Multi-source heterogeneous enterprise data driven ai service dynamic optimization method and system

Abstract

The application discloses a multi-source heterogeneous enterprise data driven AI service dynamic optimization method and system, and belongs to the field of data processing, wherein the method comprises the following steps: acquiring enterprise data in a plurality of enterprise data sources and executing a data processing chain; acquiring a service task of an AI service to be executed, determining a data object and a target business semantic, and determining a target semantic processing result; determining a semantic processing node from the data processing chain, matching the semantic processing result with the target semantic processing result, determining a reusable semantic processing node and a supplementary processing node; determining a shared processing segment according to a processing sequence, connecting the supplementary processing node to an output end to reconstruct a data processing chain of the AI service to be executed, and executing the AI service to be executed based on the reconstructed data processing chain. The application solves the technical problem that the existing AI service is usually independently constructed to form a data processing link, resulting in a long deployment period, slow response speed and poor data processing consistency.

Description

Technical Field

[0001] This invention relates to the field of data processing, specifically to a method and system for dynamic optimization of AI services driven by multi-source heterogeneous enterprise data. Background Technology

[0002] In the process of enterprise digital transformation, enterprises have accumulated massive amounts of enterprise data distributed across multiple heterogeneous business systems such as ERP (Enterprise Resource Planning), CRM (Customer Relationship Management), MES (Manufacturing Execution System), financial accounting, and OA (Office Automation), as well as various unstructured data carriers. Different data sources have significant differences in data format, storage method, update frequency, and business semantics. At the same time, enterprises have deployed various AI services based on the above data, such as sales forecasting, customer churn warning, product quality inspection, financial risk identification, and intelligent scheduling. However, in existing technologies, various AI services usually build their own dedicated data processing links. Each link needs to complete the entire process of data access, cleaning and transformation, and business semantic parsing from scratch. The semantic processing capabilities between different AI services cannot be shared and reused, which not only causes duplication and waste of development and computing resources, but also makes it difficult to unify the data business semantics of different services. When a new AI service is launched or an existing service is upgraded, a complete data processing link needs to be rebuilt, which has the defects of long deployment cycle, slow business response speed, and poor data processing consistency, making it difficult to quickly adapt to the needs of dynamically changing business scenarios. Summary of the Invention

[0003] This application provides a method and system for dynamic optimization of AI services driven by multi-source heterogeneous enterprise data, aiming to solve the technical problems of existing AI services, which usually build data processing links independently, resulting in long deployment cycles, slow response speeds, and poor data processing consistency.

[0004] In view of the above problems, this application provides a method and system for dynamic optimization of AI services driven by multi-source heterogeneous enterprise data.

[0005] The first aspect disclosed in this application provides a method for dynamic optimization of AI services driven by multi-source heterogeneous enterprise data, the method comprising: The system acquires enterprise data from multiple enterprise data sources and data processing chains corresponding to multiple AI services executed based on the enterprise data; it acquires service tasks for AI services to be executed, determines enterprise data objects to be processed and their corresponding target business semantics based on the service tasks, and determines the target semantic processing results required for the AI ​​services to be executed based on the target business semantics; it identifies semantic processing nodes corresponding to the target business semantics from the data processing chains corresponding to the multiple AI services, matches the semantic processing results formed by the semantic processing nodes with the target semantic processing results to identify reusable semantic processing nodes and supplementary processing nodes; it determines a shared processing segment based on the processing order of the reusable semantic processing nodes in the corresponding data processing chain, connects the supplementary processing nodes to the output of the shared processing segment to reconstruct the data processing chain of the AI ​​services to be executed, and executes the AI ​​services to be executed based on the reconstructed data processing chain.

[0006] Another aspect disclosed in this application provides a dynamic optimization system for AI services driven by multi-source heterogeneous enterprise data, the system comprising: The system comprises the following modules: a data acquisition module for acquiring enterprise data from multiple enterprise data sources and data processing chains corresponding to multiple AI services executed based on the enterprise data; a processing module for acquiring service tasks of AI services to be executed, determining enterprise data objects to be processed and their corresponding target business semantics based on the service tasks, and determining the target semantic processing results required for the AI ​​services to be executed based on the target business semantics; a matching module for determining semantic processing nodes corresponding to the target business semantics from the data processing chains corresponding to the multiple AI services, matching the semantic processing results formed by the semantic processing nodes with the target semantic processing results to determine reusable semantic processing nodes and supplementary processing nodes; and an execution module for determining a shared processing segment based on the processing order of the reusable semantic processing nodes in the corresponding data processing chain, connecting the supplementary processing nodes to the output of the shared processing segment to reconstruct the data processing chain of the AI ​​services to be executed, and executing the AI ​​services to be executed based on the reconstructed data processing chain.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: By standardizing and decomposing the entire data processing flow of existing AI services and marking the semantic processing nodes, and using the target business semantics of the AI ​​service to be executed as a benchmark, the output results of existing semantic processing nodes are matched in multiple dimensions, including data objects, business semantics, data fields, data range, and data status. Reusable semantic processing nodes are identified and integrated into a shared processing segment. Only supplementary processing nodes that adapt to differences in fields, ranges, and statuses are needed to reconstruct the data processing chain of the target AI service. This shortens the deployment and iteration cycle of AI services, reduces development and computing resource consumption, and ensures a unified standard of data business semantics among different AI services, thereby improving the response efficiency and data processing quality of enterprise AI services.

[0008] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0009] Figure 1 A flowchart illustrating the dynamic optimization method for AI services driven by multi-source heterogeneous enterprise data is provided for embodiments of this application. Figure 2 This application provides a schematic diagram of the structure of a dynamic optimization system for AI services driven by multi-source heterogeneous enterprise data.

[0010] Explanation of reference numerals in the attached diagram: Acquisition module 11, Processing module 12, Matching module 13, Execution module 14. Detailed Implementation

[0011] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0012] The overall concept of the technical solution provided in this application is as follows: This application provides a method and system for dynamic optimization of AI services driven by multi-source heterogeneous enterprise data. By disassembling the entire data processing chain of existing AI services and marking semantic processing nodes, reusable semantic processing nodes are selected based on the target business semantic matching of the service to be executed. Adjacent reusable nodes are integrated into a shared processing segment. Only the differences in data fields, data ranges, and data states are addressed by adding corresponding processing nodes. This allows for rapid reconstruction of the target AI service's data processing chain, ultimately achieving cross-service reuse of business semantic processing capabilities. This reduces redundant development work, shortens the deployment cycle of new AI services, and unifies the business semantic processing standards for multi-source heterogeneous data, effectively improving the dynamic adaptability and construction efficiency of the data processing chain.

[0013] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0014] Example 1, as Figure 1 As shown in the embodiments of this application, a method for dynamic optimization of AI services driven by multi-source heterogeneous enterprise data is provided. The method includes: S100: Obtain enterprise data from multiple enterprise data sources, and the data processing chain corresponding to multiple AI services executed based on the enterprise data.

[0015] Specifically, the first step is to acquire enterprise data from multiple enterprise data sources. These multiple enterprise data sources refer to heterogeneous business systems and data carriers distributed across the entire enterprise. Typical examples include structured business databases such as ERP (Enterprise Resource Planning) systems, CRM (Customer Relationship Management) systems, MES (Manufacturing Execution System), financial accounting systems, and OA (Office Automation) systems, as well as semi-structured and unstructured data carriers such as production logs, business documents, and report files. Different data sources have significant differences in data format, storage method, update frequency, and business domain, which is a concrete manifestation of the multi-source heterogeneity characteristic of the solution. The corresponding enterprise data acquisition action is to complete the standardized collection of data across the entire domain through an adaptive data access mechanism: for structured databases, data connections are established through JDBC / ODBC protocols for batch or incremental extraction; for semi-structured files and interface data, timed synchronization is achieved through API gateways and file acquisition components; and for unstructured documents, text parsing and entity extraction tools are used to complete the structuring conversion. At the same time, metadata annotation is completed during the data access process, registering the source system, business object, field meaning, update cycle, and data quality level of each piece of data, forming a unified enterprise data resource catalog.

[0016] Based on this, the acquisition and sorting of data processing chains corresponding to multiple AI services are completed simultaneously. These multiple AI services refer to various artificial intelligence application scenarios that have been deployed and implemented in enterprises and operate based on the aforementioned enterprise data, such as sales volume prediction AI, customer churn warning AI, product quality inspection AI, financial risk identification AI, and human resource intelligent scheduling AI. Each type of AI service needs to process enterprise data in a series of ways around specific business objectives before outputting decision results.

[0017] The data processing chain refers to a series of data processing operations that are sequentially linked according to the data flow and have clear input-output dependencies in the complete process of a single AI service from receiving raw enterprise data to outputting the final AI result. Typical processing nodes include data cleaning, field mapping, data aggregation, business semantic parsing, feature engineering, and model inference.

[0018] The corresponding data processing chain execution action is to standardize and decompose the entire processing flow of existing AI services and to deposit it as an asset: traverse each online AI service one by one, starting from its original data input source, and decompose each data processing operation along the data flow direction. Completely record the processing logic, input data format and content, output data format and content, and upstream and downstream dependencies of each processing operation. At the same time, mark the operation nodes responsible for performing business semantic parsing on the data and outputting semantic processing results with business meaning during the decomposition process. These are the semantic processing nodes used for matching and reuse in subsequent steps. Finally, connect all operation nodes in the order of processing to form the data processing chain corresponding to each AI service, and store the data processing chains of all AI services in a unified processing chain resource library for centralized management and retrieval.

[0019] S200: Obtain the service task of the AI ​​service to be executed, determine the enterprise data object to be processed and the target business semantics corresponding to the enterprise data object to be processed according to the service task, and determine the target semantic processing result required for the AI ​​service to be executed according to the target business semantics.

[0020] Specifically, the first step is to acquire the service tasks for the AI ​​services to be executed. These AI services specifically refer to newly deployed, iteratively upgraded, or temporarily invoked AI application services for specific business scenarios, distinguishing them from the existing AI services already running in the previous steps. This is the core object of this dynamic optimization. The service task is a collection of concrete business goals and execution constraints that the AI ​​service to be executed needs to achieve. It is a complete task definition that includes core elements such as a clear business scenario, output delivery format, data time granularity, and applicable links of the results. For example, monthly regional dealer credit risk rating output and weekly failure probability prediction of key equipment on the production line. The acquisition action here is achieved by parsing the service request work orders submitted by the business department, reading the task parameters in the AI ​​service call request, and decomposing the goal definition in the service deployment specification. At the same time, the business boundaries, time requirements, and output verification rules of the task are sorted out to complete the structured decomposition of the service task.

[0021] Based on this, the action of determining the enterprise data object to be processed and the target business semantics corresponding to the enterprise data object according to the service task is executed. The enterprise data object to be processed is the collection of all enterprise data required to support the operation of this service task, corresponding to all business entity data covered by the service task. For example, the data object corresponding to the dealer credit rating task includes dealer entity file data, historical transaction performance data, financial reconciliation data, legal litigation record data, etc. The implementation logic of this action is to deduce the data requirements in reverse based on the business objectives of the service task, match the dataset of the corresponding business domain from the multi-source enterprise data sources collected in the previous steps, clarify the data subject, time span, statistical granularity and source system, and define the complete data to be processed. The scope; while target business semantics refers to the definition of the business meaning carried by the enterprise data object to be processed in the specific business scenario of the current service task. It is different from the physical meaning of the data fields themselves. It is a unified interpretation of the value of data from a business perspective. For example, the same overdue days field corresponds to the semantics of collection risk in the accounts receivable scenario and the semantics of performance capability in the supplier management scenario. The implementation of this part is to combine the specific business links corresponding to the service task, map the corresponding business meaning to each type of data field and data set in the data object to be processed, upgrade the physical meaning of the data fields to the semantic definition at the business level, and associate and bind all semantic definitions with the business links of the service task to form a unified target business semantic system, ensuring that subsequent data processing always revolves around business semantics.

[0022] Subsequently, the action of determining the target semantic processing result required for the AI ​​service to be executed based on the target business semantics is performed. The target semantic processing result is the final output standard of the AI ​​service to be executed for semantic-level data processing. It is a standardized data form with clear business semantics, meeting the input requirements of subsequent model inference or business computation, and is also the core benchmark for subsequent matching and comparison with existing AI service semantic processing nodes. The specific implementation path of this action is as follows: First, based on the business objectives of the service task, the corresponding business data requirements are decomposed to clarify what semantic dimensions of data support are needed for the inference or computation logic of the AI ​​service to be executed. Then, the corresponding semantic dimensions are clarified one by one according to the target business semantics. The data field list, the legal value range of each field, and the data status requirements, including missing value handling standards, normalization requirements, aggregation granularity, and anonymization levels, are all included. Ultimately, the data fields, value ranges, and data status requirements are associated and bound with the corresponding target business semantics to form a structured and quantifiable target semantic processing result definition. For example, the target semantic processing result for the dealer credit rating task can be defined as a structured dataset with a single dealer as the statistical subject, covering three core business semantics: capital strength, performance credit, and operational capabilities, corresponding to 12 quantitative indicator fields, with values ​​uniformly mapped to a standardized score of 0-100, and the data status being a structured dataset with missing values ​​filled and monthly dimension aggregation.

[0023] S300: Determine the semantic processing node corresponding to the target business semantic from the data processing chain corresponding to the multiple AI services, and match the semantic processing result formed by the semantic processing node with the target semantic processing result to determine the reusable semantic processing node and the supplementary processing node.

[0024] Specifically, the first step is to determine the semantic processing node corresponding to the target business semantics from the data processing chains corresponding to the multiple AI services. The semantic processing node is a specific processing unit embedded in the data processing chains of each existing AI service, responsible for business semantic parsing and transformation, and capable of outputting data results with clear business meaning. It differs from nodes that only perform technical processing such as data cleaning and format conversion; it is the core unit in the data processing chain that carries business value. The determination process here employs a full-domain traversal plus semantic initial screening logic. Using the core business meaning of the target business semantics and the corresponding business links as the retrieval benchmark, the complete data processing chains of all existing AI services are scanned one by one. The functional definition and business attributes of the output data of each processing node in the chain are checked one by one. All processing nodes whose output results carry the same type of business semantics are selected, forming a candidate semantic processing node set. This completes the initial screening of locating business-related core nodes from a massive number of technical processing nodes, filtering out irrelevant purely technical processing nodes.

[0025] Based on this, the action of matching the semantic processing results generated by the semantic processing nodes with the target semantic processing results is performed. The semantic processing results are standardized data forms with complete business semantics that are actually output by the candidate semantic processing nodes in the original AI service operation. They contain five core attributes: the corresponding data object subject, business semantic dimension, covered data fields, field value range, and data status specification. The target semantic processing results are the standardized semantic output benchmarks that the AI ​​service to be executed is required to achieve, output by the preceding steps. The matching here adopts a multi-dimensional item-by-item alignment execution method. The five core attributes of the candidate semantic processing results and the target semantic processing results are extracted respectively: data object, business semantics, data fields, data range, and data status. An item-by-item comparison list is established. With business semantics and data object as the core mandatory matching items and the other three as detailed verification items, the overlap and difference points of the two results in each dimension are verified one by one. A matching relationship graph with each attribute matching or difference status is generated, clearly showing the consistent items and specific deviation positions between the existing processing results and the target requirements.

[0026] Subsequently, a determination action is executed to identify reusable semantic processing nodes and supplementary processing nodes. Reusable semantic processing nodes refer to those whose core business processing logic can be directly reused without redevelopment. Supplementary processing nodes are new processing units used to correct discrepancies between the output of existing nodes and target requirements, filling capability gaps. This determination action is performed in two scenarios based on the aforementioned matching relationship graph: when all five attributes of the candidate semantic processing result—data object, business semantics, data fields, data range, and data status—correspond consistently to the target semantic processing result, the candidate semantic processing node is directly determined as a reusable semantic processing node, and its output can directly meet the input requirements of the target AI service. When the core attributes of the candidate semantic processing result, namely the data object and the business semantics, are consistent with the target, but there are differences in at least one of the data fields, data range, and data status, the node is still determined as a reusable semantic processing node to reuse its core semantic parsing capabilities. At the same time, supplementary processing nodes are generated according to the type of difference. If the difference is a missing data field, a data extraction node is generated; if the difference is a mismatch in the value range, a data filtering node is generated; if the difference is a data status that does not meet the standard, a data status processing node is generated. The input of the supplementary processing node is clearly the output result of the reusable semantic processing node, forming a combination of reusable core capabilities and supplementary difference capabilities.

[0027] S400: Determine a shared processing segment based on the processing order of the reusable semantic processing nodes in the corresponding data processing chain, connect the supplementary processing node to the output of the shared processing segment to reconstruct the data processing chain of the AI ​​service to be executed, and execute the AI ​​service to be executed based on the reconstructed data processing chain.

[0028] Specifically, the first step is to determine the shared processing segment based on the processing order of the reusable semantic processing nodes in their corresponding data processing chains. The processing order refers to the sequential execution order and upstream / downstream input / output dependencies formed by each reusable semantic processing node within its original existing AI service data processing chain, reflecting the inherent logical chain of semantic processing. The shared processing segment is a link fragment with complete semantic processing capabilities, composed of reusable semantic processing nodes with continuous and closely connected processing orders. It is a collective carrier of reusable capabilities, distinct from scattered individual processing nodes. The specific implementation of this action... The current approach is as follows: First, extract the position index and connection relationship of all reusable semantic processing nodes in the original data processing chain. Arrange all reusable semantic processing nodes strictly according to their original processing order, restore the input-output docking logic between nodes, and then select at least two reusable semantic processing nodes that are consecutively adjacent in processing order, have no non-reusable processing nodes inserted in between, and whose output of the previous node can be directly used as the input of the next node. Combine them to determine a shared processing segment, while retaining the processing rules, parameter configuration and execution logic of this segment in the original data processing chain to ensure that the reused processing logic is completely consistent with the original valid logic.

[0029] Based on this, the action of connecting the supplementary processing node to the output end of the shared processing segment to reconstruct the data processing chain of the AI ​​service to be executed is carried out. The reconstruction is a link reorganization performed with the target semantic processing result as the final output standard and following the sequential logic of data flow. The connection adopts the docking logic of output end serialization, that is, the output data of the final output node of the shared processing segment is used as the input data source of the supplementary processing node. According to the logical order of difference correction (first data field extraction and completion, then value range filtering and adaptation, and finally data state standardization adjustment), all supplementary processing nodes are sequentially connected, so that the supplementary processing node takes over the semantic processing result of the shared processing segment and completes the correction and completion of the difference items in a targeted manner. This action will eventually form a complete new data processing chain dedicated to the AI ​​service to be executed. The input end of this chain corresponds to the access interface of the enterprise data object to be processed. The chain contains the shared processing segment and the supplementary processing node in sequence. The output end of the chain directly outputs data that meets the target semantic processing result standard, completing the complete processing logic closed loop from the original enterprise data to the target semantic data.

[0030] Subsequently, the action of executing the AI ​​service to be executed is performed based on the reconstructed data processing chain. The execution flow of this action is as follows: the enterprise data objects to be processed identified in the previous steps are imported from the corresponding data source into the input end of the reconstructed data processing chain. The data flows through each semantic processing node of the shared processing segment to complete the parsing and transformation of the core business semantics, and then flows into the supplementary processing node to complete the difference correction processing such as missing field completion, value range adaptation, and data state standardization. Finally, a standardized semantic dataset that fully meets the requirements of the target semantic processing result is output. Then, the semantic dataset is input into the core service modules of the AI ​​service to be executed, such as subsequent model inference, business calculation, and result output, to drive the AI ​​service to complete the complete business calculation and result output, and finally deliver the AI ​​service result that meets the service task requirements.

[0031] Furthermore, in the method provided in this application embodiment, determining the enterprise data object to be processed and the target business semantics corresponding to the enterprise data object to be processed according to the service task includes: determining the enterprise business object and business process corresponding to the AI ​​service to be executed according to the service task; determining the corresponding enterprise data object to be processed from the multiple enterprise data sources according to the enterprise business object; determining the corresponding business semantic information from the enterprise data object to be processed according to the business process, and determining the business semantic information as the target business semantics.

[0032] Specifically, the first step involves determining the enterprise business objects and business processes corresponding to the AI ​​service to be executed based on the service task. The enterprise business objects are the core business entities targeted by the AI ​​service, independent entities within the business scenario, such as distributors, production equipment, customers, and purchase orders. These are distinct from data objects at the technical level and are the core carriers of business needs. The business processes are the specific process nodes and business scenarios within the enterprise's complete business process. For example, distributor credit rating corresponds to the distributor's ongoing risk control process, and equipment failure prediction corresponds to the predictive maintenance process for production equipment, determining the business interpretation perspective of the data. This determination is achieved through a business architecture mapping method. Based on the decomposed service task objectives, the entity list and process nodes in the enterprise's business architecture are matched in reverse, clarifying the scope of business entities covered by the service and their respective process stages. Simultaneously, business boundaries and responsibilities are defined. For example, for a monthly regional distributor credit risk rating task, the corresponding enterprise business objects are all registered cooperative distributors within the enterprise's jurisdiction, and the corresponding business process is the ongoing credit risk control process under the distributor management system, thus anchoring the core elements at the business level.

[0033] Based on this, the action of determining the corresponding enterprise data object to be processed from the multiple enterprise data sources according to the enterprise business object is executed. The enterprise data object to be processed is the collection of all enterprise data that supports the operation of this AI service and is associated with the established enterprise business object. It covers the related data distributed in different heterogeneous data sources and is the data carrier of the business object. The determination action here uses the business object as the search keyword, traverses the full-domain multi-source enterprise data sources with metadata marking completed in the previous steps, and matches all datasets in the data sources that have a relationship with the business object. For example, around the dealer business object, the dealer basic file dataset is matched from the CRM system, the historical transaction order dataset is matched from the ERP system, the accounts receivable dataset is matched from the financial system, and the litigation record dataset is matched from the legal system. These cross-system and cross-structure related data are integrated and defined as the enterprise data object to be processed. At the same time, the boundary conditions such as the time span, statistical granularity and update frequency of the data are defined to complete the accurate mapping from business entities to data scope, ensuring that the data scope completely matches the business needs and is without redundancy.

[0034] Subsequently, the action of determining the corresponding business semantic information from the enterprise data object to be processed based on the business process is executed, and the business semantic information is determined as the target business semantic. Here, business semantic information is the business meaning and value definition carried by the data to be processed under a specific business process, which is different from the physical attribute meaning of the data field itself. The same data field corresponds to different business semantics under different business processes. The determination action here combines the business rules, assessment logic and management objectives of the business process, and interprets and classifies the meaning of all data fields and data sets in the enterprise data object to be processed from a business perspective. For example, in the dealer credit risk control process, the registered capital field is interpreted as capital strength semantic, the historical overdue payment number field is interpreted as performance credit level semantic, and the annual purchase amount field is interpreted as cooperation scale semantic. Then, all the interpreted business meanings are integrated, uniformly named and defined by boundaries to form a complete set of business semantic information, and finally determined as the target business semantic, completing the upgrade and transformation from data physical attributes to business semantic.

[0035] Furthermore, in the method provided in this application embodiment, the corresponding business semantic information is determined from the enterprise data object to be processed according to the business process. The enterprise data object to be processed includes related data objects of the same enterprise business object in different enterprise data sources; the business fields and field values ​​corresponding to each related data object are obtained; the business meaning corresponding to the business fields and field values ​​is determined according to the business process corresponding to each related data object; the business meaning is associated with the corresponding business process to obtain the business semantic information corresponding to each related data object.

[0036] Specifically, the first step is to clarify the constituent characteristics of the enterprise data objects to be processed, namely, they include related data objects of the same enterprise business object in different enterprise data sources. The related data objects are subsets of data that point to the same core business entity but are scattered and stored in different heterogeneous enterprise data sources. For example, for the same distributor business object, the distributor file data stored in the CRM system, the transaction order data in the ERP system, the payment collection data in the financial system, and the litigation record data in the legal system are all related data objects of the same business object. Different related data objects have typical heterogeneous differences in data structure, field naming, storage format, and update frequency. This definition clarifies that the object of semantic parsing is a heterogeneous dataset with cross-source association, which clarifies the processing boundaries and scope for subsequent cross-source semantic alignment.

[0037] Based on this, the action of obtaining the business fields and field values ​​corresponding to each of the aforementioned related data objects is performed. The business fields are the specific data attribute items that carry business information in the data source to which each related data object belongs. Fields carrying the same business meaning in different data sources often have differences in naming, data type, and storage format. The field values ​​are the specific data content and built-in value rules corresponding to the business fields, including various forms such as numerical range, enumeration type, date format, and encoding rules. The acquisition here first establishes a cross-data source entity mapping relationship through the globally unique identifier of the enterprise business object, including internal entity code, unified social credit code, etc., to accurately anchor the related data objects in different data sources to the same business entity. Then, the full business field metadata of each related data object is extracted one by one, covering attributes such as field name, data type, length constraint, and value logic. At the same time, typical value samples of the corresponding fields are extracted, and finally a cross-source business field list labeled with the data source to which it belongs is formed, completing the full inventory of fields of heterogeneous data sources.

[0038] Next, the action of determining the business fields and their corresponding business meanings based on the business processes corresponding to each associated data object is executed. The business meaning is a unified definition of the field's value from a business management perspective, independent of the data source's technical naming rules. It is an abstract description of the business role and management value carried by the field. The same business meaning can correspond to multiple heterogeneous technical fields from different data sources. This determination adopts a business-caliber mapping method, using the established business rules, management objectives, and assessment criteria of the current business process as a unified standard. Each field in the cross-source business field list is semantically interpreted from a business perspective, while simultaneously... The business logic of the combined field values ​​maps scattered technical fields to a unified business meaning. For example, in the dealer credit risk control business process, the registered subscribed amount field in the CRM system and the paid-in capital balance field in the financial system are uniformly mapped to the business meaning of capital strength. The order overdue days field in the ERP system and the overdue payment duration field in the financial system are uniformly mapped to the business meaning of performance credit risk. At the same time, the business meaning corresponding to the field value is clearly defined. For example, the larger the value, the higher the risk level. This completes the essential transformation from technical field attributes to business value meaning and unifies the semantics of cross-source heterogeneous fields.

[0039] The following action is then performed to associate the business meaning with the corresponding business process to obtain the business semantic information corresponding to each associated data object. The business semantic information is a set of semantic units bound to a specific business scenario and possessing unified business interpretation rules. It is the final form of cross-source heterogeneous data after semantic standardization and can be directly used for subsequent semantic matching and processing reuse. The association here adopts a dimension aggregation binding method, which classifies and aggregates all mapped business meanings according to the management dimensions of the corresponding business process. At the same time, semantic alignment rules for fields with the same business meaning in different associated data objects are established, clarifying the conversion, completion, and aggregation logic between fields of different data sources. Then, the aggregated set of business meanings is strongly bound to the corresponding business process to form the complete business semantic information corresponding to each set of associated data objects, realizing the unified expression of cross-source heterogeneous data at the business semantic level.

[0040] Furthermore, in the method provided in this application embodiment, determining the target semantic processing result required for the AI ​​service to be executed based on the target business semantics includes: determining the business data requirements corresponding to the AI ​​service to be executed under the target business semantics based on the service task; determining the data fields corresponding to the target business semantics and the data range of each data field based on the business data requirements, and determining the data status requirements corresponding to each data field; associating the data fields, the data range, and the data status requirements with the target business semantics to generate the target semantic processing result corresponding to the AI ​​service to be executed.

[0041] Specifically, the first step is to determine the business data requirements corresponding to the AI ​​service to be executed under the target business semantics based on the service task. These business data requirements are a set of business information dimensions necessary for the AI ​​service to complete its core calculations and outputs, from a business function perspective. They differ from technical field requirements and serve as the carrier for the demand docking between business semantics and AI service capabilities. This determination employs a task-reverse derivation method. Based on the final output target of the service task and the core processing logic of the AI ​​service, the various business information dimensions required to support the AI ​​service's computation, decision-making, and output are decomposed in reverse. These information dimensions are then mapped and matched one by one with the defined target business semantic dimensions. The core essential business semantic dimensions and auxiliary supplementary dimensions are selected, clarifying the business role, priority weight, and coverage boundaries of each requirement dimension. Finally, a structured list of business data requirements is formed, achieving precise docking from the service task to business semantic requirements and avoiding redundancy or omissions in subsequent data standards.

[0042] Based on this, the following actions are performed: determining the data fields corresponding to the target business semantics and the data range of each data field according to the business data requirements; and determining the data status requirements corresponding to each data field. Here, a data field is the specific attribute carrier that translates the business data requirements into the data layer; it is the concrete representation of the target business semantics at the data layer, corresponding to standardized data attribute items uniformly defined after cross-source association. The data range refers to the value boundaries and statistical constraints corresponding to each data field, including multiple boundary conditions such as legal value intervals, enumerated value sets, statistical time spans, aggregation granularity, and statistical frequency. The data status requirements are the quality standards and format specifications that the data fields must meet, covering specific requirements such as missing value allowance rate, outlier handling rules, data format standards, numerical normalization methods, and information anonymization levels.

[0043] The determination process here is carried out simultaneously in three layers: The first layer decomposes and maps each semantic dimension of the business data requirements into several specific standardized data fields, clarifying the business meaning of the fields and their correspondence with cross-data source sources; The second layer, in conjunction with business rules, task timeliness, and management standards, defines the data range of each field one by one, clarifying constraints such as upper and lower limits of values, time span, and aggregation granularity; The third layer, in conjunction with the model input specifications and data quality requirements of the AI ​​service to be executed, clarifies the data status requirements of each field one by one, and formulates specific rules for missing value filling, outlier removal, format conversion, and standardization processing, completing the layer-by-layer decomposition from business requirements to data technology standards.

[0044] The following action involves associating the data fields, data ranges, and data status requirements with the target business semantics to generate the target semantic processing result corresponding to the AI ​​service to be executed. This association is a hierarchical binding action with the target business semantics as the top-level dimension. By establishing a four-level mapping relationship—target business semantic dimension, corresponding data field, field data range, and field status requirements—dispersed technical standards are attached to the corresponding business semantics, forming a unified benchmark with a hierarchical structure. The target semantic processing result is the standardized benchmark output at the end of this step. It is a semantic processing output specification that can be directly used for comparison and verification, fully defining the business semantics that the semantic processing output needs to cover, the included data fields, value boundaries, and quality standards. The generation action here is implemented through structured definition, solidifying and encapsulating the four-level mapping relationship, clarifying the verification rules and matching judgment logic for each standard, and ultimately forming a comparable target semantic processing result benchmark.

[0045] Furthermore, in the method provided in this application embodiment, matching the semantic processing result formed by the semantic processing node with the target semantic processing result to determine reusable semantic processing nodes and supplementary processing nodes includes: determining the data object, business semantics, data field, data range, and data status corresponding to the semantic processing result and the target semantic processing result respectively; performing item-by-item mapping on the data object, the business semantics, the data field, the data range, and the data status to generate a matching relationship between the semantic processing result and the target semantic processing result; and determining the reusable semantic processing node and the supplementary processing node based on the matching relationship.

[0046] Specifically, the process first involves determining the data objects, business semantics, data fields, data range, and data status corresponding to the semantic processing result and the target semantic processing result. The semantic processing result is a standardized data product with business semantics, actually output by candidate semantic processing nodes in the existing AI service. The target semantic processing result is the semantic processing output benchmark of the AI ​​service to be executed, defined in the preceding steps. These two are the main components of this matching comparison. The data object is the core business entity scope covered by the semantic result; the business semantics is the business meaning and value proposition carried by the data result; the data fields are the specific data attribute items included in the semantic result; and the data range is the range of each field. The five dimensions—value boundaries, statistical granularity, and time span constraints, data status (which serves as the data quality standard and format specification)—constitute the complete attribute framework of the semantic result. The determination of actions here employs a unified dimensional decomposition method. For the output of each candidate semantic processing node, by parsing the node's processing rules, output metadata, and runtime logs, the corresponding data object boundaries, core business semantic definitions, output field list, field value range, and data status standards are extracted and structured. Simultaneously, the target semantic processing result is structurally decomposed according to the same five dimensions, forming an attribute list with unified dimensional standards for both parties, establishing a unified benchmarking framework for subsequent item-by-item comparisons.

[0047] Based on this, an action is performed to match the data object, business semantics, data fields, data range, and data status item by item, generating a matching relationship between the semantic processing result and the target semantic processing result. This item-by-item matching is a layer-by-layer comparison conducted with a priority order of core dimensions first, followed by detailed dimensions. Data objects and business semantics are core dimensions that determine the essence of semantics, while data fields, data range, and data status are secondary dimensions that affect output details. The matching action here is performed using a matrix-style benchmarking verification method. Using the five attributes of the target semantic processing result as the benchmark value, each candidate semantic processing result's five attributes are verified against the benchmark value one by one. The core dimensions use a complete consistency judgment rule, while the detailed dimensions use a difference refinement judgment rule. The comparison result for each dimension is marked as one of three states: complete match, partial difference, or complete mismatch. Simultaneously, the specific difference content and deviation degree of some difference items are recorded, ultimately forming a matching relationship map annotated with the full-dimensional matching status and difference details, clearly presenting the fit and specific deviation points between each candidate node and the target benchmark.

[0048] Subsequently, the action of determining the reusable semantic processing nodes and the supplementary processing nodes based on the matching relationship is executed. The reusable semantic processing nodes are existing nodes whose core semantic processing logic can be directly reused without redevelopment. The supplementary processing nodes are new processing units used to correct differences and fill capability gaps. The determination action here is based on the matching relationship graph and classified by scenario: when all five dimensions are marked as a complete match, it means that the output result of the node fully meets the target requirements, and it is directly determined as a reusable semantic processing node; when the two core dimensions of data object and business semantics are completely matched, and only one or more of the data fields, data range, and data status have partial differences, it means that the core semantic parsing capability of the node is reusable, and only the output details have deviations. Therefore, it is still determined as a reusable semantic processing node. At the same time, according to the type of difference and the specific deviation content, the direction of the processing capability that needs to be supplemented is clearly defined; when the core dimensions do not match, the node is determined to be unreusable and is directly excluded from the candidate range, and finally a clear list of reusable nodes and a list of capabilities to be supplemented are formed.

[0049] Furthermore, in the method provided in this application embodiment, determining the reusable semantic processing node and the supplementary processing node based on the matching relationship includes: when the data object, business semantics, data field, data range, and data state in the semantic processing result are consistent with the corresponding data item in the target semantic processing result, the corresponding semantic processing node is determined as the reusable semantic processing node; when the data object and business semantics in the semantic processing result are consistent with the corresponding data item in the target semantic processing result, and at least one of the data field, the data range, and the data state is inconsistent with the corresponding data item in the target semantic processing result, the corresponding semantic processing node is determined as the reusable semantic processing node, and the supplementary processing node is determined based on the inconsistent data field, data range, or data state.

[0050] Specifically, firstly, corresponding to the first type of judgment rule, the judgment scenario is when the data object, business semantics, data fields, data range, and data status in the semantic processing result are consistent with the corresponding data items in the target semantic processing result. Here, consistency with the corresponding data items means that the candidate semantic processing result completely matches the corresponding benchmark items of the target semantic processing result in five dimensions: the scope of business entities covered by the data object, the definition of the core business semantics, the complete list of output data fields, the boundaries and statistical granularity of field values, and the standard specifications of data quality and form. There are no differences such as range deviation, missing fields, out-of-bounds values, or substandard states. The judgment execution logic here is to iterate through the data objects in the target semantic processing result. Based on the aforementioned matching relationship graph, the matching status markers of each candidate node across five dimensions are verified one by one. When all five dimensions are marked as a complete match, the corresponding semantic processing node is identified as the reusable semantic processing node. This action simultaneously completes the reuse and authorization of the node's capabilities, clarifying that the core processing logic, parameter configuration, and execution rules of the node can be directly reused without modification. Its output semantic processing results can directly meet the input requirements of the AI ​​service to be executed without any additional pre- or post-processing. At the same time, the node is included in the list of reusable node components, and its output interface standards and processing performance parameters are marked, providing standardized components that can be directly called for subsequent processing chain reconstruction.

[0051] The second type of judgment rule is then applied, which is the judgment scenario where the data object and business semantics in the semantic processing result are consistent with the corresponding data items in the target semantic processing result, and at least one of the data field, the data range, and the data state is inconsistent with the corresponding data item in the target semantic processing result. Among them, the consistency of the data object and the business semantics is the core judgment premise, which means that the core business entity coverage and core semantic parsing capability of the candidate node fully match the target requirements. This is the core basis for the reusability of the node. The differences in data field, data range, and data state are at the level of output details and do not affect the reusability of the core semantic processing logic. The inconsistency of at least one covers a variety of deviation scenarios such as single-dimensional differences and multi-dimensional combination differences.

[0052] The judgment and execution are divided into two steps. The first step is to determine the corresponding semantic processing node as the reusable semantic processing node, that is, to retain the core semantic processing capabilities of the node, confirm that its core processing link can be fully reused, and not to negate the overall reusability value of the node due to differences in details, thus avoiding the repeated development of core semantic parsing capabilities. The second step is to simultaneously determine the supplementary processing node based on inconsistent data fields, data ranges, or data states. The determination based on the type of inconsistency is a differentiated node derivation logic. If the inconsistency is a data field, a corresponding supplementary processing node of data extraction type is deduced to complete the missing target field or trim redundant non-target fields, thereby realizing the field list. The system ensures precise adaptation. If the inconsistency is a data range, corresponding supplementary processing nodes for data filtering or aggregation are generated to adjust the value range, convergence statistical granularity, or matching time span, achieving precise alignment of data boundaries. If the inconsistency is a data state, corresponding supplementary processing nodes for data state processing are generated to complete state corrections such as missing value filling, outlier management, format standardization, or numerical normalization, achieving precise compliance with data quality standards. At the same time, it is clear that the input interfaces of all supplementary processing nodes are connected to the output interfaces of the aforementioned reusable semantic processing nodes, forming a combination mode of core reuse and post-completion capabilities, ensuring that the corrected output results fully match the requirements of the target semantic processing results.

[0053] Furthermore, in the method provided in this application embodiment, the supplementary processing node is determined based on inconsistent data fields, data ranges, or data states. The method includes: determining a corresponding data extraction node based on inconsistent data fields, determining a corresponding data filtering node based on inconsistent data ranges, and determining a corresponding data state processing node based on inconsistent data states; determining at least one of the data extraction node, the data filtering node, and the data state processing node as the supplementary processing node, and using the semantic processing result output by the reusable semantic processing node as the input of the supplementary processing node.

[0054] Specifically, the process first executes node-oriented derivation actions corresponding to three types of differences. The first type involves determining the corresponding data extraction node based on inconsistent data fields. Inconsistent data fields refer to deviations in the field dimension between the output fields of the reusable semantic processing node and the field requirements of the target semantic processing result, such as missing fields, redundant fields, mismatched field names, or inconsistent field mapping relationships. These are detailed differences in the attribute dimension of the semantic output. The data extraction node is a processing unit specifically responsible for field-level adaptation processing. Its core capabilities are field filtering, field completion, field renaming, and cross-field mapping conversion. The determination action here is driven by field difference analysis. First, the field lists of both parties are compared item by item to clarify whether the difference type is missing, redundant, or mapping deviation. Then, the processing rules of the node are configured accordingly: if the field is missing, field completion rules are configured to extract or derive the target field from the associated attributes of the output result of the reusable node; if the field is redundant, field filtering rules are configured to retain only the attributes in the target field list; if the field name or mapping deviation is a problem, field mapping rules are configured to achieve unified alignment between field names and semantics. Finally, a field-level adaptation node with clear processing rules is generated to achieve accurate matching at the field dimension.

[0055] The second category involves determining the corresponding data filtering node based on inconsistent data ranges. Inconsistent data ranges refer to deviations between the boundary conditions (value range, time span, statistical granularity, business dimension coverage, etc.) of the reusable semantic processing node's output data and the target requirements. This represents a difference in the semantic output's range across boundary dimensions. The data filtering node is a dedicated processing unit responsible for data range calibration, with core capabilities including range filtering, granularity aggregation, dimension splitting, and time slicing. The determination process here is driven by range difference analysis. First, it identifies the type of deviation: too wide a range, too narrow a range, too coarse granularity, or dimension mismatch. Then, it configures the node's processing logic accordingly: if the range is too wide, it configures filtering conditions to remove data exceeding the target value range or business boundary; if the range is too narrow, it configures range expansion rules to supplement data within the boundary range based on existing data; if the statistical granularity is mismatched, it configures aggregation or splitting rules to summarize or decompose data according to the target granularity; if the time span is mismatched, it configures time slicing rules to filter data within the target time range. Finally, it generates a range adaptation node with clear boundary calibration rules, achieving precise alignment of data boundaries.

[0056] The third category involves determining the corresponding data status processing node based on inconsistent data states. Inconsistent data states refer to discrepancies between the quality standards and format specifications of the output data from reusable semantic processing nodes and the target requirements. These discrepancies encompass deviations in various status dimensions, such as excessive missing values, unmanaged outliers, inconsistent data formats, unnormalized values, and lack of data anonymization. Data status processing nodes are dedicated processing units responsible for data quality governance and format standardization. Their core capabilities include missing value imputation, outlier correction, format conversion, numerical normalization, and information anonymization. The determination process here is driven by status difference analysis. First, deviations in the data state are identified one by one. Then, the processing algorithms for the nodes are configured accordingly: for missing values, mean, median, or business rule imputation logic is configured; for outliers, removal, correction, or marking rules are configured; for format differences, unified format conversion rules are configured; for numerical magnitude differences, standardized normalization algorithms are configured; and for anonymization requirements, data anonymization rules are configured. Ultimately, a status calibration node with clear quality control rules is generated, achieving accurate compliance with data status standards.

[0057] Based on this, the action of identifying at least one of the data extraction node, the data filtering node, and the data status processing node as the supplementary processing node, and using the semantic processing result output by the reusable semantic processing node as the input of the supplementary processing node, is executed. At least one of these corresponds to a combination of differences in the actual scenario. If only a single-dimensional difference exists, the corresponding single-type node is identified as the supplementary processing node. If there are multi-dimensional combinations of differences, the nodes corresponding to multiple types are linked together to form a complete supplementary processing link, constituting a set of supplementary processing nodes, following the inherent data processing logic sequence of field extraction - range filtering - status processing. Here, "as input" clarifies the upstream and downstream connection between the supplementary processing node and the reusable semantic processing node. That is, the supplementary processing node does not directly access the original enterprise data; its only input source is the semantic processing result output by the reusable semantic processing node. All difference corrections are post-processed based on data that has already undergone core semantic parsing, ensuring the complete reuse of core semantic processing capabilities and avoiding resource consumption caused by repeated calls to the original data during supplementary processing.

[0058] Furthermore, in the method provided in this application embodiment, the method for determining a shared processing segment based on the processing order of the reusable semantic processing nodes in the corresponding data processing chain includes: arranging the reusable semantic processing nodes according to their processing order in the corresponding data processing chain; and determining at least two reusable semantic processing nodes with consecutive and adjacent processing orders as the shared processing segment.

[0059] Specifically, the process begins by arranging the reusable semantic processing nodes according to their processing order within the corresponding data processing chain. The processing order refers to the sequential execution order of each reusable semantic processing node within its original existing AI service data processing chain, determined by the data flow direction and upstream / downstream input / output dependencies. This reflects the inherent logical chain of semantic processing; changes in the node order directly lead to confusion in the semantic processing logic and result deviations. This arrangement employs the original chain restoration method. First, the position index, preceding dependent nodes, and subsequent output node information of each reusable semantic processing node in the original data processing chain are extracted one by one. Strictly following the data flow direction from original input to semantic output, the sequential arrangement of all reusable semantic processing nodes is restored. During the arrangement process, the original input / output docking rules, parameter passing logic, and processing connections between nodes are fully preserved. The execution order of any node is not adjusted, ensuring that the arranged node sequence completely reproduces the processing logic in the original chain, avoiding disruption of the continuity and accuracy of semantic processing due to disordered order.

[0060] Based on this, the action of determining at least two reusable semantic processing nodes that are consecutive and adjacent in processing order as the shared processing segment is performed. The consecutive and adjacent processing order is the core determination condition of the shared processing segment, which means that in the arranged reusable node sequence, there are no non-reusable processing nodes, additional data processing operations or link interruptions between nodes in the original data processing chain. The output data of the previous node can be directly used as the input data of the next node. The processing logic between nodes is closely connected and there are no additional intermediate links.

[0061] At least two criteria were defined for the minimum size of a shared processing segment. A single reusable node can only be reused as an independent component and cannot form a link-level processing capability. Only when two or more consecutively adjacent nodes are connected in series can a link segment with complete stage-based semantic processing capability be formed. The determination action here adopts the continuous segment recognition method, which traverses the arranged reusable node sequence, scans and identifies all node groups that meet the condition of consecutive adjacency, and delineates two or more consecutively adjacent reusable semantic processing nodes in each group as a shared processing segment. At the same time, a unified input interface, output interface and internal execution rules are encapsulated for each shared processing segment, integrating the originally scattered multiple nodes into a standardized processing unit that can be called as a whole and run independently, and finally forming several sets of shared processing segments with different granularities.

[0062] Furthermore, in the method provided in this application embodiment, the method for obtaining the data processing chain corresponding to multiple AI services executed by the enterprise data includes: obtaining the data processing operations sequentially executed on the enterprise data object during the execution of each AI service, as well as the input data and output data corresponding to each data processing operation; determining the processing order and connection relationship between each data processing operation based on the correspondence between the input data and output data between adjacent data processing operations; determining the multiple data processing operations connected according to the processing order as the data processing chain of the corresponding AI service, and determining the data processing operation that performs business semantic parsing on the enterprise data object and outputs semantic processing results as the semantic processing node in the data processing chain.

[0063] Specifically, the process first involves acquiring the data processing operations sequentially performed on the enterprise data objects during the execution of each AI service, as well as the input and output data corresponding to each data processing operation. Each data processing operation is the smallest independent execution unit constituting the AI ​​service data processing flow. Each operation possesses a single, clearly defined processing function, encompassing various technical and business processing actions such as data cleaning, field mapping, data aggregation, format conversion, business semantic parsing, feature engineering, and model inference. Input and output data define the boundaries of each data processing operation, corresponding to the data form before execution and the output data form after execution, respectively, including data structures and field groups. The data acquisition process employs a full-process reverse decomposition method, starting with the final output of each existing AI service and tracing the entire data flow path backward. By analyzing the AI ​​service's process configuration file, operator orchestration rules, execution logs, interface call relationships, and data lineage records, each independent data processing operation from the initial enterprise data access to the final service output is decomposed. Simultaneously, the source of input data, input data structure, output data form, and output data destination corresponding to each operation are extracted. This completes a full inventory and boundary definition of all processing units in the entire AI service process, avoiding any omission of intermediate processing links.

[0064] Based on this, the action of determining the processing order and connection relationship between each data processing operation is performed according to the correspondence between the input and output data of adjacent data processing operations. The correspondence between the input and output data of adjacent data processing operations refers to the data dependency relationship between upstream and downstream operations, that is, whether the output data of the preceding operation is used as the input data of the subsequent operation, which is the core basis for deriving the processing link. The processing order is the order in which each data processing operation is executed in the direction of data flow. The connection relationship is the topological structure between each operation, including various forms such as serial connection, branching, and multi-input convergence. The determination action here is implemented by the data lineage matching method. The input and output data characteristics of all data processing operations are used as the matching basis to match the output and input correspondence between adjacent operations one by one, verifying the consistency of data structure, field semantics, and data granularity. Based on this, the upstream and downstream dependency logic between all operations is derived. According to the flow direction of data from the original input to the final output, the processing order of each operation is determined. At the same time, the connection forms such as serial connection, branching, and convergence between operations are clarified, and finally a complete processing operation topology diagram is formed, clarifying the position of each operation in the link and the upstream and downstream connection nodes.

[0065] Subsequently, the data processing chain for the corresponding AI service is determined by connecting multiple data processing operations in the processing order. The data processing operation that performs business semantic parsing on enterprise data objects and outputs semantic processing results is determined as the action of the semantic processing node in the data processing chain. The data processing chain is a complete link with complete data processing capabilities formed by connecting multiple data processing operations in the processing order and connection relationship. It corresponds to the entire process logic of a single AI service from data access to result output and is a link-level asset that can be reused as a whole or in segments. Semantic processing nodes are core operational units with special functional positioning in the data processing chain. Unlike ordinary operations that only perform technical data processing, their core function is to perform semantic parsing and transformation of enterprise data objects from a business perspective, and output semantic processing results with clear business meaning. They are the core nodes that carry business value.

[0066] The process involves two steps. First, all data processing operations connected in the processing order are encapsulated as a whole and defined as the data processing chain for the corresponding AI service. The input interface, output interface, and overall processing logic of the chain are clearly defined. Second, all data processing operations on the chain are functionally classified and attribute-labeled. Based on whether business semantic parsing is performed and whether results with business meaning are output, all processing operations that perform business semantic parsing are identified and labeled as semantic processing nodes. The business semantic type, corresponding business link, and output semantic result features of each semantic processing node are simultaneously labeled, thus completing the construction of a structured data processing chain with core node labels.

[0067] In summary, the multi-source heterogeneous enterprise data-driven AI service dynamic optimization method provided in this application has the following technical effects: By standardizing and decomposing the entire data processing flow of existing AI services and marking the semantic processing nodes, and using the target business semantics of the AI ​​service to be executed as a benchmark, the output results of existing semantic processing nodes are matched in multiple dimensions, including data objects, business semantics, data fields, data range, and data status. Reusable semantic processing nodes are identified and integrated into a shared processing segment. Only supplementary processing nodes that adapt to differences in fields, ranges, and statuses are needed to reconstruct the data processing chain of the target AI service. This shortens the deployment and iteration cycle of AI services, reduces development and computing resource consumption, and ensures a unified standard of data business semantics among different AI services, thereby improving the response efficiency and data processing quality of enterprise AI services.

[0068] Example 2, based on the same inventive concept as the multi-source heterogeneous enterprise data-driven AI service dynamic optimization method in the foregoing examples, such as... Figure 2 As shown in the embodiment of this application, an AI service dynamic optimization system driven by multi-source heterogeneous enterprise data is provided. The system includes: The acquisition module 11 is used to acquire enterprise data from multiple enterprise data sources, and data processing chains corresponding to multiple AI services executed based on the enterprise data; the processing module 12 is used to acquire the service tasks of the AI ​​services to be executed, determine the enterprise data objects to be processed and the target business semantics corresponding to the enterprise data objects to be processed according to the service tasks, and determine the target semantic processing results required for the AI ​​services to be executed according to the target business semantics; the matching module 13 is used to determine the semantic processing nodes corresponding to the target business semantics from the data processing chains corresponding to the multiple AI services, match the semantic processing results formed by the semantic processing nodes with the target semantic processing results to determine reusable semantic processing nodes and supplementary processing nodes; the execution module 14 is used to determine a shared processing segment according to the processing order of the reusable semantic processing nodes in the corresponding data processing chain, connect the supplementary processing nodes to the output end of the shared processing segment to reconstruct the data processing chain of the AI ​​services to be executed, and execute the AI ​​services to be executed based on the reconstructed data processing chain.

[0069] Furthermore, the processing module 12 is also used to perform the following steps: determine the enterprise business object and business process corresponding to the AI ​​service to be executed according to the service task; determine the corresponding enterprise data object to be processed from the multiple enterprise data sources according to the enterprise business object; determine the corresponding business semantic information from the enterprise data object to be processed according to the business process, and determine the business semantic information as the target business semantic.

[0070] Furthermore, the processing module 12 is also used to perform the following steps: determining the corresponding business semantic information from the enterprise data object to be processed according to the business link, wherein the enterprise data object to be processed includes related data objects of the same enterprise business object in different enterprise data sources; obtaining the business fields and field values ​​corresponding to each of the related data objects; determining the business meaning corresponding to the business fields and field values ​​according to the business link corresponding to each of the related data objects, and associating the business meaning with the corresponding business link to obtain the business semantic information corresponding to each of the related data objects.

[0071] Furthermore, the processing module 12 is also used to perform the following steps: determining the business data requirements corresponding to the AI ​​service to be executed under the target business semantics according to the service task; determining the data fields corresponding to the target business semantics and the data range of each data field according to the business data requirements, and determining the data status requirements corresponding to each data field; associating the data fields, the data range and the data status requirements with the target business semantics to generate the target semantics processing result corresponding to the AI ​​service to be executed.

[0072] Furthermore, the matching module 13 is also used to perform the following steps: determining the data object, business semantics, data field, data range, and data status corresponding to the semantic processing result and the target semantic processing result respectively; performing item-by-item mapping of the data object, the business semantics, the data field, the data range, and the data status to generate a matching relationship between the semantic processing result and the target semantic processing result; and determining the reusable semantic processing node and the supplementary processing node based on the matching relationship.

[0073] Furthermore, the matching module 13 is also used to perform the following steps: when the data object, business semantics, data field, data range, and data status in the semantic processing result are consistent with the corresponding data item in the target semantic processing result, the corresponding semantic processing node is determined as the reusable semantic processing node; when the data object and business semantics in the semantic processing result are consistent with the corresponding data item in the target semantic processing result, and at least one of the data field, the data range, and the data status is inconsistent with the corresponding data item in the target semantic processing result, the corresponding semantic processing node is determined as the reusable semantic processing node, and the supplementary processing node is determined according to the inconsistent data field, data range, or data status.

[0074] Furthermore, the matching module 13 is also used to perform the following steps: determining the corresponding data extraction node based on the inconsistent data field, determining the corresponding data filtering node based on the inconsistent data range, and determining the corresponding data state processing node based on the inconsistent data state; determining at least one of the data extraction node, the data filtering node, and the data state processing node as the supplementary processing node, and using the semantic processing result output by the reusable semantic processing node as the input of the supplementary processing node.

[0075] Furthermore, the execution module 14 is also used to perform the following steps: arranging the reusable semantic processing nodes according to their processing order in the corresponding data processing chain; and determining at least two reusable semantic processing nodes with consecutive and adjacent processing orders as the shared processing segment.

[0076] Furthermore, the execution module 14 is also used to perform the following steps: obtaining the data processing operations sequentially executed on the enterprise data object during the execution of each AI service, as well as the input data and output data corresponding to each data processing operation; determining the processing order and connection relationship between each data processing operation based on the correspondence between the input data and output data between adjacent data processing operations; determining multiple data processing operations connected according to the processing order as the data processing chain of the corresponding AI service, and determining the data processing operation that performs business semantic parsing on the enterprise data object and outputs semantic processing results as the semantic processing node in the data processing chain.

[0077] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for dynamic optimization of AI services driven by multi-source heterogeneous enterprise data, characterized in that, The method includes: Acquire enterprise data from multiple enterprise data sources, and the data processing chain corresponding to multiple AI services executed based on the enterprise data; Obtain the service task of the AI ​​service to be executed, determine the enterprise data object to be processed and the target business semantics corresponding to the enterprise data object to be processed based on the service task, and determine the target semantic processing result required for the AI ​​service to be executed based on the target business semantics; From the data processing chains corresponding to the multiple AI services, determine the semantic processing nodes corresponding to the target business semantics, and match the semantic processing results formed by the semantic processing nodes with the target semantic processing results to determine reusable semantic processing nodes and supplementary processing nodes. The shared processing segment is determined according to the processing order of the reusable semantic processing nodes in the corresponding data processing chain. The supplementary processing node is connected to the output of the shared processing segment to reconstruct the data processing chain of the AI ​​service to be executed, and the AI ​​service to be executed is executed based on the reconstructed data processing chain.

2. The method as described in claim 1, characterized in that, The method for determining the enterprise data object to be processed and the target business semantics corresponding to the enterprise data object based on the service task includes: The enterprise business object and business process corresponding to the AI ​​service to be executed are determined based on the service task. Based on the enterprise business object, the corresponding enterprise data object to be processed is determined from the multiple enterprise data sources; Based on the business process, the corresponding business semantic information is determined from the enterprise data object to be processed, and the business semantic information is determined as the target business semantic.

3. The method as described in claim 2, characterized in that, Based on the business process, the corresponding business semantic information is determined from the enterprise data object to be processed. The enterprise data object to be processed includes related data objects of the same enterprise business object in different enterprise data sources. Obtain the business fields and field values ​​corresponding to each of the associated data objects; Based on the business process corresponding to each of the associated data objects, the business fields and the business meanings corresponding to their values ​​are determined, and the business meanings are associated with the corresponding business processes to obtain the business semantic information corresponding to each of the associated data objects.

4. The method as described in claim 1, characterized in that, The method for determining the target semantic processing result required for the AI ​​service to be executed based on the target business semantics includes: Based on the service task, determine the business data requirements corresponding to the AI ​​service to be executed under the target business semantics; Based on the business data requirements, determine the data fields corresponding to the target business semantics and the data range of each data field, and determine the data status requirements corresponding to each data field; The data fields, data range, and data status requirements are associated with the target business semantics to generate the target semantic processing result corresponding to the AI ​​service to be executed.

5. The method as described in claim 4, characterized in that, The semantic processing results generated by the semantic processing nodes are matched with the target semantic processing results to determine reusable semantic processing nodes and supplementary processing nodes, including: Determine the data object, business semantics, data fields, data range, and data status corresponding to the semantic processing result and the target semantic processing result, respectively; The data object, the business semantics, the data fields, the data range, and the data status are matched item by item to generate a matching relationship between the semantic processing result and the target semantic processing result; The reusable semantic processing node and the supplementary processing node are determined based on the matching relationship.

6. The method as described in claim 5, characterized in that, Determining the reusable semantic processing node and the supplementary processing node based on the matching relationship includes: When the data object, business semantics, data field, data range, and data status in the semantic processing result are consistent with the corresponding data item in the target semantic processing result, the corresponding semantic processing node is determined as the reusable semantic processing node. When the data object and business semantics in the semantic processing result are consistent with the corresponding data item in the target semantic processing result, and at least one of the data field, the data range, and the data state is inconsistent with the corresponding data item in the target semantic processing result, the corresponding semantic processing node is determined as the reusable semantic processing node, and the supplementary processing node is determined according to the inconsistent data field, data range, or data state.

7. The method as described in claim 6, characterized in that, The method for determining the supplementary processing node based on inconsistent data fields, data ranges, or data states includes: The corresponding data extraction node is determined based on the inconsistent data field, the corresponding data filtering node is determined based on the inconsistent data range, and the corresponding data status processing node is determined based on the inconsistent data status. At least one of the data extraction node, the data filtering node, and the data state processing node is identified as the supplementary processing node, and the semantic processing result output by the reusable semantic processing node is used as the input of the supplementary processing node.

8. The method as described in claim 1, characterized in that, The method for determining a shared processing segment based on the processing order of the reusable semantic processing nodes in the corresponding data processing chain includes: The reusable semantic processing nodes are arranged according to their processing order in the corresponding data processing chain. The shared processing segment is defined as at least two reusable semantic processing nodes that are consecutive and adjacent in processing order.

9. The method as described in claim 1, characterized in that, The method for obtaining the data processing chain corresponding to multiple AI services executed by the enterprise data includes: Obtain the data processing operations sequentially performed on the enterprise data object during the execution of each AI service, as well as the input and output data corresponding to each data processing operation; Based on the correspondence between input and output data between adjacent data processing operations, the processing order and connection relationship between each data processing operation are determined; Multiple data processing operations connected in the processing order are identified as the data processing chain of the corresponding AI service, and the data processing operations that perform business semantic parsing on enterprise data objects and output semantic processing results are identified as the semantic processing nodes in the data processing chain.

10. A dynamic optimization system for AI services driven by multi-source heterogeneous enterprise data, characterized in that: The system is used to implement the multi-source heterogeneous enterprise data-driven AI service dynamic optimization method according to any one of claims 1 to 9, wherein the system comprises: The acquisition module is used to acquire enterprise data from multiple enterprise data sources, as well as the data processing chain corresponding to multiple AI services executed based on the enterprise data; The processing module is used to obtain the service task of the AI ​​service to be executed, determine the enterprise data object to be processed and the target business semantics corresponding to the enterprise data object to be processed according to the service task, and determine the target semantic processing result required for the AI ​​service to be executed according to the target business semantics. The matching module is used to determine the semantic processing node corresponding to the target business semantic from the data processing chain corresponding to the multiple AI services, and match the semantic processing result formed by the semantic processing node with the target semantic processing result to determine the reusable semantic processing node and the supplementary processing node. The execution module is used to determine a shared processing segment based on the processing order of the reusable semantic processing nodes in the corresponding data processing chain, connect the supplementary processing node to the output of the shared processing segment to reconstruct the data processing chain of the AI ​​service to be executed, and execute the AI ​​service to be executed based on the reconstructed data processing chain.

Images