights. The graph merging algorithm generates a transaction-related entity topology G by combining the union of the node sets and the edge sets. t =(V t E t V t =V s ∪V h E t =E s ∪E h ∪E o E o To handle edge weight merging when two graphs have the same pair of nodes, the maximum value, weighted average, or normalized summation methods can be used. For example, the edge weight merging formula is: , where m is the coefficient for the fusion of explicit and implicit edge weights, which can be adjusted according to business preferences.

[0060] Using the entity's credit status feature vector as the central node feature, and based on the transaction-related entity topology, a dynamic weighted fusion of explicit and implicit related credit status vectors is performed based on business scenario context awareness. Context awareness refers to dynamically adjusting the weight allocation during the fusion process based on the business scenario, reflecting the focus of different business scenarios. For example, in a loan approval scenario, explicit equity control chain associations may have a higher weight, while in a risk monitoring scenario, the weight of frequent large-amount implicit transaction associations is increased. This can be dynamically set according to actual needs and experience. The weight calculation can use a weighted average method, where the central node v... c Its neighbor node v i Credit vector according to weight w i The fusion generates a new credit feature representation: This enables differentiated credit assessments for various business objectives. New credit characteristics are transformed into interpretable credit reports, which include correlation analysis and risk transmission paths. The interpretable credit report comprises the entity's credit status, correlation analysis, and risk transmission paths. Correlation analysis refers to a weighted analysis of the contributions of explicit and implicit neighbor nodes, illustrating the risk impact of each related entity on the entity's credit. The risk transmission path is based on the transaction-related entity topology and edge weight tracing of the potential impact links from the risk source node to the entity node.

[0061] By comprehensively considering the impact of both explicit and implicit related entities on a subject's creditworthiness, and through dynamic weighted fusion and business scenario context awareness, credit assessment becomes more comprehensive and accurate. Simultaneously, the generated interpretable credit reports provide decision-makers with clear decision-making support, enhancing the transparency and credibility of credit assessments, reducing credit risk, and improving the effectiveness and accuracy of digital credit service management.

[0062] Example 2 is based on the same inventive concept as the digital credit service management method based on multi-source data in the foregoing examples, such as... Figure 2 As shown, this application provides a digital credit service management platform based on multi-source data, wherein the digital credit service management platform based on multi-source data includes: The event message construction module 11 is used to construct a standardized event message based on the credit query request received by the digital credit service platform after the query gateway performs basic authorization verification. The standardized event message includes a query object identifier and a query context. The data extraction module 12 is used to extract multimodal heterogeneous raw data from the multi-source data lake according to the query object identifier. The feature vector construction module 13 is used to aggregate the multimodal heterogeneous raw data and perform vectorization encoding to construct a subject credit status feature vector. The association mining module 14 is used to perform entity association mining based on the query object identifier as the query starting point and the query context to construct an entity association network. The feature extraction module 15 is used to construct an asynchronous query task based on the entity association network, extract full-dimensional time-series credit features from the multi-source data lake, and construct an associated entity credit feature vector. The credit report output module 16 is used to perform context-aware dynamic weighted fusion of the associated entity credit feature vector with the subject credit status feature vector as the central node feature, and output an interpretable credit report.

[0063] Furthermore, the system is also used to: take the query object identifier as the query starting point, perform related transaction queries according to the query context, and construct an explicit related entity topology; trace back the transaction behavior pattern of the target query object corresponding to the query object identifier, perform implicit relationship mining and expansion on the explicit related entity topology, and construct an implicit related entity topology; use the explicit and implicit related entity topologies as the entity association network to construct an asynchronous query task, extract full-dimensional time-series credit features from the multi-source data lake, and obtain the related entity credit feature vector including explicit related credit state vector and implicit related credit state vector.

[0064] Furthermore, the credit report output module 16 is also used to: take the subject credit status feature vector as the central node feature, and perform dynamic weighted fusion of the explicit related credit status vector and the implicit related credit status vector according to the explicit related entity topology and the implicit related entity topology, and output the interpretable credit report.

[0065] Furthermore, the event message construction module 11 is also used to: extract request verification elements by parsing the credit query request, wherein the request verification elements include the requesting financial institution code, the unique identifier of the query subject, and the query scenario purpose; call the authorization rule service to traverse the basic authorization verification of the request verification elements; if the verification passes, construct the query context based on the requesting financial institution code and the query scenario purpose, wherein the query context includes the querying party identifier, query request metadata, and audit trail information; use the unique identifier of the query subject as the query object identifier, and combine it with the query context to construct the standardized event message.

[0066] Furthermore, the feature vector construction module 13 is also used to: extract subject data from the multi-source data lake according to the query object identifier to obtain the multimodal heterogeneous raw data, wherein the multimodal heterogeneous raw data includes basic information, business information, financial information, property information, positive information, negative information, and credit rating information; apply a credit feature rule base to perform feature engineering pipeline processing on the multimodal heterogeneous raw data to obtain multi-dimensional credit features, wherein the feature engineering pipeline covers classification feature encoding, numerical index standardization, existence feature labeling, and time series index derivation processing; and sequentially concatenate the multi-dimensional credit features to generate the subject credit status feature vector.

[0067] Furthermore, the event message construction module 11 is also used to: determine the initiator identifier and associated user information based on the requesting financial institution code to form the querying party identifier; perform business association mining based on the query scenario purpose to obtain business purpose information and construct the query request metadata; and generate a unique tracking serial number as the audit trail information, wherein the querying party identifier, query request metadata and audit trail information constitute the query context.

[0068] Furthermore, the system is also used for: transforming association mining instructions based on the business purpose information of the query request metadata, wherein the association mining instructions include mining target depth, mining target breadth, and association type preference; retrieving business registration information based on the query party identifier to obtain explicit legal association information, wherein the explicit legal association information includes a list of entities associated with equity control chains and a list of entities associated with cross-appointment of personnel; and integrating the entity associations of the explicit legal association information based on the association mining instructions, using the query object identifier as the query starting point, to construct the explicit association entity topology.

[0069] Furthermore, the system is also used to: extract multiple transaction entities from the explicit related entity topology, map them into the target query object, and construct multiple transaction entities to be analyzed; retrieve multiple corporate transaction records of the multiple transaction entities to be analyzed; filter multiple implicit related entities from the multiple corporate transaction records based on preset transaction frequency thresholds and transaction amount thresholds; quantify transaction frequency and transaction amount into edge weight attributes, construct the topological association relationship between the multiple implicit related entities and the target query object, and generate the implicit related entity topology.

[0070] Furthermore, the credit report output module 16 is also used to: fuse the explicit related entity topology and the implicit related entity topology to obtain the transaction related entity topology; take the subject credit status feature vector as the central node feature, and based on the transaction related entity topology, dynamically weightedly fuse the explicit related credit status vector and the implicit related credit status vector based on business scenario context awareness, and output the interpretable credit report.

[0071] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The digital credit service management method and specific examples based on multi-source data in the aforementioned embodiment one are also applicable to the digital credit service management platform based on multi-source data in this embodiment. Through the foregoing detailed description of the digital credit service management method based on multi-source data, those skilled in the art can clearly understand the digital credit service management platform based on multi-source data in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0072] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0073] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A digital credit service management method based on multi-source data, characterized in that, The method includes: After the query gateway performs basic authorization verification on the credit query request received by the digital credit service platform, it constructs a standardized event message based on the credit query request. The standardized event message includes the query object identifier and the query context. Multimodal heterogeneous raw data is extracted from the multi-source data lake based on the query object identifier; The multimodal heterogeneous raw data is aggregated and vectorized to construct a subject credit status feature vector; Starting with the query object identifier, entity association mining is performed based on the query context to construct an entity association network; An asynchronous query task is constructed based on the entity association network, and full-dimensional time-series credit features are extracted from the multi-source data lake to construct a credit feature vector of associated entities. Using the subject's credit status feature vector as the central node feature, the associated entity's credit feature vector is dynamically weighted and fused based on context awareness to output an interpretable credit report.

2. The digital credit service management method based on multi-source data as described in claim 1, characterized in that, The method further includes: Starting with the query object identifier, perform related transaction queries based on the query context to construct an explicit related entity topology; By tracing back the transaction behavior patterns of the target query objects corresponding to the query object identifiers, implicit relationship mining and expansion are performed on the explicit related entity topology to construct an implicit related entity topology; The explicit and implicit associated entity topologies are used as the entity association network to construct an asynchronous query task. Full-dimensional time-series credit features are extracted from the multi-source data lake to obtain the associated entity credit feature vector, which includes explicit and implicit associated credit state vectors.

3. The digital credit service management method based on multi-source data as described in claim 2, characterized in that, Using the subject credit status feature vector as the central node feature, and based on the explicit and implicit related entity topologies, the explicit and implicit related credit status vectors are dynamically weighted and fused to output the interpretable credit report.

4. The digital credit service management method based on multi-source data as described in claim 2, characterized in that, After the query gateway performs basic authorization verification on the credit query request received by the digital credit service platform, a standardized event message is constructed based on the credit query request. The method includes: By parsing the credit inquiry request, the request verification elements are extracted, including the requester's financial institution code, the unique identifier of the inquiry subject, and the purpose of the inquiry scenario. Invoke the authorization rules service to traverse the basic authorization verification of the requested verification elements; If the verification passes, the query context is constructed based on the requester's financial institution code and the purpose of the query scenario. The query context includes the requester's identifier, query request metadata, and audit trail information. The standardized event message is constructed by using the unique identifier of the query subject as the identifier of the query object and combining it with the query context.

5. The digital credit service management method based on multi-source data as described in claim 1, characterized in that, The method involves aggregating the multimodal heterogeneous raw data and performing vectorized encoding to construct a subject credit status feature vector. Based on the query object identifier, the main data is extracted from the multi-source data lake to obtain the multimodal heterogeneous raw data, wherein the multimodal heterogeneous raw data includes basic information, business information, financial information, property information, positive information, negative information and credit rating information; The feature engineering pipeline is applied to the multimodal heterogeneous raw data using a credit feature rule base to obtain multidimensional credit features. The feature engineering pipeline covers classification feature encoding, numerical index standardization, existence feature labeling, and time series index derivation processing. The multi-dimensional credit features are sequentially concatenated to generate the subject's credit status feature vector.

6. The digital credit service management method based on multi-source data as described in claim 4, characterized in that, The method for constructing the query context based on the requesting financial institution code and the intended use of the query scenario includes: The initiator identifier and associated user information are determined based on the requesting financial institution code, which constitutes the querying party identifier; Based on the purpose of the query scenario, business association mining is performed to obtain business purpose information, and the query request metadata is constructed. A unique tracking serial number is generated as the audit trail information, wherein the query party identifier, query request metadata, and audit trail information constitute the query context.

7. The digital credit service management method based on multi-source data as described in claim 6, characterized in that, Using the query object identifier as the starting point, and based on the query context, performing related transaction queries to construct an explicit related entity topology, the method includes: The association mining instructions are transformed based on the business purpose information of the query request metadata, wherein the association mining instructions include the depth of the mining target, the breadth of the mining target, and the association type preference; The business registration information is retrieved based on the query party identifier to obtain explicit legal association information, which includes a list of entities associated with the equity control chain and a list of entities associated with cross-appointment of personnel. Using the query object identifier as the query starting point, entity association integration of the explicit legal association information is performed according to the association mining instruction to construct the explicit association entity topology.

8. The digital credit service management method based on multi-source data as described in claim 7, characterized in that, The method involves tracing back the transaction behavior patterns of the target query object corresponding to the query object identifier, and then performing implicit relationship mining and expansion on the explicit related entity topology to construct an implicit related entity topology. After extracting multiple transaction entities from the explicit associated entity topology, they are mapped and added to the target query object to construct multiple transaction entities to be analyzed; Retrieve multiple corporate transaction records of the multiple transaction entities to be analyzed; Multiple implicitly related entities are filtered from multiple corporate transaction records based on preset transaction frequency and transaction amount thresholds; The transaction frequency and transaction amount are quantified as edge weight attributes, and the topological relationship between the multiple implicitly related entities and the target query object is constructed to generate the implicitly related entity topology.

9. The digital credit service management method based on multi-source data as described in claim 2, characterized in that, Using the subject's credit status feature vector as the central node feature, the associated entity's credit feature vector is dynamically weighted and fused based on context awareness to output an interpretable credit report. The method includes: By integrating the explicit and implicit related entity topologies, a transaction related entity topology is obtained. Using the subject credit status feature vector as the central node feature, and based on the transaction-related entity topology, the explicit and implicit related credit status vectors are dynamically weighted and fused according to business scenario context awareness, and the interpretable credit report is output.

10. A digital credit service management platform based on multi-source data, characterized in that: The step of implementing the digital credit service management method based on multi-source data according to any one of claims 1 to 9, wherein the digital credit service management platform based on multi-source data comprises: The event message construction module is used to construct a standardized event message based on the credit query request after the query gateway performs basic authorization verification on the credit query request received by the digital credit service platform. The standardized event message includes a query object identifier and a query context. The data extraction module is used to extract multimodal heterogeneous raw data from the multi-source data lake based on the query object identifier; The feature vector construction module is used to aggregate the multimodal heterogeneous raw data, perform vectorized encoding, and construct the subject credit status feature vector. The association mining module is used to perform entity association mining based on the query object identifier as the query starting point and the query context to construct an entity association network. The feature extraction module is used to construct an asynchronous query task based on the entity association network, extract full-dimensional time-series credit features from the multi-source data lake, and construct a credit feature vector of associated entities. The credit report output module is used to perform context-aware dynamic weighted fusion of the credit feature vectors of the associated entities, with the subject's credit status feature vector as the central node feature, and output an interpretable credit report.

Images