A method, apparatus, and data reporting method for constructing data lineage based on multicaliber data.

By constructing semantic lineage graphs in the financial and insurance industries, and automatically extracting and transforming business logic from regulatory policy documents, the flexibility and traceability issues of data lineage construction are resolved, enabling efficient and accurate multi-caliber data reporting.

CN123068415APending Publication Date: 2026-11-10SHANGHAI FAIRYLAND INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611107577.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-24
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

In highly regulated industries such as finance and insurance, existing technologies struggle to adapt data lineage construction methods to policy changes quickly. Furthermore, the conceptual lineage mapping is disconnected from the underlying execution logic, resulting in poor data lineage traceability and high system maintenance and link reconstruction costs.

Method used

By acquiring unstructured rule documents, the understanding engine extracts business concepts, categorical entities, and derived logic, generates rule triples, converts them into executable code, and stores them in the derivation relationship edges of the semantic lineage graph. This realizes the data lineage between business concepts and categorical entities. A confidence score and graph update mechanism are introduced to ensure the validity and consistency of the logic.

Benefits of technology

It enables rapid response to changes in regulatory policies, reduces development and maintenance costs, improves the traceability of data lineage and the accuracy of data reporting, and reduces the risks of human error and cross-platform data inconsistency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123068415A_ABST
    Figure CN123068415A_ABST
Patent Text Reader

Abstract

This application provides a method, apparatus, and data reporting method for constructing data lineages based on multiple calibers, belonging to the field of data processing technology. The data lineage construction method includes: obtaining an unstructured first rule document; extracting business concept entities, caliber entities, and derived logic from the first rule document using an understanding engine, and combining them into rule triples; converting the rule triples into corresponding first executable code according to the logic type determined by the derived logic; establishing a first derived relationship edge in a semantic lineage graph connecting the business concept node corresponding to the business concept entity and the caliber node corresponding to the caliber entity, and storing the first executable code in the edge attribute of the first derived relationship edge to generate the data lineage between the business concept entity and the caliber entity. This application reduces rule response time and achieves efficient structured expression of the business data flow chain.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus and data reporting method for constructing data lineage based on multiple calibers. Background Technology

[0002] In heavily regulated sectors such as finance and insurance, institutions typically need to report business data to multiple different regulatory platforms (such as the State Financial Supervision and Administration Bureau, the Provincial Department of Finance, and the Provincial Department of Agriculture). Because different regulatory agencies have varying statistical standards and requirements for the same business indicators, companies need to build a clear data lineage tracking system to establish a mapping relationship between the underlying data of their core business systems and the reporting standards of external multi-dimensional regulators, thereby ensuring the accuracy of the reported data.

[0003] To meet the multi-dimensional regulatory reporting requirements, the existing data processing workflow typically involves business personnel or data analysts studying various regulatory policy documents and manually extracting statistical rules for reporting indicators. Subsequently, developers write standardized Extract-Transform-Load (ETL) scripts based on these rules to cleanse the source data in the internal database and map it to target data that conforms to specific regulatory standards. During this process, the correspondence between internal source fields (Data Definition Language, DDL) and external regulatory fields is usually recorded in static tables or data dictionaries as a record of data lineage.

[0004] However, existing data lineage construction methods have significant limitations when facing increasingly complex and dynamically changing regulatory environments. On the one hand, regulatory rules are mostly issued in unstructured text form, relying entirely on manual reading and analysis to sort out the business concept system and complex derived calculation rules. This is not only time-consuming but also prone to deviations in rule extraction due to differences in individual subjective understanding, making it difficult to respond nimbly to frequent policy changes. On the other hand, in traditional construction models, the conceptual lineage mapping is separated from the underlying actual data transformation logic. Mapping records exist only as static metadata, while the script code that undertakes the actual transformation tasks is scattered across various scheduling systems. This statically separated architecture results in a lack of rigorous structured expression of the business data flow chain. The system cannot intuitively manage the dynamic association and actual processing process between specific nodes and their source business nodes, leading to poor traceability of data lineage. When facing multi-caliber switching or rule updates, the system maintenance and link reconstruction costs are very high. Summary of the Invention

[0005] This application provides a method, apparatus, and data reporting method for constructing data lineage based on multiple calibers, aiming to solve the problems of the difficulty in responding quickly to policy changes due to the manual processing of regulatory rules, and the poor traceability of data lineage caused by the disconnect between the conceptual lineage mapping and the underlying data conversion logic.

[0006] Firstly, this application provides a method for constructing data lineage based on multiple calibers, including: Obtain the unstructured first rule document; The understanding engine is used to extract business concept entities, caliber entities, and derived logic from the first rule document, and the business concept entities, caliber entities, and derived logic are combined to generate rule triples; The logic type is determined based on the derived logic, and the rule triple is converted into the corresponding first executable code based on the logic type. In the semantic lineage graph, a first derivation relationship edge is established connecting the business concept node corresponding to the business concept entity and the caliber node corresponding to the caliber entity, and the first executable code is stored in the edge attribute of the first derivation relationship edge to generate the data lineage between the business concept entity and the caliber entity.

[0007] As an example, converting the rule triples into corresponding first executable code according to the logical type includes: If the logical type is a direct mapping, then the fields in the rule triple are converted one-to-one into field codes, which serve as the first executable code; If the logical type is a calculation conversion, then the source field identifier, calculation expression and verification rule are extracted from the rule triple, and the source field identifier, calculation expression and verification rule are assembled into formula calculation code, which is used as the first executable code; If the logical type is an external association, then the interface address, request parameters, and response parsing rules in the rule triple are extracted, and the interface address, request parameters, and response parsing rules are assembled into application interface call code, which serves as the first executable code.

[0008] As one embodiment, after storing the first executable code in the edge attribute of the first derivation relationship edge to generate the data lineage between the business concept entity and the caliber entity, the method further includes: Verify the first executable code; If the verification passes, the state of the first derived relation edge is updated based on the confidence score of the derived logic.

[0009] As an example, updating the state of the first derived relation edge based on the confidence score of the derivation logic specifically includes: If the confidence score of the derived logic output by the language model is greater than or equal to the first threshold, then the first derived relation edge is controlled to take effect, and the state of the first derived relation edge is updated to the ready state. If the confidence score is greater than or equal to the second threshold and less than the first threshold, a review request for the first derived relation edge is triggered, and after receiving the review success information, the first derived relation edge is controlled to take effect, and the state of the first derived relation edge is updated to the ready state. If the confidence score is less than the second threshold, the first derived relation edge is disabled and an alarm is triggered.

[0010] As one example, the method for constructing a multi-caliber data lineage also includes: If updated information is detected in the rule base and / or reporting platform, the semantic lineage graph is updated according to the impact of the updated information on the semantic lineage graph.

[0011] As an example, when updated information is detected in the rule library, the semantic lineage graph is updated according to the impact of the updated information on the semantic lineage graph, including: Obtain the second rule document corresponding to the updated information, and establish a second derived relation edge based on the second rule document; In the semantic lineage graph, starting from the second derivation edge, the adjacent derivation edges and aperture nodes are traversed outward layer by layer to identify all aperture nodes affected by the second derivation edge as the first target aperture node; and the first transmission path from the second derivation edge to the first target aperture node is determined. Verify the validity of the derived logic on the first transmission path, and take the first transmission path that passes the validity verification as the first valid path; Regenerate the second executable code for the first valid path, and update the first valid path based on the second executable code.

[0012] As an example, when updated information is detected in the external interface to which the derived logic depends, the semantic lineage graph is updated according to the impact of the updated information on the semantic lineage graph, including: Extract the external interface call instructions contained in the executable code corresponding to the derived logic; Test the availability status of the external system interface corresponding to the external interface call command; If the available status is unavailable, a path conflict flag is added to the third derivation edge containing the external system interface to prevent the third derivation edge from taking effect, and an alarm message is generated.

[0013] As an example, when updated information is detected in the attributes of a business concept node, the semantic lineage graph is updated according to the impact of the updated information on the semantic lineage graph, including: In the semantic lineage graph, starting from the business concept node, the adjacent derived relationship edges and caliber nodes are traversed outward layer by layer to identify all caliber nodes affected by the attribute changes of the business concept node as the second target caliber node; and the second transmission path from the business concept node to the second target caliber node is determined. Verify the validity of the derived logic on the second transmission path, and take the second transmission path that passes the validity verification as the second valid path; The third executable code for the second valid path is regenerated, and the second valid path is updated based on the third executable code.

[0014] As an example, the edge attribute also stores the lineage metadata of the first derived edge, the lineage metadata including the edge version information, source policy document identifier, and effective timestamp corresponding to the first derived edge: The method further includes: If the verification of the first derivation relationship edge fails, read the lineage metadata; Based on the bloodline metadata, trace upstream in the semantic bloodline graph to locate the problem node that caused the verification failure.

[0015] Secondly, this application also provides a multi-caliber data lineage construction device, comprising: The document acquisition module is used to acquire unstructured first-rule documents; The triple extraction module is used to extract business concept entities, caliber entities, and derived logic from the first rule document using the understanding engine, and to combine the business concept entities, the caliber entities, and the derived logic to generate rule triples. The code conversion module is used to determine the logic type based on the derived logic, and convert the rule triple into the corresponding first executable code based on the logic type. The relation edge construction module is used to establish a first derived relation edge in the semantic lineage graph connecting the business concept node corresponding to the business concept entity and the caliber node corresponding to the caliber entity, and to store the first executable code in the edge attribute of the first derived relation edge to generate the data lineage between the business concept entity and the caliber entity.

[0016] Thirdly, this application also provides a multi-caliber data reporting method, including: The target reporting platform is determined based on the data reporting task; In the semantic lineage graph, determine the target caliber node corresponding to the target reporting platform, and traverse the target business concept nodes associated with the target caliber node in the semantic lineage graph; Extract the executable code stored in the edge attributes of the derived relationship edges connecting the target business concept node and the target caliber node; Source data is extracted from the target business concept node, and the executable code is executed to transform the source data and generate target reporting data; The target reporting data is sent to the target reporting platform.

[0017] Fourthly, this application also provides a multi-caliber data reporting device, comprising: The platform determination module is used to determine the target reporting platform based on the data reporting task; The graph query module is used to determine the target caliber node corresponding to the target reporting platform in the semantic lineage graph, and to traverse the target business concept nodes associated with the target caliber node in the semantic lineage graph. The code extraction module is used to extract the executable code between the target business concept node and the target caliber node; The data conversion module is used to extract source data from the target business concept node and execute the executable code to convert the source data and generate target reporting data. The sending module is used to send the target reporting data to the target reporting platform.

[0018] Fifthly, this application also provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-described multi-caliber data lineage construction methods or implements the above-described multi-caliber data reporting methods.

[0019] Sixthly, this application also provides a non-transitory computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-described multi-caliber data lineage construction methods or the above-described multi-caliber data reporting methods.

[0020] Seventhly, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements any of the above-described methods for constructing data lineage based on multi-caliber data or implements the above-described method for transmitting data based on multi-caliber data. Attached Figure Description

[0021] 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the semantic lineage map provided in this application; Figure 2 This is one of the flowcharts illustrating the multi-caliber data lineage construction method provided in this application; Figure 3 This is the second flowchart of the multi-caliber data lineage construction method provided in this application; Figure 4 This is a flowchart illustrating the process of constructing data lineage using reverse GraphRAG technology provided in this application; Figure 5 This is a thread diagram of an example of data lineage reconstruction based on rule updates provided in this application; Figure 6 This is a schematic diagram of the structure of the multi-caliber data lineage construction device provided in this application; Figure 7 This is a flowchart illustrating the multi-caliber data reporting method provided in this application; Figure 8 This is a schematic diagram of the structure of the multi-caliber data reporting device provided in this application; Figure 9 This is a schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0025] The terms "first," "second," etc., used in this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, without limiting the number of objects; for example, a first object can be one or more. Furthermore, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects have an "or" relationship.

[0026] In heavily regulated sectors such as finance and insurance, institutions typically need to report business data to multiple different regulatory platforms (such as the State Financial Supervision and Administration Bureau, the Provincial Department of Finance, and the Provincial Department of Agriculture). The same data entry must be submitted in different formats, using different standards and verification rules according to the requirements of each regulatory body. For example, the same agricultural insurance policy data must simultaneously meet the solvency adequacy ratio calculation standards of the regulatory platform, the fiscal subsidy disbursement standards of the Provincial Department of Finance, and the agricultural insurance coverage rate statistical standards of the Provincial Department of Agriculture.

[0027] To meet the multi-dimensional regulatory reporting requirements, existing technologies manually extract statistical rules for reporting indicators and establish data lineage relationships through ETL code. Specifically, traditional ETL data lineage technology extracts field-level mapping relationships and constructs a technical lineage graph by parsing technical code such as Structured Query Language (SQL) and stored procedures. For example, it identifies the technical path "Source table A. Field 1 → Transformation function → Target table B. Field 2". This technology can only track technical-level field mappings and cannot understand business semantics. For example, traditional lineage can identify the technical path "core_policy.premium_amount → report_table.premium_col", but it cannot recognize the semantic difference between the business concept of "policy amount" representing "original currency premium" under the regulatory platform and the need to deduct the farmer's self-paid portion and convert it into "subsidy base" under the finance department's regulations. The lack of a business semantic layer necessitates hardcoding independent ETL processes for each regulatory platform, leading to repetitive development and maintenance difficulties. Furthermore, this approach is not only inefficient but also prone to deviations in rule extraction due to differences in individual subjective understanding, making it difficult to respond nimbly to frequent policy changes.

[0028] In addition, in the traditional construction model, the conceptual lineage mapping is separated from the underlying data transformation logic, which makes the business data flow chain lack a rigorous structured expression. The system cannot intuitively manage the dynamic relationship and actual processing between a specific node and its source business concept node, resulting in poor traceability of data lineage. When facing multiple switching or rule updates, the system maintenance and link reconstruction costs are very high.

[0029] The following description, in conjunction with the accompanying drawings, describes the method, apparatus, and data reporting method for constructing data lineage based on multi-caliber data provided in this application.

[0030] It should be noted that the multi-caliber data lineage construction method provided in this application embodiment is implemented using a multi-caliber data lineage construction device. The multi-caliber data lineage construction method can capture the semantic relationship between the expressed business concept and the multi-caliber derivation, and can also achieve rapid and automatic adaptation after changes in regulatory policies. At the same time, it can apply graph retrieval-enhanced generation technology (GraphRAG) to the generation of lineage rules, quickly and automatically complete the reconstruction of data lineage caused by changes in rules, business concepts and calibers, and perform data reporting on this basis, ensuring that the institution can efficiently and accurately complete the data reporting task when facing the differentiated reporting requirements of multiple regulatory platforms.

[0031] This application describes the method for constructing a data lineage based on multiple calibers using a multi-caliber data lineage construction device as the execution subject.

[0032] like Figure 1 and Figure 3 As shown, the semantic kinship graph includes the following three-layer data structure: Business Concept Layer L1: This layer includes multiple business concept nodes (e.g., policy, premium, object, etc.). Each business concept node has a node sequence number (node_ID), which records the relevant attributes and data sources of the corresponding business concept entity, such as the definition of the business concept entity (e.g., object), the data source table, primary key fields (e.g., object number), etc. Reporting Scope Layer L2: This layer includes the reporting scope nodes for each reporting platform. For example, in the insurance sector, these nodes might include the regulatory platform's solvency profile, the finance department's subsidy application profile, and the agriculture department's coverage rate profile. Each reporting scope node includes attributes such as platform identifier (e.g., agriculture department), reporting form name (e.g., subsidy distribution table), field list (e.g., subsidies at various levels, planting area), and validation rules. Derivation layer L3: Used to store the derivation mapping relationship (i.e., derivation edge) from business concept layer L1 to reporting standard layer L2. It includes fields such as derivation logic code, code version information, confidence level, source policy document identifier, and effective timestamp. The derivation logic code is the executable code that converts the business concept entity into the reporting standard entity. The confidence level is used to store the reliability score of the information extracted by the language model. When the rule confidence level is low, manual review is triggered.

[0033] During initialization, the business concept entities to be reported in the core business system are registered in the business concept layer, and the corresponding scope nodes and derivation relationship edges are initialized. For the same business concept, according to the differentiated requirements of each reporting platform, derivation relationship edges pointing to different scope nodes are established, and each derivation relationship edge independently maintains its derivation logic and version information.

[0034] Given the characteristics of data reporting, when the same business data needs to be adapted to different reporting standards across multiple reporting platforms, it either requires adding new reporting requirements or changing the field definitions or calculation logic of existing standards. To address this, a three-layer structure of "Business Concept Layer - Reporting Standard Layer - Derivation Relationship Layer" is defined in the semantic lineage graph. This avoids the need to develop independent ETL processes for each reporting platform, significantly reducing development and maintenance costs.

[0035] Figure 2 This is one of the flowcharts illustrating the multi-caliber data lineage construction method provided in this application. For example... Figure 2 As shown, the multi-caliber data lineage construction method provided in this application includes: S210: Obtain the unstructured first rule document.

[0036] Rule documents are various policy documents, notices, announcements, or reporting guidelines issued by different reporting platforms. These documents are usually presented in natural language text formats, such as portable document format files, text documents, or rich text web pages, and have not yet been parsed into two-dimensional tables or key-value pair formats that databases can directly read. Therefore, rule documents are unstructured data.

[0037] In one possible implementation, after obtaining the rule document, it is stored in a rule base. During data lineage construction, the rule document is retrieved from the rule base.

[0038] S220: Use the understanding engine to extract business concept entities, caliber entities, and derived logic from the first rule document, and combine the business concept entities, caliber entities, and derived logic to generate rule triples.

[0039] refer to Figure 3 As shown, after extracting the rule document, the rule document is input into the understanding engine to extract lineage rules. The understanding engine is a large-scale language model built on a deep learning architecture and pre-trained on professional data from the application domain.

[0040] Understandably, after obtaining the first rule document, the understanding engine first preprocesses the rule document, converting unstructured documents such as PDF / Word / web pages into standardized text, preserving the chapter structure and table data, and generating structured text blocks to be processed, in order to prepare high-quality text feature data for subsequent processing.

[0041] Then, the understanding engine uses entity recognition and dependency parsing techniques from natural language processing to extract three types of information from the rule document: business concept entities, scope entities, and derived logic. Business concept entities refer to the basic business elements corresponding to the underlying data architecture of the enterprise's core business system, such as policy amount, insured's age, or insurance type code. Scope entities refer to the target statistical indicators or data fields required to be reported by the reporting platform in the first rule document, such as the combined expense ratio or specific risk exposure values. Derived logic is the specific calculation process or conversion conditions for transforming business concept entities into external scope entities.

[0042] After acquiring these three types of key information, the understanding engine uses business concept entities as source endpoints, caliber entities as destination endpoints, and derived logic as a relational description connecting the two endpoints to generate structured rule triples. These rule triples clearly depict what kind of processing the source data needs to undergo to meet the reporting requirements of the regulatory dimension.

[0043] In one possible implementation, this application uses GraphRAG technology to extract three types of key information.

[0044] S230: Determine the logic type based on the derived logic, and convert the rule triplet into the corresponding first executable code based on the logic type.

[0045] Logical types reflect the essential characteristics of data transformation. They can include types that read the underlying data as is, types that perform mathematical operations on multiple underlying fields, or types that require supplementary data from external systems.

[0046] After determining the logic type, a code generation engine or a pre-set code transformation template is used to directly translate the rule triples in natural language or semi-structured form into first executable code that the computer can run directly. This first executable code can be a structured query language script, a data processing script in a specific programming language, or configuration file code that an extraction, transformation, and loading tool can recognize. During this process, business concept entities and categorical entities are converted into input and output variables in the code, while derived logic is instantiated into data processing functions or logical operators in the code, thus achieving rule structuring.

[0047] S240: In the semantic lineage graph, establish a first derivation relationship edge connecting the business concept node corresponding to the business concept entity and the caliber node corresponding to the caliber entity, and store the first executable code in the edge attribute of the first derivation relationship edge to generate the data lineage between the business concept entity and the caliber entity, thereby realizing content-level lineage injection into the semantic lineage graph. Please refer to [reference needed]. Figure 3 As shown.

[0048] The first derivation edge clearly defines the start and end points of the data flow. Edge attributes are the data storage carriers attached to this derivation edge, used to record the metadata of the link, thereby transforming the triple into a lineage rule. (Reference) Figure 1 As shown, each box in the derivation relationship layer L3 represents a derivation relationship edge, which is used to connect a business concept node in the business concept layer L1 and a reporting caliber node in the reporting caliber layer L2.

[0049] In the GraphRAG technology of this application, the input data is a rule document, and the output is executable code representing lineage rules, with the aim of driving lineage reconstruction. For example... Figure 4 As shown, the reverse extraction process is as follows: First, the rule document (unstructured) undergoes reverse GraphRAG entity extraction to identify business concept entities, policy entities, and derived logic. Then, a graph triplet is constructed to create triples (business concept - derivation relationship - reporting policy). Next, rule generation technology converts the triples into executable ETL code with lineage attributes, achieving rule structuring. Finally, lineage injection embeds semantic lineage graph node identifiers, derived logic versions, and policy sources.

[0050] It should be noted that in existing technologies, GraphRAG technology is used in question-answering retrieval scenarios. As a static knowledge retrieval engine, it assists large models in locating context in massive amounts of text, employing a passive retrieval triggering mechanism. Furthermore, the graph itself lacks any execution capabilities, making it impossible to automate the process from regulatory policy release to automatic adaptation to the reporting system. This application, however, utilizes GraphRAG technology to proactively extract three key pieces of information from relevant materials in regulatory policies and reporting systems, generating ETL code to structure lineage rules. This process essentially leverages reverse GraphRAG thinking.

[0051] This application uses GraphRAG technology to automatically extract structured lineage rules from unstructured regulatory policy documents through reverse engineering. All rule extraction and graph construction are completed through a large language model, which significantly improves the speed of policy response.

[0052] This application's embodiments automatically extract entities and derived logic from unstructured rule documents and generate rule triples by introducing a language model, replacing the traditional model that relies on manual reading and analysis. This reduces rule response time from weeks to hours, achieving an order-of-magnitude efficiency improvement and eliminating subjective comprehension errors. Furthermore, this application's embodiments convert derived logic into executable code and store it directly in the edge attributes of derived relationship edges in the semantic lineage graph. Through lineage injection technology, the graph node structure is strongly bound to the underlying execution script, breaking the previous static architecture where the concept mapping graph and the actual execution code were separated. This achieves efficient and structured expression of the business data flow chain, enhances the traceability of data lineage, and makes each data field traceable. Tracing back to the original rule document, the system meets the requirements of the reporting platform for penetrating review, enabling it to reconstruct and maintain its lineage at extremely low cost when facing frequent changes in regulatory rules. This fundamentally solves the efficiency, cost, and review challenges of multi-caliber reporting. Furthermore, regarding development costs, traditional methods hard-code independent ETL processes for each regulatory platform, with maintenance costs increasing linearly with the number of platforms. This application's embodiment, however, achieves zero-code automatic adaptation across multiple reporting platforms based on a unified lineage graph, significantly reducing development and maintenance costs. Regarding data consistency, traditional multi-caliber reporting is prone to data conflicts due to differences in manual conversion logic. This application's embodiment fundamentally eliminates the risk of cross-platform data inconsistency through an automatic verification mechanism based on the same source and structure.

[0053] During rule structuring, different derived logic types employ different code conversion methods. In one possible implementation, in step S230, the rule triplet is converted into corresponding first executable code based on the logic type, including: If the logical type is a direct mapping, then the fields in the rule triple are converted one-to-one into field codes, which serve as the first executable code.

[0054] If the logic type is a calculation transformation, then the source field identifier, calculation expression, and validation rule are extracted from the rule triple, and the source field identifier, calculation expression, and validation rule are assembled into formula calculation code, which serves as the first executable code. In other words, when the derived logic type is a calculation transformation, the ETL code is formula calculation logic, with the first two fields being the source field identifier, followed by the calculation expression and validation rule.

[0055] If the logical type is an external association, then the interface address, request parameters, and response parsing rules are extracted from the rule triple, and the interface address, request parameters, and response parsing rules are assembled into application interface call code, which serves as the first executable code. In other words, when the derived type is an external association, the ETL code is the application programming interface (API) call logic, containing the external system interface address, request parameters, and response parsing rules.

[0056] Therefore, the length of the generated ETL code is uncertain, and the processing branch is determined by the derived logic type during code execution. For the transformation of data submitted for regulatory reporting, it is either direct mapping, formula calculation, or obtaining supplementary information from external sources. This layered design of code transformation logic can effectively support the integration of complex multi-source heterogeneous data with minimal storage cost.

[0057] This application embodiment refines the derived logic into three specific logic types: direct mapping, computational transformation, and external association. Differentiated field replacement, formula assembly, or application programming interface splicing strategies are adopted for different types to achieve accurate translation from unstructured triples to actual scenario code. This multi-branch code conversion mechanism can comprehensively cover various business scenarios commonly encountered in financial regulatory reporting, such as original value extraction, complex indicator calculation, and external data supplementation, effectively ensuring the business adaptability and accuracy of the generated first executable code.

[0058] In one possible implementation, such as Figure 3 As shown, after storing the first executable code in the edge attribute of the first derivation relationship edge to generate the data lineage between the business concept entity and the caliber entity (S240), the method further includes: S250: Verify the first executable code.

[0059] In S250, the first executable code is verified, including: S2510: Mark the state of the first derived relation edge as a state to be verified; S2520: In an isolated test environment (e.g., a sandbox environment), the first executable code is called to process the sample data and obtain the processing result; S2530: If the difference between the processing result and the expected result meets the requirements, the verification is successful.

[0060] S2540: If the difference between the processing result and the expected result meets the requirements, the test fails. The executable code for the business concept node and the caliber node is rolled back to the previous version, and a repair suggestion is generated. Please refer to [the relevant documentation]. Figure 3 As shown.

[0061] S260: If the verification passes, update the state of the first derived relation edge based on the confidence score of the derived logic.

[0062] Specifically, when the confidence score meets the requirements, the first derived relation edge is activated, and its state is updated to the ready state. The ready state means that the newly generated first derived relation edge can be used for subsequent actual data reporting tasks, thereby achieving state-level lineage injection into the semantic lineage graph. Please refer to [reference needed]. Figure 3 As shown. If the confidence score does not meet the requirements, the control over the first derived relation edge will not take effect.

[0063] This application introduces an isolated testing environment to test and verify the first executable code using sample data before it officially takes effect. It establishes a strict lineage logic admission mechanism to automatically intercept erroneous derived logic caused by model parsing deviations or abnormal source data before the code is put into the production environment. This effectively avoids the pollution of the underlying business system's real data and the final reported results by erroneous logic, and improves the security of the data lineage construction process and the reliability of the reported data.

[0064] In one possible implementation, the edge attribute also stores lineage metadata of the first derived edge, the lineage metadata including edge version information, source policy document identifier, and effective timestamp corresponding to the first derived edge: Accordingly, the method further includes: If the verification of the first derivation relationship edge fails, read the lineage metadata; Based on the bloodline metadata, trace upstream in the semantic bloodline graph to locate the problem node that caused the verification failure.

[0065] Specifically, since the core business system will still be processing normal business during the data lineage construction, in order to avoid the situation where other businesses cannot be processed normally due to the construction process, lineage metadata is added to the edge attributes of the derived relationship edges, including edge version information, source policy document identifier, and effective timestamp, etc., and when the verification fails, it will revert to the previous version based on the lineage metadata to prevent derived relationship edges that have not completed testing and verification from taking effect directly and to prevent the continuous propagation of erroneous rules.

[0066] This application embodiment embeds lineage metadata such as edge version information, source policy document identifier, and effective timestamp into the edge attributes, giving each derived relationship edge a multi-dimensional traceability label. When the testing process fails or subsequent data verification fails, the system can accurately trace back along the data link in the graph based on this lineage metadata, quickly locating which specific version of the policy rule or which upstream node caused the logical conflict, thereby significantly shortening the problem investigation cycle and improving the efficiency of data quality governance.

[0067] In one possible implementation, S260 updates the state of the first derived relation edge based on the confidence score of the derived logic, specifically including: If the confidence score of the derived logic output by the language model is greater than or equal to the first threshold, then the first derived relation edge is controlled to take effect, and the state of the first derived relation edge is updated to the ready state. If the confidence score is greater than or equal to the second threshold and less than the first threshold, a review request for the first derived relation edge is triggered, and after receiving the review success information, the first derived relation edge is controlled to take effect, and the state of the first derived relation edge is updated to the ready state. If the confidence score is less than the second threshold, the first derived relation edge is disabled and an alarm is triggered.

[0068] In one possible implementation, for high-confidence derived logic (e.g., confidence ≥ 0.9), the first derived relation edge automatically takes effect; for medium-confidence derived logic (e.g., 0.5 ≤ confidence < 0.9), manual review is requested, and the first derived relation edge automatically takes effect after the review is passed; for low-confidence derived logic (e.g., confidence < 0.5), the first derived relation edge is rejected and an alarm is issued.

[0069] This application's embodiments introduce confidence thresholds at various levels for lineage rules, enabling high-confidence rules to take effect directly, medium-confidence rules to undergo manual review to determine if they are correct, and low-confidence rules to not take effect. By controlling the quality of rules, the quality of data transmission is ensured.

[0070] In one possible implementation, during the operation of the data reporting system, the following is also included: If updated information is detected in the rule base and / or reporting platform, the semantic lineage graph is updated according to the impact of the updated information on the semantic lineage graph.

[0071] Specifically, the aforementioned updates may include updated information in the rule library (i.e., updated rule documents corresponding to new regulatory policies), changes to external interfaces that derived logic depends on, and changes to the attributes of business concept nodes (corresponding to core system DDL changes). These three scenarios are explained in detail below.

[0072] In one possible implementation, when the policy rule library contains updated information in the rule document, steps such as reverse GraphRAG extraction and rule structuring are immediately triggered (see the descriptions in S220 to S240), and impact scope analysis and path recalculation steps are performed.

[0073] If the rule library contains updated information, update the semantic lineage graph, including: P10: When an update is detected in the rule library, obtain the second rule document corresponding to the update information, and establish a second derived relation edge based on the second rule document.

[0074] Changes in the rule base can be detected by periodically polling or listening to the message queue. When an update is detected, it indicates that the reporting platform has issued new requirements or modified existing rules. At this time, the second rule document corresponding to the update information is obtained, and the entity extraction and code conversion steps in the aforementioned embodiments are reused (please refer to the descriptions in S210-S240 above) to construct a second derived relationship edge reflecting the latest regulatory intent in the semantic lineage graph.

[0075] P20: In the semantic lineage graph, starting from the second derivation edge, traverse the adjacent derivation edges and caliber nodes layer by layer outward, identify all caliber nodes affected by the second derivation edge as the first target caliber node, and determine the first transmission path from the second derivation edge to the first target caliber node.

[0076] Because insurance or financial business data often has a complex one-to-many network structure, an update to a policy rule can trigger a chain reaction. Therefore, starting with the newly generated second derived relation edge, a breadth-first search or depth-first search graph traversal algorithm is used to explore outwards layer by layer along the edge pointing relationships of the graph. During this process, all nodes that logically depend on the derived relation edge are identified and marked as the first target node. At the same time, an ordered set of nodes and derived relation edges that must be traversed from the update starting point to each first target node is recorded. This set is determined as the first transmission path, which reflects the influence range of the rule update in the graph.

[0077] P30: Verify the validity of the derived logic on the first transmission path, and take the first transmission path that passes the validity verification as the first valid path.

[0078] While the derived logic on the first transmission path is affected in terms of topology, this does not mean it can run directly. The logic on these paths is validated, for example, by checking whether the source fields referenced in the original calculation formulas still exist, whether the required data types match, and whether external interfaces are available. Paths that pass all validations and show no logical conflicts or omissions are designated as the first valid paths that can run normally. Warnings are issued for transmission paths with conflicts, and repair suggestions are generated.

[0079] P40: Regenerate the second executable code for the first valid path, and update the first valid path based on the second executable code.

[0080] For the first valid path that has passed verification, although its underlying business concept entity remains unchanged, it needs to incorporate new regulatory intent from the second rule document. Therefore, the code conversion module needs to be invoked to rewrite the changed business logic mapping relationships and new lineage metadata into the script, generating the second executable code. It should be noted that the newly generated executable code needs to be verified, and its effectiveness is determined based on a confidence score after successful verification.

[0081] Because continuous refactoring may affect the execution of other reporting tasks, if the refactoring verification is not completed within the specified time, the refactoring will be stopped, and processing will begin from the current rule to be verified when the next refactoring cycle starts.

[0082] Figure 5 A thread diagram is provided for an example of data lineage reconstruction based on rule updates. For example... Figure 5As shown, during the design phase, the second rule document is released at T0 minutes; from T0 to T1 minutes, preprocessing of the second rule document is completed to obtain structured text blocks; from T1 to T2 minutes, intent recognition, reverse GraphRAG extraction, triple generation, and confidence scoring of the second rule document are completed; then, path code is generated. Specifically, from T2 to T3 minutes, impact scope analysis and path recalculation are completed to generate valid paths; subsequently, code verification is performed. Specifically, from T3 to T4 minutes, ETL code generation and sandbox testing verification are completed, and the system marks the generated derived relationship edges as ready. At T4 minutes, if the test verification passes, the generated derived relationship edges are made effective directly based on the confidence score, realizing code implementation; or it is pushed to manual review. If the test verification fails, the executable code is rolled back to the previous version, and recovery suggestions are generated. Figure 5 It can be seen that the entire data lineage reconstruction process took T4 minutes.

[0083] This application embodiment achieves top-down adaptive reconstruction of rule changes by performing influence range analysis based on breadth or depth traversal in the graph. As a result, when the rules change, the system can automatically identify the affected node paths and regenerate the code, eliminating the need for developers to conduct a full investigation and manually modify scattered scripts, thus greatly reducing system maintenance costs.

[0084] In one possible implementation, dependency checks and path revalidation are triggered when the external interface of a derived logic dependency changes. Specifically, for cases where the external interface of a derived logic dependency changes, the semantic lineage graph is updated, including: Q10: When a change is detected in the external interface that the derived logic depends on, extract the external interface call instructions contained in the executable code corresponding to the derived logic.

[0085] External interfaces are third-party application service interfaces that provide currency conversion, external credit inquiries, or industry blacklist verification. Changes to external interfaces may include changes in service address, modification of request parameter format, or service shutdown. When the service registry or interface gateway detects such changes, it retrieves the relevant executable code from the semantic lineage graph and parses out the external interface call instruction responsible for initiating the network request.

[0086] Q20: Test the availability status of the external system interface (API) corresponding to the external interface call command.

[0087] Specifically, by constructing a simulated request packet or sending a network connectivity probe message, a test call is initiated to the external system interface pointed to by the obtained external interface call instruction, in order to obtain the actual response code or connectivity delay of the external system interface, and then assess its current availability status.

[0088] Q30: If the available status is unavailable, add a path conflict flag to the third derivation edge containing the external system interface to prevent the third derivation edge from taking effect and generate an alarm message.

[0089] If the probe results show a connection timeout or return an error code, the available state is determined to be unavailable. At this time, the third derived relationship edge containing the failed interface call is found in the semantic lineage graph, and a path conflict flag is written into its edge attribute. This flag will serve as an interception credential to prevent the scheduling system from sending and executing the abnormal code corresponding to the derived relationship edge. At the same time, an alarm message is pushed to the operation and maintenance terminal to prompt manual intervention.

[0090] Q40: If the available status is available, the third derived relation edge will automatically take effect.

[0091] This application's embodiments enhance the robustness of the data lineage network by detecting changes to external dependent interfaces and testing their availability. When an external third-party interface fails, the system can proactively trigger circuit breakers and alarms, preventing widespread failures of reporting tasks and cascading errors caused by upstream interface downtime, thus improving the system's fault tolerance.

[0092] In one possible implementation, when the attributes of a business concept layer node change (i.e., the core system source field DDL changes), an impact scope analysis and path recalculation are triggered. Specifically, for cases where the attributes of a business concept node change, the semantic lineage graph is updated, including: R1: When a change in the attribute of a business concept node is detected, in the semantic lineage graph, starting from the business concept node, the adjacent derived relationship edges and caliber nodes are traversed outward layer by layer, and all caliber nodes affected by the attribute change of the business concept node are identified as the second target caliber node; and a second transmission path from the business concept node to the second target caliber node is determined.

[0093] Changes to the attributes of business concept nodes typically correspond to database-level operations within the core business systems of an enterprise, such as modifications to database table structures, renaming tables, or splitting fields. When these change events are captured by database log monitoring or metadata synchronization tools, the changed business concept node is used as the origin. The semantic lineage graph is then traced downstream layer by layer along the data flow to identify all nodes that logically depend on the underlying source data for extraction and transformation. These nodes are designated as the second target nodes. Simultaneously, the directed graph connectivity trajectories extending from the change origin to each second target node are recorded, forming a second transmission path.

[0094] R2: Verify the validity of the derived logic on the second transmission path, and take the second transmission path that passes the validity verification as the second valid path.

[0095] R3: Regenerate the third executable code for the second valid path, and update the second valid path based on the third executable code. It should be noted that the newly generated executable code needs to be verified, and its effectiveness is determined based on a confidence score after successful verification.

[0096] Similar to the refactoring caused by rule changes, the existence of the second transmission path is checked. For those valid second paths where the interface remains connected and the logic is valid, the new underlying data source information is adapted for them. For example, the old table names in the original script are updated to the new table names, and obsolete single-field extraction instructions are split into multi-field concatenation instructions under the new structure. Then, a third executable code that can accurately interface with the current underlying physical architecture is generated and updated to the corresponding edge attributes.

[0097] This application embodiment implements a bottom-up data link tracing mechanism through graph search technology, enabling the data reporting system to adapt quickly to the physical architecture upgrade of the internal core system. When the underlying database table structure changes, the system can automatically execute script rewriting of the passively affected edges, ensuring the continuous synchronization of internal data asset evolution and external regulatory reporting business.

[0098] Based on the above, this application also provides a multi-caliber data lineage construction apparatus. The multi-caliber data lineage construction apparatus and the aforementioned multi-caliber data lineage construction method can be referred to and correspond to each other.

[0099] As an example, such as Figure 6 As shown, the multi-caliber data lineage construction device includes: Document acquisition module 610 is used to acquire unstructured first rule documents; The triple extraction module 620 is used to extract business concept entities, caliber entities and derived logic from the first rule document using the understanding engine, and combine the business concept entities, the caliber entities and the derived logic to generate rule triples. The code conversion module 630 is used to determine the logic type according to the derived logic, and convert the rule triplet into the corresponding first executable code according to the logic type; The relation edge construction module 640 is used to establish a first derived relation edge in the semantic lineage graph connecting the business concept node corresponding to the business concept entity and the caliber node corresponding to the caliber entity, and to store the first executable code in the edge attribute of the first derived relation edge to generate the data lineage between the business concept entity and the caliber entity.

[0100] This application's embodiments automatically extract entities and derived logic from unstructured rule documents and generate rule triples by introducing a language model, replacing the traditional model that relies on manual reading and analysis. This reduces rule response time from weeks to hours, achieving an order-of-magnitude efficiency improvement and eliminating subjective comprehension errors. Furthermore, this application's embodiments convert derived logic into executable code and store it directly in the edge attributes of derived relationship edges in the semantic lineage graph. Through lineage injection technology, the graph node structure is strongly bound to the underlying execution script, breaking the previous static architecture where the concept mapping graph and the actual execution code were separated. This achieves efficient and structured expression of the business data flow chain, enhances the traceability of data lineage, and makes each data field traceable. Tracing back to the original rule document, the system meets the requirements of the reporting platform for penetrating review, enabling it to reconstruct and maintain its lineage at extremely low cost when facing frequent changes in regulatory rules. This fundamentally solves the efficiency, cost, and review challenges of multi-caliber reporting. Furthermore, regarding development costs, traditional methods hard-code independent ETL processes for each regulatory platform, with maintenance costs increasing linearly with the number of platforms. This application's embodiment, however, achieves zero-code automatic adaptation across multiple reporting platforms based on a unified lineage graph, significantly reducing development and maintenance costs. Regarding data consistency, traditional multi-caliber reporting is prone to data conflicts due to differences in manual conversion logic. This application's embodiment fundamentally eliminates the risk of cross-platform data inconsistency through an automatic verification mechanism based on the same source and structure.

[0101] Based on the above, this application also provides a data reporting method based on multiple calibers. For example... Figure 7 As shown, the multi-caliber data reporting method provided in this application includes: S710: Determine the target reporting platform based on the data reporting task; S720: Determine the target caliber node corresponding to the target reporting platform in the semantic lineage graph, and traverse the target business concept nodes associated with the target caliber node in the semantic lineage graph; S730: Extract the executable code stored in the edge attributes of the derived relationship edges connecting the target business concept node and the target caliber node; S740: Extract source data from the target business concept node and execute the executable code to transform the source data and generate target reporting data; S750: Send the target reporting data to the target reporting platform.

[0102] In practice, when a data reporting task is triggered, the corresponding target reporting platform node is located according to the task's indication. Then, the derivation layer is traversed in reverse to obtain the corresponding target business concept node, derivation edges, and executable code stored in the edge attributes. Next, source data is extracted from the target business concept node of the core business system and transformed based on the executable code to generate target reporting data that meets the reporting requirements, while embedding lineage metadata in the output. Finally, the target reporting data is sent to the target reporting platform.

[0103] Upon receiving confirmation of the target data submission, record the lineage path version for this submission and continue processing other submission tasks. If no response is received within a specified period or verification fails, the problem node can be quickly located by tracing the lineage metadata, and retrying using historical versions or switching to the previous version of executable code.

[0104] In executing specific data reporting tasks, this application's embodiments effectively solve the lag problem of relying on manual reading and verification and manual maintenance of conversion scripts in traditional data reporting processes by calling the executable code generated from the automated parsing of unstructured rule documents based on language models in the early stage for source data transformation. This ensures that the final generated target reporting data can quickly and accurately meet the requirements of multi-dimensional external regulatory standards, significantly reducing the data reporting risk caused by human misunderstanding. On the other hand, when extracting and transforming source data, the semantic lineage graph is directly traversed, and the executable code stored in the edge attributes of the derivation relationship is extracted. This achieves seamless connection between upper-level logical concept routing and lower-level data execution operations, breaking the architectural barrier of the original static separation between concept mapping records and external scheduling scripts. This enables multi-caliber reporting tasks in complex networks to have a high degree of automated operation capability and clear data traceability, thereby effectively reducing the daily operation and maintenance pressure and link reconstruction cost of the reporting system.

[0105] Based on the above, this application also provides a multi-caliber data reporting device. The multi-caliber data reporting device and the aforementioned multi-caliber data reporting method can be referred to and correspond to each other.

[0106] As an example, such as Figure 8 As shown, the multi-caliber data reporting device includes: Platform determination module 810 is used to determine the target reporting platform based on the data reporting task; The graph query module 820 is used to determine the target caliber node corresponding to the target reporting platform in the semantic lineage graph, and to traverse the target business concept nodes associated with the target caliber node in the semantic lineage graph. Code extraction module 830 is used to extract executable code between the target business concept node and the target caliber node; The data conversion module 840 is used to extract source data from the target business concept node and execute the executable code to convert the source data and generate target reporting data. The sending module 850 is used to send the target reporting data to the target reporting platform.

[0107] In executing specific data reporting tasks, this application's embodiments effectively solve the lag problem of relying on manual reading and verification and manual maintenance of conversion scripts in traditional data reporting processes by calling the executable code generated from the automated parsing of unstructured rule documents based on language models in the early stage for source data transformation. This ensures that the final generated target reporting data can quickly and accurately meet the requirements of multi-dimensional external regulatory standards, significantly reducing the data reporting risk caused by human misunderstanding. On the other hand, when extracting and transforming source data, the semantic lineage graph is directly traversed, and the executable code stored in the edge attributes of the derivation relationship is extracted. This achieves seamless connection between upper-level logical concept routing and lower-level data execution operations, breaking the architectural barrier of the original static separation between concept mapping records and external scheduling scripts. This enables multi-caliber reporting tasks in complex networks to have a high degree of automated operation capability and clear data traceability, thereby effectively reducing the daily operation and maintenance pressure and link reconstruction cost of the reporting system.

[0108] Figure 9 This is a schematic diagram of the structure of the electronic device provided in this application, such as... Figure 9 As shown, the electronic device may include a processor 910, a communications interface 920, a memory 930, and a communication bus 940. The processor 910, communications interface 920, and memory 930 communicate with each other via the communication bus 940. The processor 910 can call logical instructions from the memory 930 to execute a multi-caliber data lineage construction method or to implement the aforementioned multi-caliber data reporting method.

[0109] Furthermore, the logical instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0110] On the other hand, this application also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer is able to execute the multi-caliber data lineage construction method or the multi-caliber data reporting method provided in the above embodiments.

[0111] In another aspect, this application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the multi-caliber data lineage construction method or the multi-caliber data reporting method provided in the above embodiments.

[0112] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0113] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for constructing a multi-caliber data lineage, characterized in that, include: Obtain the unstructured first rule document; The understanding engine is used to extract business concept entities, caliber entities, and derived logic from the first rule document, and the business concept entities, caliber entities, and derived logic are combined to generate rule triples; The logic type is determined based on the derived logic, and the rule triple is converted into the corresponding first executable code based on the logic type. In the semantic lineage graph, a first derivation relationship edge is established connecting the business concept node corresponding to the business concept entity and the caliber node corresponding to the caliber entity, and the first executable code is stored in the edge attribute of the first derivation relationship edge to generate the data lineage between the business concept entity and the caliber entity.

2. The method for constructing a multi-caliber data lineage according to claim 1, characterized in that, Based on the logical type, the rule triple is converted into the corresponding first executable code, including: If the logical type is a direct mapping, then the fields in the rule triple are converted one-to-one into field codes, which serve as the first executable code; If the logical type is a calculation conversion, then the source field identifier, calculation expression and verification rule are extracted from the rule triple, and the source field identifier, calculation expression and verification rule are assembled into formula calculation code, which is used as the first executable code; If the logical type is an external association, then the interface address, request parameters, and response parsing rules in the rule triple are extracted, and the interface address, request parameters, and response parsing rules are assembled into application interface call code, which serves as the first executable code.

3. The method for constructing a multi-caliber data lineage according to claim 1, characterized in that, If the rule library is found to have updated information, the second rule document corresponding to the updated information is obtained, and a second derivation relationship edge is established based on the second rule document; In the semantic lineage graph, starting from the second derivation edge, the adjacent derivation edges and aperture nodes are traversed outward layer by layer to identify all aperture nodes affected by the second derivation edge as the first target aperture node; and the first transmission path from the second derivation edge to the first target aperture node is determined. Verify the validity of the derived logic on the first transmission path, and take the first transmission path that passes the validity verification as the first valid path; Regenerate the second executable code for the first valid path, and update the first valid path based on the second executable code.

4. The method for constructing a multi-caliber data lineage according to claim 1, characterized in that, If it is detected that there is an update information in the external interface that the derived logic depends on, the external interface call instructions contained in the executable code corresponding to the derived logic are extracted; Test the availability status of the external system interface corresponding to the external interface call command; If the available status is unavailable, a path conflict flag is added to the third derivation edge containing the external system interface to prevent the third derivation edge from taking effect, and an alarm message is generated.

5. The method for constructing a multi-caliber data lineage according to claim 1, characterized in that, If updated information is detected in the attributes of a business concept node, in the semantic lineage graph, starting from the business concept node, the adjacent derived relationship edges and caliber nodes are traversed outward layer by layer, and all caliber nodes affected by the attribute change of the business concept node are identified as the second target caliber node; and a second transmission path from the business concept node to the second target caliber node is determined. Verify the validity of the derived logic on the second transmission path, and take the second transmission path that passes the validity verification as the second valid path; The third executable code for the second valid path is regenerated, and the second valid path is updated based on the third executable code.

6. A device for constructing a data lineage based on multi-caliber data, characterized in that, include: The document acquisition module is used to acquire unstructured first-rule documents; The triple extraction module is used to extract business concept entities, caliber entities, and derived logic from the first rule document using the understanding engine, and to combine the business concept entities, the caliber entities, and the derived logic to generate rule triples. The code conversion module is used to determine the logic type based on the derived logic, and convert the rule triple into the corresponding first executable code based on the logic type. The relation edge construction module is used to establish a first derived relation edge in the semantic lineage graph connecting the business concept node corresponding to the business concept entity and the caliber node corresponding to the caliber entity, and to store the first executable code in the edge attribute of the first derived relation edge to generate the data lineage between the business concept entity and the caliber entity.

7. A data reporting method based on multiple calibers, characterized in that, include: The target reporting platform is determined based on the data reporting task; In the semantic lineage graph, determine the target caliber node corresponding to the target reporting platform, and traverse the target business concept nodes associated with the target caliber node in the semantic lineage graph; Extract the executable code stored in the edge attributes of the derived relationship edges connecting the target business concept node and the target caliber node; Source data is extracted from the target business concept node, and the executable code is executed to transform the source data and generate target reporting data; The target reporting data is sent to the target reporting platform.

8. A data reporting device based on multi-caliber transmission, characterized in that, include: The platform determination module is used to determine the target reporting platform based on the data reporting task; The graph query module is used to determine the target caliber node corresponding to the target reporting platform in the semantic lineage graph, and to traverse the target business concept nodes associated with the target caliber node in the semantic lineage graph. The code extraction module is used to extract the executable code between the target business concept node and the target caliber node; The data conversion module is used to extract source data from the target business concept node and execute the executable code to convert the source data and generate target reporting data. The sending module is used to send the target reporting data to the target reporting platform.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the multi-caliber data lineage construction method as described in any one of claims 1 to 5, or the multi-caliber data reporting method as described in claim 7.

10. A non-transitory computer-readable storage medium, wherein a computer program is stored on the non-transitory computer-readable storage medium, characterized in that, When the computer program is executed by the processor, it implements the multi-caliber data lineage construction method as described in any one of claims 1 to 5, or the multi-caliber data reporting method as described in claim 7.