CN-121981617-A - Index semantic layer construction and full life cycle management method

CN121981617ACN 121981617 ACN121981617 ACN 121981617ACN-121981617-A

Abstract

The invention discloses an index semantic layer construction and full life cycle management method, and belongs to the field of data management and enterprise data asset management. The method comprises the steps of completing index standardization definition and core parameter configuration through constructing a three-layer index semantic model, generating index metadata and calculation logic, executing index duplicate removal detection through combining a rule matching scoring model and a semantic similarity judging model, outputting detection results and risk grade dividing results, completing index lifecycle state management and standardization approval process execution based on the detection results, completing full-process index dependency verification through constructing an index dependency graph, and achieving index lifecycle state update and version archiving. The method can realize standardization unification of the index caliber and full-flow compliance control, reduce redundant content of an index library, reduce index management and maintenance cost, and pre-identify the index change influence range, thereby providing complete technical support for standardized construction of an enterprise-level index system.

Inventors

Wen Maoyuan
WU HAO
LENG YUJIE
HUANG LANG

Assignees

猪哥云(四川)数字科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. The index semantic layer construction and full life cycle management method is characterized by comprising the following steps of: S1, constructing a three-layer index semantic model, completing index standardization definition and core parameter configuration based on the three-layer index semantic model, and generating index metadata and index calculation logic; S2, extracting multi-dimensional characteristic information based on index metadata and index calculation logic generated by the three-layer index semantic model, and performing index duplicate removal detection through a rule matching scoring model and a semantic similarity judgment model to generate an index duplicate removal detection result and a risk grade division result; s3, based on the index duplicate removal detection result, completing life cycle state assignment and compliance conversion by combining index information corresponding to the three-layer index semantic model, inputting the index information into a standardized approval process to execute approval operation, and generating an approval result; s4, constructing an index dependency relationship map based on index metadata corresponding to the three-layer index semantic model and the approval result, executing full-flow index dependency verification based on the index dependency relationship map, and finishing the final life cycle state updating and version archiving of the index according to the verification result.
2. The method according to claim 1, wherein step S1 specifically comprises: Setting an atomic index module, a derived index module and a composite index module in the constructed three-layer index semantic model, wherein the atomic index module is a base layer, the derived index module is a middle layer and directly depends on the atomic index module, and the composite index module is an upper layer and directly depends on the atomic index module or the derived index module; S1.1, configuring business process mapping parameters, measurement mode aggregation parameters and data source association parameters of an atomic index module of a three-layer index semantic model, setting threshold ranges of the business process mapping parameters, the measurement mode aggregation parameters and the data source association parameters, carrying out standardized definition on the atomic index module according to association rules of the business process and the measurement mode, and outputting index metadata of the atomic index through the atomic index module; S1.2, configuring time limiting parameters, service limiting parameters and basic index association parameters of a derived index module of a three-layer index semantic model, carrying out standardized definition on the derived index module based on index metadata output by an atomic index module, superposing standardized time limiting conditions and service limiting conditions, restricting the basic definition content of the atomic index module not to be modified, and outputting the index metadata of the derived index through the derived index module; S1.3, configuring operation rule parameters, index reference parameters and precision retention parameters of a composite index module of the three-layer index semantic model, carrying out standardized definition on the composite index module through a standardized four-rule operation expression based on index metadata output by the atomic index module or index metadata output by the derivative index module, and outputting index metadata and index calculation logic of a composite index through the composite index module.
3. The method according to claim 1, wherein step S2 specifically comprises: S2.1, constructing a rule matching scoring model, setting a feature extraction module, a weight distribution module, a scoring calculation module and a candidate screening module by the rule matching scoring model, wherein the output end of the feature extraction module of the rule matching scoring model is directly connected with the input end of the weight distribution module of the rule matching scoring model, the output end of the weight distribution module of the rule matching scoring model is directly connected with the input end of the scoring calculation module of the rule matching scoring model, the output end of the scoring calculation module of the rule matching scoring model is directly connected with the input end of the candidate screening module of the rule matching scoring model, the feature extraction module of the rule matching scoring model, the weight distribution module of the rule matching scoring model, the scoring calculation module of the rule matching scoring model, the feature dimension, the feature duty ratio, the matching threshold and the result number parameter of the candidate screening module of the rule matching scoring model are input into the rule matching scoring model, the feature information is subjected to multidimensional rule matching scoring, the candidate index with the similarity reaching a set value is screened, and the rough recall processing of index de-duplication detection is completed; S2.2, a semantic similarity judgment model is constructed, an input layer, an encoding layer, a feature fusion layer, a similarity calculation layer and a result grading layer are arranged in the semantic similarity judgment model, an input layer output end of the semantic similarity judgment model is directly abutted to the encoding layer input end of the semantic similarity judgment model, an encoding layer output end of the semantic similarity judgment model is directly abutted to a feature fusion layer input end of the semantic similarity judgment model, a feature fusion layer output end of the semantic similarity judgment model is directly abutted to a similarity calculation layer input end of the semantic similarity judgment model, a similarity calculation layer output end of the semantic similarity judgment model is directly abutted to the result grading layer input end of the semantic similarity judgment model, batch sizes, iteration times and learning rate parameters trained by the semantic similarity judgment model are configured, candidate indexes and new indexes are input into the semantic similarity judgment model, the semantic similarity judgment model is input through consistency in a business process, consistency of measurement objects, time limit equivalence, business limit difference attribute and calculation logic equivalence dimension execute semantic similarity judgment, and an index duplicate removal detection result and risk grade division result are generated.
4. The method according to claim 1, wherein step S3 specifically comprises: S3.1, assigning the initial life cycle state of the defined index as a draft state, checking the prepositive setting condition of index submission approval based on the index duplicate removal detection result, converting the index life cycle state into an approval state after the verification is passed, and executing the locking operation of the index basic definition content; S3.2, inputting index information in the approval state into a standardized approval process, executing decision operation of approval nodes, converting the index life cycle state into an online state when the approval passes based on the final decision result of all the approval nodes, and converting the index life cycle state into a draft state when the approval fails.
5. The method according to claim 1, wherein step S4 specifically comprises: S4.1, analyzing the index calculation logic to extract the reference relation among the indexes, constructing an index dependency relation map in the form of a directed acyclic graph, configuring node identification parameters, side association parameters and map updating parameters of the index dependency relation map, and executing updating operation of corresponding content of the index dependency relation map according to the index reference relation; S4.2, sequentially executing index cyclic dependency detection, pre-online dependency integrity check and pre-offline influence range analysis operation based on an index dependency relationship map, configuring a detection range, result judgment and result output parameters of a verification process, and completing index dependency check of a whole process through the index cyclic dependency detection, the pre-online dependency integrity check and the pre-offline influence range analysis operation to generate an index dependency check result; S4.3, based on the index dependence checking result and the approval result, executing the final updating operation of the index on-line state or off-line state, and executing the full content archiving operation of the corresponding version of the index according to the version management rule.
6. The method according to claim 1, wherein in step S1, the index metadata includes basic information, attribution information, technical information and version information, corresponding mandatory fields and standardized verification rules are set for the basic information, attribution information, technical information and version information, respectively, field matching parameters, content compliance parameters and verification result feedback parameters of the standardized verification rules are configured, the basic information includes an index unique identifier, an index display name, an index standard name and a service caliber description, the attribution information includes a service domain, an affiliated department, a service responsible person and a technical responsible person, the technical information includes data sources, calculation logic, dependency relationships and update frequencies, and the version information includes a version number, creation time and a modification record.
7. The method of claim 1, wherein in step S2, the multidimensional feature information includes an index standard name, an index display name, a service caliber, a belonging field, a time limit, a service limit, and a calculation logic, wherein corresponding weight duty ratios are respectively set for the index standard name, the index display name, the service caliber, the belonging field, the time limit, the service limit, and the calculation logic, standardized preprocessing operations are respectively executed for the index standard name, the index display name, the service caliber, the belonging field, the time limit, the service limit, and the calculation logic, standardized preprocessed text format parameters, expression specification parameters, and data cleaning parameters are configured, unified operations of feature information formats and expression specifications are executed, and the processed feature information is input into a rule matching scoring model to execute the calculation operations.
8. The method of claim 1, wherein in step S3, a standardized approval process sets a service approval node, a technical approval node and a data administration approval node, a node circulation parameter, a decision result parameter and a process triggering parameter of the standardized approval process are configured, a countersign rule or a countersign and conditional branch circulation rule is set for the standardized approval process, the countersign rule is that all approval nodes can pass through the rear side and can circulate, or the countersign rule is that any approval node can pass through, the conditional branch rule is that the approval nodes are matched according to service scenes, the service approval node executes accuracy and necessity verification operation of a service caliber, the technical approval node executes correctness and compliance verification operation of calculation logic, the data administration node executes integrity and normalization verification operation of metadata, and the matching setting operation of the circulation rule is executed according to index type and influence range.
9. The method according to claim 1, wherein in step S4, when the index dependency check result is failed, a blocking operation of an index lifecycle state update operation is performed, a corresponding influence range report is output, index identification parameters, influence link parameters and content display parameters of the influence range report are configured, when the loop dependency check result is failed, an output operation of a loop-referenced index unique identification and a reference link is performed, when the on-line dependency check result is failed, an output operation of an unfinished on-line dependency index unique identification and a corresponding lifecycle state is performed, and when the off-line influence analysis result is failed, an output operation of an affected downstream index unique identification and a corresponding service scene is performed.
10. The method according to claim 4, wherein in step S3.1, when performing an editing operation on an index in an online state, a new independent index version generating operation is performed, a version number generating parameter, a content synchronizing parameter and a version associating parameter of the index version are configured, an initial lifecycle state of the newly generated index version is assigned as a draft state, a state of the index version in the original online state is kept unchanged until the newly generated index version completes all standardized approval processes and is converted into the online state, a smooth switching operation of the version is performed, an execution time parameter and a downstream adapting parameter of the smooth switching are configured, a marking operation of a history version is performed on the original index version, and a full content archiving operation of the history version is performed.

Description

Index semantic layer construction and full life cycle management method Technical Field The invention relates to the field of data management and enterprise data asset management, in particular to an index semantic layer construction and full life cycle management method. Background Along with the continuous promotion of the enterprise digital transformation process, the data asset becomes a core support element for enterprise operation decision, service optimization and risk management and control, and the index is used as a key carrier for landing the data value, is a core bridge for converting the bottom data into the landing service insight by the enterprise, and is widely and deeply applied to various service scenes such as daily operation, strategic planning, performance management and control of the enterprise. Currently, enterprises in various industries gradually develop the construction work of an enterprise-level unified index system, the index management related technology also continuously develops and iterates, and related technologies such as metadata management, data blood-margin analysis, natural language processing and the like are gradually integrated into the whole flow of index management, so that the concept of index whole life cycle management is widely accepted and practiced in the industry. The technical scheme of index management at the present stage is gradually extended from traditional index calculation, storage and visual display to the directions of index standardization definition, full-flow compliance management and control, intelligent operation and maintenance and the like, and various index management architectures and implementation schemes adapting to different business scenes are also developed in the industry aiming at the management requirements of a large-scale index library, so that diversified technical paths are provided for standardized construction of an enterprise index system. In the current practice process of index management technology, a series of technical pain points to be solved still exist, firstly, a unified and standardized index semantic layer construction framework is lacking, the index definition under different service scenes lacks a layered and standardized unified rule, index metadata and calculation logic with consistent caliber and unified logic are difficult to form, and stable and unified basic support cannot be provided for the whole flow management and control of the indexes. Secondly, the prior art scheme lacks a systematic index repetition detection mechanism, can not finish the pre-identification and hierarchical control of repeated risks in an index creation link, easily causes a large number of redundant repeated indexes in an enterprise index library, increases the maintenance cost of index management, and easily causes the problem of disordered business statistical caliber. Meanwhile, the existing scheme cannot form a closed-loop index full life cycle management and control flow, the index lacks a standardized check mechanism and a compliance approval flow from creation to archiving, and the standardization and compliance of the index full flow are difficult to guarantee. In addition, the prior art lacks complete analysis, visualization and full-flow verification capability on the dependency relationship among indexes, cannot construct a clear and complete index dependency relationship link, is difficult to finish the pre-identification of the influence range before the index state is changed, and cannot realize standardized archiving and full-flow traceability management of the index version. Disclosure of Invention The invention aims to overcome the defects of the prior art and provides an index semantic layer construction and full life cycle management method. The aim of the invention is realized by the following technical scheme: The utility model provides an index semantic layer construction and full life cycle management method, which comprises the following steps: S1, constructing a three-layer index semantic model, completing index standardization definition and core parameter configuration based on the three-layer index semantic model, and generating index metadata and index calculation logic; S2, extracting multi-dimensional characteristic information based on index metadata and index calculation logic generated by the three-layer index semantic model, and performing index duplicate removal detection through a rule matching scoring model and a semantic similarity judgment model to generate an index duplicate removal detection result and a risk grade division result; s3, based on the index duplicate removal detection result, completing life cycle state assignment and compliance conversion by combining index information corresponding to the three-layer index semantic model, inputting the index information into a standardized approval process to execute approval operation, and generating an approval result; s4, constructing