Methods, systems, and storage media for the automatic generation and evolution of enterprise executable ontology
By extracting candidate ontology elements from multi-source enterprise observation data, constructing evidence graphs and verifying them, and generating and evolving enterprise executable ontology, the shortcomings of existing enterprise ontology construction are addressed, and efficient and reliable enterprise ontology management and secure execution of intelligent agents are achieved.
Patent Information
- Application Number
- CN202611635645.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-22
- Publication Date
- 2026-11-17
AI Technical Summary
Existing enterprise ontology construction methods rely on static data sources, which makes it difficult to reflect the actual operating status of enterprise business, lacks modeling of business actions, lacks credibility and auditability in the generation process, lacks incremental update mechanisms, makes it difficult to learn from operational failures, and makes it difficult to automatically convert interfaces of intelligent agent tools, resulting in high development costs and great governance difficulties.
By extracting candidate ontology elements from multi-source enterprise observation data, constructing evidence graphs, generating ontology change patches, and performing conflict detection, permission verification, and sandbox verification, the ontology continuously evolves based on runtime data, enabling intelligent agent querying and business execution.
It reduces the cost of building and maintaining enterprise ontology, improves the credibility and auditability of the generated ontology, reduces the security risks of intelligent agent execution, and enables enterprise AI systems to continuously learn and evolve.
Smart Images

Figure CN123114537A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of artificial intelligence, enterprise data management and business process automation, and in particular to a method, system and storage medium for the automatic generation and evolution of an enterprise executable ontology. Background Technology
[0002] As enterprises become increasingly digitalized, they typically deploy customer relationship management (CRM) systems, enterprise resource planning (ERP) systems, financial systems, billing systems, customer service systems, data warehouses, business intelligence (BI) systems, workflow systems, approval systems, access control systems, API gateways, and various internal application systems. These systems accumulate a large amount of business objects, business relationships, business rules, and business operation records.
[0003] In recent years, enterprises have begun to introduce large language models and intelligent agent systems, enabling them to perform tasks such as information retrieval, business analysis, report generation, work order processing, data entry, approval assistance, and process automation based on natural language. To ensure that intelligent agents can accurately understand enterprise semantics and securely execute enterprise operations, the system typically needs to construct an enterprise knowledge graph, enterprise ontology, or unified semantic layer. Existing methods for constructing enterprise ontology or knowledge graphs mainly include manual modeling, database schema mapping, document extraction, automatic knowledge graph construction, and semantic extraction based on large language models. While these technologies can structure objects, fields, relationships, and some rules within an enterprise to a certain extent, they still have significant shortcomings in the actual operational scenarios of enterprise AI agents.
[0004] First, existing ontology construction methods primarily rely on static data sources. Traditional solutions typically generate ontology from database structures, table fields, business documents, or expert definitions. However, the true semantics of an enterprise are often reflected in the operational processes such as API calls, user operations, approval workflows, workflows, query logs, task failures, and user error correction. Ontologies generated solely from static data sources cannot accurately reflect the real operational status of an enterprise's business in a timely manner.
[0005] Second, existing ontology models are mostly used to describe objects, attributes, and relationships, lacking modeling of business actions. Enterprise AI agents not only need to answer questions but also perform operations, such as creating orders, approving refunds, initiating purchases, changing customer status, generating invoices, or closing work orders. Traditional ontology models typically cannot fully describe action parameters, preconditions, post-effects, permission requirements, approval conditions, audit fields, and rollback actions, thus making it difficult to support intelligent agents in securely executing enterprise actions.
[0006] Third, directly using large language models to generate or modify ontology presents challenges in terms of credibility and auditability. Large language models may generate ontology elements based on incomplete or conflicting information, and it is difficult to clearly explain the data source from which each ontology element originates, what evidence supports it, whether it conflicts with existing ontology, or whether it will affect existing applications or tasks.
[0007] Fourth, existing ontology updates lack incremental, verifiable, and rollback mechanisms. Once an enterprise ontology is used by multiple application systems, query systems, and intelligent agents, any modification may disrupt existing queries, tool calls, permission policies, or business processes. Existing solutions typically lack ontology change patching, sandbox verification, historical task playback, and version rollback mechanisms.
[0008] Fifth, existing semantic layers struggle to learn continuously from operational failures. When user queries fail, intelligent agent tool calls fail, or business process execution fails, traditional systems typically only return error results without identifying the cause of the failure as an ontology gap and further addressing objects, attributes, relationships, actions, constraints, or permission rules.
[0009] Sixth, there is a lack of automatic conversion mechanisms between existing ontology and intelligent agent tool interfaces. Even if enterprises have built knowledge graphs or ontology, engineers often need to manually write APIs, SDKs, tool schemas, permission verification, approval logic, and audit logic, resulting in high development costs, long deployment cycles, and significant governance difficulties.
[0010] Therefore, the market urgently needs a technical solution that can automatically generate enterprise ontology from multi-source enterprise observation data and continuously evolve the ontology based on user dialogue, user error correction, operation logs, task failures, and manual review results during enterprise operation. Summary of the Invention
[0011] In view of the above problems, the present invention is proposed to provide a method, system and storage medium for the automatic generation and evolution of an enterprise executable ontology that overcomes or at least partially overcomes the above problems.
[0012] This invention provides a method for the automatic generation and evolution of an enterprise executable ontology, the method comprising: S1. Extract candidate ontology elements from multi-source enterprise observation data; S2. Use the enterprise observation data of the extracted candidate ontology elements as the source evidence for the corresponding candidate ontology elements, and construct an evidence graph between the candidate ontology elements and the source evidence. S3. Generate an ontology change patch for the enterprise executable ontology based on the candidate ontology elements and their evidence graph; S4. Perform conflict detection, permission verification, constraint verification, and impact scope analysis on ontology change patches to generate risk scores; S5. Sandbox verification of ontology change patches that meet the preset low-risk scoring threshold. S6. Release an ontology change patch that has passed sandbox verification to generate a new enterprise executable ontology version, so that the intelligent agent can perform query tasks, inference tasks and execute enterprise business actions based on the new enterprise executable ontology version. S7. Collect the running data and task failure records during the operation of the intelligent agent, write the error correction records in the running data into the evidence graph as new evidence nodes, and generate candidate ontology change patches for error correction based on the new evidence graph. Determine the ontology gap type according to the failure reasons in the task failure records. Extract candidate ontology elements from the multi-source enterprise observation data based on the ontology gap type and construct the evidence graph to generate candidate ontology change patches for repairing ontology gaps. S8. Use the original failed task or a similar task to perform sandbox verification on the candidate ontology change patch for error correction and the candidate ontology change patch for fixing ontology gaps. S9. Release candidate ontology change patches that have passed sandbox verification to generate new enterprise executable ontology versions.
[0013] Another aspect of the present invention provides an automatic generation and evolution system for an enterprise executable ontology, the system comprising functional modules for implementing the automatic generation and evolution method for an enterprise executable ontology as described above, including: The ontology element extraction module is used to extract candidate ontology elements from multi-source enterprise observation data. The evidence graph construction module is used to construct an evidence graph between candidate ontology elements and source evidence by taking enterprise observation data of extracted candidate ontology elements as source evidence for the corresponding candidate ontology elements. The patch generation module is used to generate ontology change patches for an enterprise executable ontology based on candidate ontology elements and their evidence graphs. The risk assessment module is used to perform conflict detection, permission verification, constraint verification, and impact scope analysis on ontology change patches to generate a risk score. The sandbox verification module is used to perform sandbox verification on ontology change patches that meet the preset low-risk scoring threshold. The release module is used to release ontology change patches that have passed sandbox verification to generate a new enterprise executable ontology version, so that the intelligent agent can perform query tasks, inference tasks, and execute enterprise business actions based on the new enterprise executable ontology version. The evolution module is used to collect operational data and task failure records during the operation of the intelligent agent. It writes the error correction records in the operational data into the evidence graph as new evidence nodes and generates candidate ontology change patches for error correction based on the new evidence graph. It determines the ontology gap type according to the failure reasons in the task failure records. Based on the ontology gap type, it extracts candidate ontology elements from multi-source enterprise observation data and constructs the evidence graph to generate candidate ontology change patches for repairing ontology gaps. The sandbox verification module is also used to perform sandbox verification on candidate ontology change patches for error correction and candidate ontology change patches for fixing ontology gaps using the original failed task or a similar task. The release module is also used to release candidate ontology change patches that have passed sandbox validation to generate new enterprise executable ontology versions.
[0014] In another aspect, the present invention provides a computer storage medium storing a computer program that, when executed, implements the method for the automatic generation and evolution of an enterprise executable ontology as described above.
[0015] In another aspect, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, it implements the automatic generation and evolution method of the enterprise executable ontology as described above.
[0016] This invention provides a method, system, and storage medium for the automatic generation and evolution of an executable enterprise ontology. It acquires multi-source enterprise observation data, extracts candidate ontology elements from this data, constructs an evidence graph between the candidate ontology elements and source evidence, and generates ontology change patches based on this graph. Subsequently, conflict detection, permission verification, constraint verification, impact analysis, and sandbox task replay are performed on the ontology change patches. Based on the verification results, candidate ontology change patches that meet a preset low-risk scoring threshold and pass sandbox verification are released, generating a new executable enterprise ontology version. This allows intelligent agents to perform query tasks, inference tasks, and execute enterprise business actions based on this new version. The invention collects operational data and task failure records during the intelligent agent's operation and uses these as feedback input for the next round of ontology element extraction, evidence graph updates, and ontology change patch generation, thus achieving the automatic generation and evolution of the executable enterprise ontology. This invention reduces the cost of enterprise ontology construction and maintenance, improves the credibility and auditability of ontology generation, reduces the security risks of intelligent agents executing enterprise operations, and enables enterprise AI systems to continuously learn and evolve from operational feedback.
[0017] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description
[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. In the drawings: Figure 1 This is a flowchart illustrating an automatic generation and evolution method for an enterprise executable ontology according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the specific implementation of a method for automatically generating and evolving an enterprise executable ontology, according to another embodiment of the present invention. Figure 3 This is a structural block diagram of an automatic generation and evolution system for an enterprise executable ontology according to an embodiment of the present invention. Detailed Implementation
[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
[0020] It is understood that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art and should not be interpreted in an idealized or overly formal sense unless specifically defined.
[0021] In this embodiment of the invention, enterprise ontology refers to a structured semantic model used to describe enterprise business objects, object attributes, object relationships, business actions, business constraints, permission rules, approval rules, audit rules, rollback rules, and intelligent agent callable interfaces. Enterprise ontology may include at least one of object type, attribute type, relationship type, action type, constraint type, strategy type, function type, tool interface definition, application interface definition, and software development interface definition.
[0022] An executable ontology is an enterprise ontology that describes not only the business objects, attributes, and relationships existing within an enterprise, but also the business actions, action parameters, preconditions, post-effects, permission conditions, approval conditions, audit fields, rollback actions, and risk control rules that can be executed by application systems or intelligent agents. Compared to traditional knowledge graphs or static ontology, executable ontology not only answers "what exists in the enterprise," but also "what actions can be performed on a given object, who can perform them, under what conditions they can be performed, what the effects of execution are, how to audit them, and how to roll back them."
[0023] Multi-source enterprise observation data refers to internal or external data sources used to infer, generate, verify, and update the enterprise ontology, including but not limited to database structure, data table fields, sample data, primary and foreign key relationships, data lineage, business documents, operation manuals, standard operating procedures, product documents, user dialogues, user error correction, API call logs, workflow logs, approval workflow logs, application operation logs, query logs, reports, permission policies, historical task execution records, and intelligent agent task failure records.
[0024] Candidate ontology elements refer to objects, attributes, relationships, actions, constraints, rules, strategies, or interface definitions extracted by the system from multi-source enterprise observation data that may be added to or used to modify the enterprise ontology. Candidate ontology elements may include candidate object types, candidate attribute types, candidate relationship types, candidate action types, candidate constraint rules, candidate access control rules, candidate approval rules, candidate audit fields, candidate rollback actions, and candidate tool interface definitions.
[0025] An evidence graph is a graph structure connecting candidate ontology elements with their source evidence. Nodes in the evidence graph can include candidate ontology element nodes, data source nodes, document fragment nodes, log event nodes, user dialogue nodes, permission policy nodes, historical task nodes, and failure record nodes. Edges in the evidence graph can represent extraction sources, semantic support, conflict relationships, co-occurrence relationships, lineage relationships, permission dependencies, action dependencies, or verification relationships. The evidence graph is used to calculate the confidence level, conflict degree, scope of influence, and review priority of candidate ontology elements.
[0026] An Ontology Delta refers to a patch to an ontology change, not the entire ontology. An Ontology Delta can include the change type, the target ontology element, the content to be added, modified, or deleted, the source of evidence, the confidence level, the conflict detection results, the scope of impact, the verification results, whether manual review is required, the release status, and rollback information.
[0027] Operational feedback refers to the feedback information generated during the operation of an enterprise's AI Agent, application system, query system, or business process, including user error correction, query failure, tool call failure, permission verification failure, API call exception, workflow execution exception, historical task execution results, user confirmation results, and manual review results.
[0028] The intelligent agent tool interface refers to the structured interface definition that can be called by AI Agent or other intelligent systems, including tool name, function description, input parameters, output results, preconditions, post-effects, permission requirements, approval requirements, risk level, dry-run support, rollback operation and audit fields.
[0029] Cross-system concept alignment refers to the process of identifying, comparing, and unifying business concepts with identical, similar, hierarchical, role-based, or conflicting relationships across multiple enterprise systems. For example, Account in a Customer Relationship Management System, Customer in an Enterprise Resource Planning System, Payer in a Billing System, Organization in a Customer Service System, and Legal Entity in a Contract System may represent the same business entity, the same business entity's role in different scenarios, or different levels of business concepts. Cross-system concept alignment can be based on at least one of the following: field name, field description, field value overlap, primary / foreign key relationships, data lineage, query co-occurrence, report co-occurrence, API call co-occurrence, business action co-occurrence, permission policy co-occurrence, user dialogue co-occurrence, and business document description co-occurrence.
[0030] Ontology constraints are structured rules written into an enterprise ontology and used to restrict rules for objects, attributes, relationships, actions, or permissions. Ontology constraints can include at least one of the following: uniqueness constraints, non-null constraints, enumeration value constraints, cardinality constraints, lifecycle state machine constraints, time sequence constraints, monetary range constraints, permission constraints, approval constraints, data quality constraints, action preconditions, and action postconditions.
[0031] like Figure 1 As shown, the automatic generation and evolution method of enterprise executable ontology proposed in this invention includes the following steps: S1. Extract candidate ontology elements from multi-source enterprise observation data. Candidate ontology elements include at least one of the following: object type, attribute type, relation type, action type, constraint rule, and access control rule.
[0032] In this embodiment, the system can access multiple enterprise information sources. These enterprise information sources include at least two of the following: databases, data warehouses, business document libraries, knowledge bases, work order systems, customer relationship management systems, enterprise resource planning systems, billing systems, financial systems, customer service systems, workflow systems, approval systems, permission systems, API gateways, application log systems, and intelligent agent operation log systems. This invention obtains multi-source enterprise observation data from these enterprise information sources. The enterprise observation data includes at least one of the following: database structure, data table fields, field types, field value samples, primary and foreign key relationships, index information, data lineage, document paragraphs, process descriptions, user dialogues, user error corrections, API requests, API responses, workflow nodes, approval records, status transition records, permission policies, query logs, historical task execution records, and intelligent agent failure records. The system generates a data source identifier, timestamp, source system identifier, access permission identifier, and traceable location information for each piece of observation data, so that candidate ontology elements can be subsequently bound to source evidence.
[0033] S2. Use the enterprise observation data of the extracted candidate ontology elements as the source evidence for the corresponding candidate ontology elements, and construct an evidence graph between the candidate ontology elements and the source evidence.
[0034] S3. Generate an ontology change patch for the enterprise executable ontology based on the candidate ontology elements and their evidence graph.
[0035] S4. Perform conflict detection, permission verification, constraint verification, and impact scope analysis on ontology change patches to generate risk scores.
[0036] S5. Sandbox verification of ontology change patches that meet the preset low-risk scoring threshold.
[0037] S6. Release an ontology change patch that has passed sandbox verification to generate a new enterprise executable ontology version, enabling the intelligent agent to perform query tasks, inference tasks, and execute enterprise business actions based on the new enterprise executable ontology version.
[0038] S7. Collect the running data and task failure records during the operation of the intelligent agent, write the error correction records in the running data into the evidence graph as new evidence nodes, and generate candidate ontology change patches based on the new evidence graph. Determine the ontology gap type according to the failure reasons in the task failure records. Extract candidate ontology elements from the multi-source enterprise observation data based on the ontology gap type and construct the evidence graph to generate candidate ontology change patches for repairing ontology gaps.
[0039] In this embodiment, subsequent operational feedback can be used as new multi-source enterprise observation data for the next round of enterprise ontology evolution. Specifically, the system continuously collects operational data and task failure records during the intelligent agent's operation. Operational data includes user dialogues, user corrections, query logs, tool call logs, permission verification results, approval results, API execution results, and workflow execution results. When a user corrects the intelligent agent's understanding of concepts, fields, relationships, actions, or permissions, the system writes this correction record as a new evidence node into the evidence graph and generates candidate ontology change patches for correction. When an intelligent agent task fails, the system categorizes the reasons for failure, determines the corresponding ontology gap type, retrieves relevant enterprise observation data based on the ontology gap type, extracts candidate ontology elements, constructs an evidence graph, and generates candidate ontology change patches to repair the ontology gap. The ontology gap types include missing objects, missing attributes, missing relationships, missing actions, missing constraints, missing permissions, missing tool interfaces, and conceptual ambiguities.
[0040] S8. Use the original failed task or a similar task to perform sandbox verification on the candidate ontology change patch for error correction and the candidate ontology change patch for fixing ontology gaps.
[0041] S9. Release candidate ontology change patches that have passed sandbox verification to generate new enterprise executable ontology versions.
[0042] Specifically, the system uses the original failed task or a similar task to perform sandbox verification on the candidate ontology change patch. If the verification passes, it proceeds to the release or manual review process; if the verification fails, it remains in the candidate status or requests manual supplementation of information.
[0043] This invention provides a method for the automatic generation and evolution of an executable enterprise ontology. It acquires multi-source enterprise observation data, extracts candidate ontology elements from this data, constructs an evidence graph between the candidate ontology elements and source evidence, and generates ontology change patches based on this graph. Subsequently, conflict detection, permission verification, constraint verification, impact analysis, and sandbox task replay are performed on the ontology change patches. Based on the verification results, candidate ontology change patches that meet a preset low-risk scoring threshold and pass sandbox verification are released, generating a new executable enterprise ontology version. This allows intelligent agents to perform query tasks, inference tasks, and execute enterprise business actions based on this new version. The method collects operational data and task failure records during the intelligent agent's operation and uses these as feedback input for the next round of ontology element extraction, evidence graph updates, and ontology change patch generation, thus achieving the automatic generation and evolution of the executable enterprise ontology. This invention reduces the cost of enterprise ontology construction and maintenance, improves the credibility and auditability of ontology generation, reduces the security risks of intelligent agents executing enterprise operations, and enables enterprise AI systems to continuously learn and evolve from operational feedback.
[0044] In this embodiment of the invention, step S1, which involves extracting candidate ontology elements from multi-source enterprise observation data, specifically includes the following steps (not shown in the accompanying drawings): S11. Perform structured data parsing on enterprise observation data of various data types in the multi-source enterprise observation data, and perform semantic normalization on the parsing results to transform the parsing results into intermediate semantic representations.
[0045] Specifically, the system selects the corresponding parser based on the observed data type to perform structured data parsing on the relevant enterprise observed data. For database structures, the system parses table names, field names, field types, primary keys, foreign keys, indexes, and constraints; for business documents, the system parses titles, paragraphs, tables, process descriptions, and business rules; for API call logs, the system parses call paths, request parameters, response fields, operating users, operating roles, status changes, and exception information; for user dialogues, the system parses user intent, error correction statements, business terminology, and contextual references. After obtaining the parsing results, semantic normalization is performed. This semantic normalization includes terminology standardization, synonym merging, naming standardization, field type standardization, time format standardization, role name standardization, and system identifier standardization.
[0046] S12. The semantically normalized parsing result is transformed into an intermediate semantic representation. The intermediate semantic representation may include object candidates, attribute candidates, relation candidates, action candidates, constraint candidates, strategy candidates, evidence fragments, and runtime events, etc.
[0047] S13. Extract candidate object types, candidate attribute types, candidate relationship types, candidate action types, candidate constraint rules, candidate access control rules, candidate approval rules, candidate audit fields, candidate rollback actions, and candidate tool interface definitions from the intermediate semantic representations corresponding to enterprise observation data of various data types to obtain candidate ontology elements.
[0048] Specifically, candidate object types, candidate attribute types, and candidate relationship types can be extracted from the database structure. For example, the system can infer object types based on table names or entity sets, attribute types based on field names and types, and relationship types based on primary and foreign key relationships, related tables, and query co-occurrence. Candidate business concepts, business rules, state machine constraints, amount range constraints, time sequence constraints, approval conditions, and permission rules can be extracted from business documents. Candidate action types can be extracted from API call logs, workflow logs, approval flow logs, and application operation logs. These candidate action types can include target objects, input parameters, output results, preconditions, post-effects, exception handling, permission requirements, approval requirements, audit fields, and rollback actions. Candidate terminology definitions, concept boundaries, aliases, exclusion relationships, mapping relationships, and ontology correction suggestions can be extracted from user dialogues and user error corrections. Candidate ontology gaps can be extracted from historical task failure records. These ontology gaps include missing objects, missing attributes, missing relationships, missing actions, missing constraints, missing permissions, missing tool interfaces, or conceptual ambiguities. For each candidate ontology element, generate a candidate element identifier, element type, name, description, source evidence set, initial confidence level, and pending verification status.
[0049] In this embodiment of the invention, step S2, which involves using the enterprise observation data from the extracted candidate ontology elements as source evidence for the corresponding candidate ontology elements and constructing an evidence graph between the candidate ontology elements and the source evidence, specifically includes the following steps not shown in the accompanying drawings: S21. Take the candidate ontology elements as candidate element nodes, and take the data fragments, database fields, document paragraphs, log events, user dialogues, permission policies, historical tasks and failure records in the source evidence as evidence nodes. Establish evidence edges between candidate element nodes and evidence nodes according to the evidence relationship between candidate element nodes and evidence nodes. The evidence relationship represented by the evidence edge includes at least one of the following: extraction source, semantic support, field mapping, log co-occurrence, action dependency, permission dependency, state change, user confirmation, user denial or conflict relationship.
[0050] S22. Calculate the confidence level of candidate ontology elements based on at least one of the following: number of evidences, type of evidence, credibility of evidence source, freshness of evidence time, consistency between evidences, matching degree with existing ontology, and historical verification results.
[0051] S23. Calculate the degree of conflict of the candidate ontology element based on at least one of the following: name similarity, semantic similarity, field value overlap, relation path overlap, action co-occurrence, permission co-occurrence, and constraint conflict state between the candidate ontology element and the existing ontology element.
[0052] S24. Determine the scope of influence of candidate ontology elements. Specifically, determine the scope of influence of candidate ontology elements based on their impact on enterprise queries, reports, tool interfaces, workflows, permission policies, smart agent tasks, and downstream applications.
[0053] S25. Write the confidence level, conflict level and scope of impact into the evidence diagram, and determine the review priority and processing strategy of the candidate ontology elements accordingly.
[0054] Furthermore, the system generates ontology change patches based on candidate ontology elements and their evidence graphs, generates unique identifiers and version dependency information for each ontology change patch, and stores them in the queue of patches to be verified. An ontology change patch includes the change type, target ontology element, proposed change content, evidence source, confidence level, conflict information, scope of impact, verification plan, review requirements, release strategy, and rollback information.
[0055] The change types include at least one of the following: adding new object types, adding new attribute types, adding new relationship types, adding new action types, adding new constraint rules, adding new permission rules, modifying object definitions, modifying attribute definitions, modifying relationship definitions, modifying action definitions, modifying constraint rules, merging ontology elements, splitting ontology elements, deleting ontology elements, and obliterating ontology elements. For action type changes, the system records the action name, target object, input parameters, output result, preconditions, post-effects, permission requirements, approval requirements, dry-run support, risk level, rollback action, and audit fields in the ontology change patch. For constraint rule changes, the system records the constraint object, constraint expression, triggering condition, verification timing, violation handling method, and source evidence in the ontology change patch. For permission rule changes, the system records the character, subject, resource, action, condition, approval requirements, and audit requirements in the ontology change patch.
[0056] In this embodiment of the invention, the conflict detection (i.e. consistency detection) of the ontology change patch in step S4 includes: performing at least one of the following ontology change patch: name conflict detection, type conflict detection, relation path conflict detection, constraint expression conflict detection, action precondition / precondition conflict detection, and version dependency conflict detection.
[0057] In this embodiment of the invention, step S4, which verifies the permissions of the ontology change patch, includes: for changes to the action type, tool interface, permission rules, and approval rules, verifying whether they comply with at least one of the existing access control model, role model, data permission policy, action permission policy, approval rules, and audit requirements.
[0058] In this embodiment of the invention, the constraint verification of the ontology change patch in step S4 includes: using preset sample data, historical operation records and business rules to verify whether the added or modified constraints can be satisfied, whether they conflict with existing constraints, and whether they will cause existing legal data to become illegal.
[0059] In this embodiment of the invention, step S4, which involves analyzing the impact scope of the ontology change patch, includes: identifying at least one of the following based on ontology element dependencies: query templates, report metrics, tool pattern definitions, software development kit (SDK) types, API mappings, UI forms, workflow nodes, permission policies, and smart agent tasks.
[0060] Subsequently, the system can generate a risk score based on confidence level, conflict level, scope of impact, permission risk, and constraint risk, and determine whether the ontology change patch should be placed on at least one of the following paths: automatic release, manual review, sandbox verification, downgrade processing, or rejection processing.
[0061] In this embodiment of the invention, step S5, which involves sandboxing the ontology change patch that meets the preset low-risk scoring threshold, specifically includes the following steps (not shown in the accompanying drawings): S51. Load the current stable version of the enterprise executable ontology in the sandbox environment, and apply ontology change patches to the enterprise executable ontology to generate candidate ontology versions. Specifically, the system loads the current stable version of the enterprise ontology in the sandbox environment, and applies the ontology change patches to be verified without affecting the production environment, generating candidate ontology versions.
[0062] S52. Perform at least one of the following based on the candidate ontology version: ontology consistency verification, data constraint verification, permission policy verification, tool interface compatibility check, software development kit type check, API mapping check, and UI form validation.
[0063] S53. Select historical queries, historical report tasks, historical API calls, historical workflow tasks, and historical smart agent tasks related to the scope of impact of the ontology change patch, and replay them in the sandbox environment. S54. Compare the replay results with the historical baseline results. If the replay results are consistent with the historical baseline results or improved within the preset allowable range, the verification is recorded as passed; otherwise, the verification is recorded as failed. Specifically, when query failure, permission bypass, tool call failure, abnormal results, or constraint conflicts occur, the verification is recorded as failed. The system generates a verification report based on the verification results. The verification report includes passed items, failed items, affected tasks, risk score, recommended release strategy, and rollback path.
[0064] Furthermore, for ontology change patches triggered by task failures, the system prioritizes replaying the original failed task or a similar task to verify whether the patch can fix the original ontology gaps.
[0065] Furthermore, for ontology change patches that meet preset risk scoring thresholds, involve preset high-risk business actions, or involve preset sensitive permissions, they are submitted to a manual review process so that reviewers can approve, reject, modify, or request supplementary evidence. For ontology change patches that have conflicts, fail verification, or lack sufficient evidence, release is rejected or they are downgraded to a candidate status to await more evidence or a manual review process. If a released ontology change causes query failures, tool call failures, permission anomalies, or business process anomalies in the production environment, a rollback is performed based on the rollback information in the ontology change patch to the previous stable version of the enterprise executable ontology. Rollback can be automatic or semi-automatic. The system records the source of evidence, review records, verification results, release time, release personnel, scope of impact, and rollback information for each ontology change, forming an auditable ontology version governance log.
[0066] In this embodiment, an ontology change patch that meets the preset low-risk scoring threshold refers to an ontology change patch with a confidence level higher than the first threshold, a conflict level lower than the second threshold, and an impact range lower than the third threshold. An ontology change patch that meets the preset medium-risk scoring threshold refers to an ontology change patch whose impact range is greater than or equal to the third threshold.
[0067] The method for automatically generating and evolving an enterprise executable ontology provided in this invention can automatically generate intelligent agent tool interfaces and governance strategies based on the enterprise executable ontology. Based on the published enterprise ontology, this invention automatically generates query interfaces, action interfaces, tool schemas, SDKs, APIs, or UIs that can be called by intelligent agents, and simultaneously generates permission verification, approval, dry-run, risk scoring, auditing, and rollback policies. This solution enables the enterprise ontology not only to be used for querying and inference, but also to directly drive intelligent agents to securely execute enterprise business actions.
[0068] Specifically, the system reads object types, attribute types, relationship types, action types, constraint rules, permission rules, and approval rules from the published enterprise ontology. Based on object types, attribute types, and relationship types, it generates query interfaces, query templates, semantic retrieval interfaces, or data access SDKs. Based on action types, it generates intelligent agent tool interface definitions. These tool interface definitions include tool name, tool description, input parameter schema, output result schema, parameter validation rules, preconditions, post-effects, permission validation rules, approval rules, dry-run rules, risk scoring rules, audit fields, and rollback actions. Based on permission rules and approval rules, it generates runtime governance strategies for the intelligent agent tool interfaces. These runtime governance strategies include pre-call permission checks, manual confirmation of sensitive actions, amount threshold approval, operation risk scoring, dry-run rehearsals, post-execution auditing, and exception rollback. The generated tool interfaces, SDKs, APIs, or UIs are then published to the intelligent agent runtime environment, enabling the intelligent agent to securely query, infer, and execute enterprise business actions based on the enterprise ontology.
[0069] In one embodiment of the present invention, the invention further includes cross-system concept alignment and constraint generation operations. This operation can be executed during the initial ontology generation phase, or it can be triggered during the ontology's evolution by new system access, user error correction, task failure, field conflict, or manual review results. Specifically, it includes the following steps: Candidate business concepts are obtained from multiple enterprise systems. These candidate business concepts can come from database table names, field names, data dictionaries, business documents, API paths, workflow nodes, permission policies, report metrics, user dialogues, or smart agent task records.
[0070] Calculate multidimensional evidence features for candidate business concepts in different systems. The multidimensional evidence features include at least two of the following: name similarity, description similarity, field value overlap, primary and foreign key relationships, data lineage, query co-occurrence, report co-occurrence, API call co-occurrence, business action co-occurrence, permission policy co-occurrence, user dialogue co-occurrence, and document description co-occurrence.
[0071] The semantic relationships between candidate business concepts are determined based on multidimensional evidence features. The semantic relationships include at least one of the following: equivalence relationship, inclusion relationship, parent-child relationship, role relationship, association relationship, conflict relationship, and irrelevance relationship.
[0072] Based on the semantic relationships, candidate ontology change patches are generated for ontology mapping, ontology merging, ontology splitting, ontology hierarchical relationships, or role-based relationships. Data fragments, database fields, document paragraphs, log events, queries, permission policies, and user dialogues involved in the judgment are written into the evidence graph as evidence nodes.
[0073] Ontology constraints are generated based on database constraints, sample data distribution, business documents, historical operations, workflow status transitions, and user error correction. The generated ontology constraints are then represented as candidate ontology change patches. The ontology constraints include at least one of the following: uniqueness constraints, non-null constraints, enumeration value constraints, cardinality constraints, lifecycle state machine constraints, time sequence constraints, amount range constraints, permission constraints, approval constraints, and data quality constraints.
[0074] Candidate ontology change patches for ontology mapping, ontology merging, ontology splitting, ontology hierarchical relationships, or role-based relationships, as well as candidate ontology change patches for ontology constraints, are subjected to sandbox verification. The verification includes replaying historical queries, historical report tasks, historical API calls, historical workflow tasks, and historical intelligent agent tasks to confirm that concept alignment or constraint generation will not disrupt existing tasks.
[0075] Based on the verification results, candidate ontology change patches for ontology mapping, ontology merging, ontology splitting, ontology hierarchical relationships or role-based relationships, as well as candidate ontology change patches for ontology constraints, are automatically published, submitted for manual review, rejected, downgraded, or rolled back.
[0076] Through the above methods, the present invention can automatically discover synonymous, near-synonymous, hierarchical, or role-based concepts among multiple enterprise systems, and incorporate concept mapping and business constraints into the enterprise ontology in an auditable, verifiable, and rollback-able manner.
[0077] Figure 2 This is a flowchart illustrating the specific implementation of a method for automatically generating and evolving an enterprise executable ontology, according to another embodiment of the present invention; as follows: Figure 2 As shown, the system first extracts candidate ontology elements from multi-source enterprise observation data, establishes the correspondence between candidate ontology elements and source evidence through an evidence graph, and generates an auditable, verifiable, and rollback-capable OntologyDelta (ontology increment / ontology difference). Before release, ontology change patches undergo conflict detection, permission verification, constraint verification, and sandbox task replay; based on the verification results, automatic release, manual review, rejection, downgrade, or rollback are executed. After release, the enterprise ontology further generates tool interfaces that can be invoked by intelligent agents, and the operational feedback is reintegrated into the next round of ontology evolution.
[0078] The following specific example illustrates the implementation of generating an initial enterprise ontology from a database and business documents. The enterprise system contains an `orders` table, a `customers` table, and an `invoices` table. After scanning the database structure, the system identifies three candidate object types: `Customer`, `Order`, and `Invoice`, and identifies the relationships between `Customer` and `Order`, and between `Order` and `Invoice`. Simultaneously, the system identifies business rules from the business documents, such as "orders can only be shipped after payment is completed" and "invoice amount must be greater than or equal to 0". The system uses database fields and document paragraphs as evidence nodes to construct an evidence graph between candidate ontology elements and source evidence, and generates multiple ontology change patches. After constraint verification and sandbox replay, the system releases the initial enterprise ontology version.
[0079] The following specific example illustrates the implementation of generating the shipping action ontology from API logs. The system detects multiple calls to the ` / orders / {id} / ship` interface in the API logs. It finds that the order status is typically `paid` before the call and `shipped` afterward, with the `tracking_number` field being updated. Further analysis of the workflow logs reveals that this action is part of the fulfillment process and is usually executed by a logistics role. Based on this, the system generates a `ShipOrder` action type. The target object is `Order`, and the input parameters include `Order_id`, `carrier`, and `tracking_number`. The precondition is `Order.status` = `paid`, and the postcondition is `Order.status` = `shipped`. The required permissions are `LogisticsOperator`, and the rollback is `CancelShipment`. The system uses API logs, workflow nodes, and permission policies as evidence to generate ontology change patches and replays relevant historical order tasks in a sandbox. After successful verification, the system publishes the action type and generates the ShipOrder tool schema (description specification) for intelligent agents to call.
[0080] The following specific example illustrates the implementation of user-corrected error-driven concept boundary correction. When a user interacts with the intelligent agent, they state, "The customer here is not a registered user, but a contracted enterprise account." The system recognizes this statement as user error correction and writes it as an evidence node into the evidence graph. Based on this, the system generates an ontology change patch to correct the concept boundary of "Customer," establishing an equivalence or mapping relationship between "Customer" and "ContractedAccount," and marking "RegisteredUser" as a different concept. The system replays historical queries and intelligent agent tasks using the "customer" concept in a sandbox environment to verify whether the concept boundary correction affects existing tasks. If the verification passes, the patch is released; if the impact is significant, it is submitted for manual review.
[0081] The following specific example illustrates the implementation of ontology gap repair driven by task failure. A user asks, "Which customers have recently experienced increased payment risk?" The intelligent agent fails to execute because the current enterprise ontology lacks payment risk-related attributes or indicators. The system categorizes this failure record as "missing attribute or indicator" and retrieves relevant database fields, reports, and business documents. The system identifies the fields `late_payment_count` (number of overdue payments), `credit_score` (credit score), and `payment_delay_days` (number of days of payment delay) and finds payment risk scoring rules in the risk control documents. Based on this, the system generates a candidate attribute `Customer.payment_risk_score` (payment risk score) and corresponding calculation logic, and generates an ontology change patch. The system replays the original failure question in a sandbox. If the intelligent agent can correctly answer the question based on the candidate ontology version, the patch enters the release or manual review process.
[0082] The following specific example illustrates the implementation of generating a smart agent tool interface from an executable ontology. The enterprise ontology already contains the `ApproveRefund` action type. The target object of this action type is `RefundRequest` (refund application), and the input parameters include `refund_request_id` (refund application number), `approved_amount` (approved amount), and `reason` (approval reason). Preconditions include `RefundRequest.status` (refund application status) = pending (pending processing) and `approved_amount` (approved amount) <= `RefundRequest.requested_amount` (refund amount requested). The required permissions are `FinanceManager` (financial administrator), and the approval requirement is that large refunds require secondary approval. Rollback is defined as `CancelRefund` (refund cancellation). Based on this action type, the system automatically generates the smart agent tool interface definition and generates parameter validation, permission validation, approval process, dry-run rehearsal, audit fields, and rollback rules. Before calling the tool, the intelligent agent must first perform parameter verification and permission verification; when the refund amount exceeds the threshold, the system automatically triggers manual approval; after execution, the system records the operator, time, reason, status before execution and status after execution.
[0083] The following specific example illustrates the implementation of cross-system concept alignment and constraint generation. An enterprise simultaneously exists a Customer Relationship Management System (CRM), an Enterprise Resource Planning (ERP) system, a Billing system, a Customer Service system, and a Contract system. Specifically, the CRM system contains the concept of `Account`, the ERP system contains `Customer`, the Billing system contains `Payer`, the Customer Service system contains `Organization`, and the Contract system contains `LegalEntity`. The system retrieves field names, field descriptions, field value samples, primary and foreign key relationships, historical queries, API call logs, business action logs, permission policies, and business document descriptions from each of these systems. The system calculates multi-dimensional evidentiary features between each candidate concept. When `Account` and `Customer` have a high degree of overlap in field values such as the Unified Social Credit Code, Contract Number, Invoice Title, and Customer Number, and frequently co-occur in order creation, contract signing, invoice issuance, and customer service tasks, the system determines that `Account` and `Customer` are equivalent or similar business concepts. When a Payer primarily appears in payment, invoicing, collection, and refund actions, and its field value set is a subset of the Customer field value set or a role-based view, the system determines that the Payer represents the Customer's role in the payment scenario. When a LegalEntity has a strong association with contract signing, invoicing entities, and tax information, the system determines that the LegalEntity represents the Customer's contracting entity in contract and financial / tax scenarios. Based on this, the system generates an ontology mapping patch. This ontology mapping patch may include: an equivalence mapping between Account and Customer, a role relationship between Payer and Customer, a contracting entity relationship between LegalEntity and Customer, and a customer service scenario mapping relationship between Organization and Customer. The system further generates ontology constraints based on database constraints, sample data distribution, and business documents. For example, `Customer.legal_entity_id` (customer's legal entity ID) should be unique within the same tenant; `Invoice.amount` (invoice amount) should be greater than or equal to 0; `Payer` should be associated with at least one `PaymentAccount` (payment account entity); `Refund.amount` (refund amount) should not exceed the original payment amount; and no new orders can be created after a customer's status is terminated. The system represents the alignment results of the above concepts and the generated constraints as an Ontology Delta, and replays historical queries, historical reporting tasks, historical API calls, and historical intelligent agent tasks in the sandbox environment.If the verification results show that the patch will not disrupt existing queries, permission checks, and business processes, the system will automatically release it or submit it for manual review; if there is a conflict or the impact is significant, the system will downgrade the patch to a candidate status and request manual confirmation.
[0084] This invention provides a method for the automatic generation and evolution of an enterprise executable ontology, which has the following beneficial effects: This invention does not generate a static ontology solely from a database schema or business documents. Instead, it comprehensively utilizes multi-source enterprise observation data, including database structure, sample data, business documents, user dialogues, user error correction, API call logs, workflow logs, approval workflow logs, permission policies, historical task records, and intelligent agent failure records, to generate an executable enterprise ontology that can describe objects, attributes, relationships, actions, constraints, permissions, approvals, audits, and rollback rules. This more accurately reflects the actual business operation status of the enterprise and supports intelligent agents in executing enterprise business actions.
[0085] This invention constructs an evidence graph for candidate ontology elements, binding each candidate object, attribute, relationship, action, constraint, or permission rule to its source field, document fragment, log event, user dialogue, permission policy, or historical task. The evidence graph system can calculate the confidence level, conflict level, scope of influence, and review priority of candidate ontology elements, thereby avoiding the untrustworthy and unauditable problems caused by the black-box generation of ontology from large language models.
[0086] This invention, after generating an initial executable ontology, does not directly modify the full ontology. Instead, it represents each ontology update as an ontology change patch. Each ontology change patch includes the change type, target ontology element, proposed change content, evidence source, confidence level, conflict information, scope of impact, verification result, review status, release status, and rollback information. This mechanism enables the evolution of an enterprise ontology to be explainable, auditable, verifiable, and rollbackable.
[0087] This invention applies the ontology change patch to a sandbox environment before releasing the change, and replays historical queries, report tasks, API calls, workflow tasks, and intelligent agent tasks to verify whether the change disrupts existing system behavior. This solution reduces the risk of ontology updates to the production environment, preventing query failures, tool call failures, permission bypasses, or business process anomalies caused by ontology changes.
[0088] This invention uses user error correction, intelligent agent task failure, tool call failure, permission verification failure, and process execution failure as ontology evolution signals to automatically identify ontology gaps and generate candidate ontology change patches. This mechanism enables the enterprise ontology to continuously learn from real-world operations, rather than relying on periodic manual maintenance.
[0089] For the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.
[0090] This invention provides an automatic generation and evolution system for an enterprise executable ontology. The system includes functional modules for implementing the automatic generation and evolution method of the enterprise executable ontology as described in any of the above embodiments. Figure 3 This diagram schematically illustrates the structure of the automatic generation and evolution system for an enterprise executable ontology provided in an embodiment of the present invention, with reference to... Figure 3 The system described in this embodiment of the invention includes: The ontology element extraction module 301 is used to extract candidate ontology elements from multi-source enterprise observation data. The evidence graph construction module 302 is used to construct an evidence graph between candidate ontology elements and source evidence by taking the enterprise observation data of the extracted candidate ontology elements as the source evidence of the corresponding candidate ontology elements. The patch generation module 303 is used to generate ontology change patches for the enterprise executable ontology based on candidate ontology elements and their evidence graphs. Risk assessment module 304 is used to perform conflict detection, permission verification, constraint verification, and impact scope analysis on ontology change patches to generate a risk score. Sandbox verification module 305 is used to perform sandbox verification on ontology change patches that meet the preset low-risk scoring threshold. The release module 306 is used to release ontology change patches that have passed sandbox verification to generate a new enterprise executable ontology version, so that the intelligent agent can perform query tasks, inference tasks and execute enterprise business actions based on the new enterprise executable ontology version. Evolution module 307 is used to collect operation data and task failure records during the operation of the intelligent agent, write the error correction records in the operation data as new evidence nodes into the evidence graph, and generate candidate ontology change patches for error correction based on the new evidence graph. It determines the ontology gap type according to the failure reasons in the task failure records, extracts candidate ontology elements from multi-source enterprise observation data based on the ontology gap type, and constructs an evidence graph to generate candidate ontology change patches for repairing ontology gaps. The sandbox verification module 305 is also used to perform sandbox verification on candidate ontology change patches for error correction and candidate ontology change patches for fixing ontology gaps using the original failed task or a similar task. Release module 306 is also used to release candidate ontology change patches that have passed sandbox validation to generate new enterprise executable ontology versions.
[0091] As the system implementation is basically similar to the method implementation, the description is relatively simple, and relevant parts can be found in the description of the method implementation.
[0092] Furthermore, another embodiment of the present invention provides a computer storage medium storing a computer program, which, when executed, implements the steps described in the above embodiment of the automatic generation and evolution method for an enterprise executable ontology, for example... Figure 1 Steps S1-S9 are shown.
[0093] Furthermore, another embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When executed by the processor, the computer program implements the steps described in the above embodiment of the automatic generation and evolution method for an enterprise executable ontology, for example... Figure 1 Steps S1-S9 are shown.
[0094] This invention reflects the true semantics of enterprise operations. Existing ontology construction schemes typically rely on database schemas, business documents, or manual maintenance results, making it difficult to reflect the true semantics of enterprise business operations in a timely manner. This invention introduces API call logs, workflow logs, approval flow logs, user operation logs, intelligent agent operation logs, and task failure records, enabling the ontology to continuously evolve from real-world operational data. Therefore, the enterprise ontology generated by this invention is closer to the actual business processes of enterprises.
[0095] This invention enables intelligent agents to securely execute enterprise actions. Traditional knowledge graphs primarily describe entities, attributes, and relationships, but struggle to describe action parameters, preconditions, post-effects, permissions, approvals, audits, and rollbacks. This invention models business actions as first-class elements of the ontology and automatically generates intelligent agent tool interfaces and governance policies. Therefore, this invention enables intelligent agents to expand from "answering questions" to "securely executing enterprise operations."
[0096] This invention offers enhanced credibility and auditability. Instead of directly accepting the ontology results generated by a large language model, it binds each candidate ontology element to source evidence through an evidence graph and calculates confidence, conflict level, and scope of impact. Therefore, this invention can clearly explain the origin of each ontology element, the supporting evidence, whether it conflicts with existing ontology, and the necessary verification and auditing.
[0097] This invention offers improved controllability and rollback capability. It represents each ontology update as an OntologyDelta ontology change patch, recording the changes, evidence, verification results, and rollback information within the patch. Therefore, when an ontology update causes an anomaly, the system can locate the problematic patch and perform a rollback, reducing the risk of enterprise-level ontology evolution.
[0098] This invention reduces the risk of downstream system failures. Before ontology changes are released, this invention performs conflict detection, permission verification, constraint verification, impact analysis, and sandbox task replay. By replaying historical queries, historical tasks, historical API calls, and historical intelligent proxy calls, the system can proactively detect ontology changes that may disrupt existing business logic. Therefore, this invention reduces query failures, tool call failures, permission bypassing, and business process anomalies caused by ontology updates.
[0099] This invention reduces the cost of deploying intelligent agents. Based on an enterprise-specific executable ontology automatic generation tool schema, SDK, API, UI, and runtime governance strategies, this invention ensures that the intelligent agent tool interface remains consistent with enterprise semantics, permission rules, and business constraints. Therefore, this invention reduces the development and governance costs of enterprise AI agents and improves the security and consistency of the tool interface.
[0100] This invention enables continuous learning from failures. It treats intelligent agent task failures, query failures, and tool call failures as ontology gap signals and automatically generates candidate ontology change patches, which are then verified through the original failed task or similar tasks. Therefore, this invention enables continuous self-repair and evolution of the enterprise ontology.
[0101] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, any of the claimed embodiments can be used in any combination.
[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for the automatic generation and evolution of an enterprise executable ontology, characterized in that, The method includes: S1. Extract candidate ontology elements from multi-source enterprise observation data; S2. Use the enterprise observation data of the extracted candidate ontology elements as the source evidence for the corresponding candidate ontology elements, and construct an evidence graph between the candidate ontology elements and the source evidence. S3. Generate an ontology change patch for the enterprise executable ontology based on the candidate ontology elements and their evidence graph; S4. Perform conflict detection, permission verification, constraint verification, and impact scope analysis on ontology change patches to generate risk scores; S5. Sandbox verification of ontology change patches that meet the preset low-risk scoring threshold. S6. Release an ontology change patch that has passed sandbox verification to generate a new enterprise executable ontology version, so that the intelligent agent can perform query tasks, inference tasks and execute enterprise business actions based on the new enterprise executable ontology version. S7. Collect the running data and task failure records during the operation of the intelligent agent, write the error correction records in the running data into the evidence graph as new evidence nodes, and generate candidate ontology change patches for error correction based on the new evidence graph. Determine the ontology gap type according to the failure reasons in the task failure records. Extract candidate ontology elements from the multi-source enterprise observation data based on the ontology gap type and construct the evidence graph to generate candidate ontology change patches for repairing ontology gaps. S8. Use the original failed task or a similar task to perform sandbox verification on the candidate ontology change patch for error correction and the candidate ontology change patch for fixing ontology gaps. S9. Release candidate ontology change patches that have passed sandbox verification to generate new enterprise executable ontology versions.
2. The method according to claim 1, characterized in that, Candidate ontology elements are extracted from multi-source enterprise observation data, including: Structured data parsing was performed on enterprise observation data of various data types from multiple sources, and semantic normalization was performed on the parsing results. The semantically normalized parsing results are transformed into intermediate semantic representations; Candidate ontology elements are obtained by extracting candidate object types, candidate attribute types, candidate relationship types, candidate action types, candidate constraint rules, candidate access control rules, candidate approval rules, candidate audit fields, candidate rollback actions, and candidate tool interface definitions from the intermediate semantic representations corresponding to enterprise observation data of various data types.
3. The method according to claim 1, characterized in that, Enterprise observation data from extracted candidate ontology elements are used as source evidence for the corresponding candidate ontology elements. An evidence graph is constructed between the candidate ontology elements and the source evidence, including: The candidate ontology elements are used as candidate element nodes, and the data fragments, database fields, document paragraphs, log events, user dialogues, permission policies, historical tasks and failure records in the source evidence are used as evidence nodes. Evidence edges are established between candidate element nodes and evidence nodes based on the evidence relationship between them. The confidence level of a candidate ontology element is calculated based on at least one of the following factors: the number of evidences, the type of evidence, the credibility of the evidence source, the freshness of the evidence time, the consistency between evidences, the matching degree with existing ontology, and historical verification results. The degree of conflict of a candidate ontology element is calculated based on at least one of the following: name similarity, semantic similarity, field value overlap, relation path overlap, action co-occurrence, permission co-occurrence, and constraint conflict state between the candidate ontology element and the existing ontology element. Determine the scope of influence of candidate ontology elements; The confidence level, conflict level, and scope of impact are written into the evidence graph, and the review priority and processing strategy of candidate ontology elements are determined accordingly.
4. The method according to claim 3, characterized in that, An ontology change patch includes the change type, target ontology element, proposed change content, evidence source, confidence level, conflict information, scope of impact, verification plan, review requirements, release strategy, and rollback information; The change types include at least one of the following: adding a new object type, adding a new attribute type, adding a new relationship type, adding a new action type, adding a new constraint rule, adding a new permission rule, modifying an object definition, modifying an attribute definition, modifying a relationship definition, modifying an action definition, modifying a constraint rule, merging ontology elements, splitting ontology elements, deleting ontology elements, and obliterating ontology elements.
5. The method according to claim 1, characterized in that, Conflict detection for ontology change patches includes at least one of the following: name conflict detection, type conflict detection, relation path conflict detection, constraint expression conflict detection, action precondition / precondition conflict detection, and version dependency conflict detection. The permission verification for ontology change patches includes: for changes such as adding or modifying action types, tool interfaces, permission rules, and approval rules, verifying whether they comply with at least one of the existing access control model, role model, data permission policy, action permission policy, approval rules, and audit requirements; Constraint verification for ontology change patches includes: using preset sample data, historical operation records, and business rules to verify whether the added or modified constraints can be satisfied, whether they conflict with existing constraints, and whether they will cause existing valid data to become invalid; The impact scope analysis of ontology change patches includes: identifying at least one of the following affected query templates, report metrics, tool pattern definitions, software development kit types, API mappings, UI forms, workflow nodes, permission policies, and smart agent tasks based on ontology element dependencies.
6. The method according to claim 5, characterized in that, Ontology change patches that meet the preset low-risk scoring threshold will undergo sandbox verification, including: Load the current stable version of the enterprise executable ontology in the sandbox environment, and apply ontology change patches to the enterprise executable ontology to generate candidate ontology versions; Perform at least one of the following based on the candidate ontology version: ontology consistency verification, data constraint verification, permission policy verification, tool interface compatibility check, software development kit type check, API mapping check, and UI form validation. Select historical queries, historical report tasks, historical API calls, historical workflow tasks, and historical smart agent tasks related to the impact scope of the ontology change patch, and replay them in the sandbox environment; Compare the playback results with the historical baseline results. If the playback results are consistent with the historical baseline results or improved within the preset allowable range, the record verification is successful; otherwise, the record verification fails.
7. The method according to claim 1, characterized in that, The method further includes: For ontology change patches that meet the preset medium risk scoring threshold, involve preset high-risk business actions, or involve preset sensitive permissions, submit them for manual review. For ontology change patches that have conflicts, fail verification, or lack sufficient evidence, refuse to release them or downgrade them to a candidate status to await more evidence or a manual review process. If a release of ontology changes causes query failures, tool call failures, permission errors, or business process errors in the production environment, the system will roll back to the previous stable version of the enterprise executable ontology based on the rollback information in the ontology change patch.
8. The method according to claim 1, characterized in that, The method further includes: Obtain candidate business concepts from multiple enterprise systems; Calculate multidimensional evidence features for candidate business concepts in different systems; Determine the semantic relationships between candidate business concepts based on multidimensional evidence features; Based on the semantic relationships, candidate ontology change patches are generated for ontology mapping, ontology merging, ontology splitting, ontology hierarchical relationships, or role-based relationships. Data fragments, database fields, document paragraphs, log events, queries, permission policies, and user dialogues involved in the judgment are written into the evidence graph as evidence nodes. Candidate ontology change patches are generated based on database constraints, sample data distribution, business documents, historical operations, workflow status transitions, and user error correction. Candidate ontology change patches for ontology mapping, ontology merging, ontology splitting, ontology hierarchical relationships or role-based relationships, and candidate ontology change patches for ontology constraints are all subjected to sandbox verification. Based on the verification results, candidate ontology change patches for ontology mapping, ontology merging, ontology splitting, ontology hierarchical relationships or role-based relationships, as well as candidate ontology change patches for ontology constraints, are automatically published, submitted for manual review, rejected, downgraded, or rolled back.
9. An automatic generation and evolution system for an enterprise executable ontology, characterized in that, The system includes functional modules for implementing the method as described in claim 1, including: The ontology element extraction module is used to extract candidate ontology elements from multi-source enterprise observation data. The evidence graph construction module is used to construct an evidence graph between candidate ontology elements and source evidence by taking enterprise observation data of extracted candidate ontology elements as source evidence for the corresponding candidate ontology elements. The patch generation module is used to generate ontology change patches for an enterprise executable ontology based on candidate ontology elements and their evidence graphs. The risk assessment module is used to perform conflict detection, permission verification, constraint verification, and impact scope analysis on ontology change patches to generate a risk score. The sandbox verification module is used to perform sandbox verification on ontology change patches that meet the preset low-risk scoring threshold. The release module is used to release ontology change patches that have passed sandbox verification to generate a new enterprise executable ontology version, so that the intelligent agent can perform query tasks, inference tasks, and execute enterprise business actions based on the new enterprise executable ontology version. The evolution module is used to collect operational data and task failure records during the operation of the intelligent agent. It writes the error correction records in the operational data into the evidence graph as new evidence nodes and generates candidate ontology change patches for error correction based on the new evidence graph. It determines the ontology gap type according to the failure reasons in the task failure records. Based on the ontology gap type, it extracts candidate ontology elements from multi-source enterprise observation data and constructs the evidence graph to generate candidate ontology change patches for repairing ontology gaps. The sandbox verification module is also used to perform sandbox verification on candidate ontology change patches for error correction and candidate ontology change patches for fixing ontology gaps using the original failed task or a similar task. The release module is also used to release candidate ontology change patches that have passed sandbox validation to generate new enterprise executable ontology versions.
10. A computer storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed, implements the automatic generation and evolution method of the enterprise executable ontology as described in any one of claims 1-8.
