A large model-based ontology modeling and dynamic evolution method and system

Abstract

The application provides a large model-based ontology modeling and dynamic evolution method and system, which comprises the following steps: constructing an ontology hierarchical structure of a target field; receiving a multi-source heterogeneous business data stream in real time, preprocessing the received data, performing entity alignment and standardization on the preprocessed data, and dividing the data into data partitions corresponding to the ontology hierarchical structure; performing multi-round extraction and supplement of ontology knowledge on the distributed business data of each level to generate a candidate knowledge set to be inserted into the ontology; performing multi-dimensional conflict detection on the candidate knowledge set and the current ontology to retain legal candidate knowledge; calculating an ontology update prediction impact value, and if the ontology update prediction impact value is less than a hierarchical safety threshold, inserting the candidate knowledge. The application reduces labor costs, improves ontology modeling efficiency and architecture standardization degree, improves knowledge mining accuracy and reasoning ability, eliminates semantic ambiguity and logical errors generated by ontology iteration and evolution, and the ontology evolution whole process is traceable.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an ontology modeling and dynamic evolution method and system based on a large model. Background Technology

[0002] Ontologies, as the core carrier for the standardized and structured expression of domain knowledge, are widely used in many industries such as intelligent manufacturing, healthcare, power grid operation and maintenance, and urban transportation, serving as the fundamental support for knowledge reasoning, semantic understanding, and data fusion. However, with the deepening of digitalization and intelligentization across industries, the explosive growth of multi-source heterogeneous data has exposed many shortcomings in traditional ontology modeling and evolution schemes, making them difficult to adapt to current business development needs. Existing technologies suffer from the following deficiencies: Ontology modeling is inefficient and highly dependent on manual labor. Traditional domain ontology construction mainly relies on domain experts to manually sort through literature, extract entities, and define attributes and semantic relationships. This is not only time-consuming and costly, but also easily affected by the experience of personnel, resulting in problems such as entity omissions, non-standard relationship definitions, and unreasonable hierarchical structures. When faced with complex domain knowledge, it suffers from both insufficient modeling quality and efficiency.

[0003] Traditional ontologies are mostly static and fixed structures, relying solely on periodic manual updates and cannot absorb new business knowledge in real time. Even if some solutions support semi-automatic updates, they do not perform comprehensive conflict verification or topological risk prediction on new knowledge. New content is highly likely to cause semantic contradictions, logical loops, and hierarchical disorder in the ontology, resulting in compromised ontology consistency.

[0004] Existing technologies, when updating ontology, do not assess the impact of knowledge insertion on the overall topology and the risk of its propagation. Blindly updating can easily paralyze the entire ontology system.

[0005] Therefore, the urgent technical problems to be solved are that existing technologies suffer from low efficiency and poor standardization in manual modeling, difficulty in standardizing and automatically integrating multi-source heterogeneous data into the ontology, and the tendency for conflicts to arise during ontology updates. Summary of the Invention

[0006] The purpose of this application is to provide a method and system for ontology modeling and dynamic evolution based on a large model. It utilizes the large model to automatically extract domain knowledge and construct an ontology hierarchical structure, thereby getting rid of the traditional manual modeling mode, significantly reducing labor costs, improving ontology modeling efficiency and architectural standardization, improving knowledge mining accuracy and reasoning ability, eliminating semantic ambiguity and logical errors generated by ontology iterative evolution, and achieving full traceability and quality control of the ontology evolution process.

[0007] To achieve the above objectives, this application provides a method for ontology modeling and dynamic evolution based on a large model. This method includes the following steps: constructing an ontology hierarchy structure for the target domain; receiving multi-source heterogeneous business data streams in real time, preprocessing the received data locally, performing entity alignment and standardization on the preprocessed data through a cloud-based semantic bridging layer, and dividing it into data partitions corresponding to the ontology hierarchy structure; based on a collaborative reasoning architecture of a cloud-based inference large model and local mapping, performing multiple rounds of ontology knowledge extraction and supplementation on the distributed business data at each level to generate a candidate knowledge set to be inserted into the ontology; performing multi-dimensional conflict detection between the candidate knowledge set and the current ontology, retaining legitimate candidate knowledge; calculating the ontology update prediction impact value based on the propagation range of the topological influence after inserting legitimate candidate knowledge into the pre-rendered ontology hierarchy structure; if the ontology update prediction impact value is less than the hierarchy safety threshold, then candidate knowledge is inserted; otherwise, insertion of candidate knowledge is prohibited; and through multi-layer semantic and topological structure collaborative optimization of the model, verifying ontology consistency in real time, retaining the verified qualified ontology as the optimized hierarchical ontology.

[0008] The ontology modeling and dynamic evolution method based on a large model, as described above, further includes the following steps: based on the optimized hierarchical ontology, calculate the ontology version difference entropy, generate the ontology version evolution chain, compare the current version with the historical versions, and trigger the evolution quality assessment when the difference entropy exceeds the evolution control threshold.

[0009] The ontology modeling and dynamic evolution method based on large models described above includes the following steps in constructing the ontology hierarchy of the target domain: extracting core entities from standard literature in the target domain through large models, extracting entity attribute constraint data, and semantic topological relationships between entities. Based on core entities, entity attribute data, and semantic relationships between entities, construct an ontology hierarchy for the target domain.

[0010] The ontology modeling and dynamic evolution method based on a large model, as described above, involves real-time reception of multi-source heterogeneous business data streams, preprocessing the received data locally, and performing entity alignment and standardization on the preprocessed data through a cloud-based semantic bridging layer, before allocating it to data partitions corresponding to the ontology hierarchical structure. This includes: real-time reception of multi-source heterogeneous business data streams; preliminary cleaning, deduplication, invalid data filtering, and compression of the received data streams before uploading them to the cloud; standardization and semantic alignment of the received data through a cloud-based semantic bridging layer; and automatic distribution of the standardized and semantically aligned data to data partitions corresponding to the matched ontology hierarchical structure through semantic matching and hierarchical routing algorithms deployed in the cloud, performing semantic similarity calculation, hierarchical path search, and target concept node determination.

[0011] The ontology modeling and dynamic evolution method based on a large model, as described above, involves calculating the predicted impact value of the ontology update based on the propagation range of the topological influence after inserting legitimate candidate knowledge into the pre-simulated ontology hierarchical structure. If the predicted impact value of the ontology update is less than the hierarchical safety threshold, candidate knowledge is inserted; otherwise, the insertion of candidate knowledge is prohibited. This includes: calculating the predicted topological influence value of the associated edges after inserting legitimate candidate knowledge into the ontology hierarchical structure based on the topological influence value of the associated edges; comparing the ontology update impact value with the hierarchical safety threshold; if the predicted impact value of the ontology update is less than the hierarchical safety threshold, candidate knowledge is inserted; otherwise, the insertion of candidate knowledge is prohibited.

[0012] The ontology modeling and dynamic evolution method based on large models described above includes the following: Real-time consistency verification of the ontology is performed through a multi-layer semantic and topological structure collaborative optimization model, retaining verified ontologies as the optimized hierarchical ontology. This includes: performing multi-dimensional semantic matching verification on the ontology through a multi-layer semantic and topological structure collaborative optimization model, retaining verified ontologies and removing those that fail verification; and performing consistency verification on the ontology.

[0013] The ontology modeling and dynamic evolution method based on a large model, as described above, includes the following steps for consistency verification of the ontology: Semantic disambiguation is performed on the text descriptions of newly added concept nodes to verify whether there is any ambiguity or overlap between the newly added concept nodes and existing concept nodes. If not, the current verification passes; otherwise, the current verification fails. The rationality of the direction of newly added association edges is verified, and the existence of logical loops is checked. If the direction of the newly added association edges is reasonable, the current verification passes; otherwise, the current verification fails. The connection of newly added association edges conforms to the association rules defined in the ontology. If yes, the current verification passes; otherwise, the current verification fails. The consistency between newly added candidate knowledge and the original ontology logical axioms is verified. If they are consistent, the current verification passes; otherwise, the current verification fails. If all the above verifications pass, the ontology consistency verification is successful; otherwise, the ontology consistency verification fails.

[0014] The ontology modeling and dynamic evolution method based on large models described above involves constructing an ontology hierarchy for the target domain based on core entities, entity attribute data, and semantic relationships between entities. This includes: constructing multi-level concept nodes for the ontology hierarchy; constructing attribute nodes attached to multi-level concept nodes based on the attribute constraints of the ontology; constructing association edges between concept nodes, between concept nodes and attribute nodes, and between attribute nodes based on the internal semantic relationships of the ontology; and setting semantic weights and topological influence values ​​for each association edge based on the semantic similarity and association tightness between concepts in the ontology.

[0015] As a second aspect of this application, this application provides an ontology modeling and dynamic evolution system based on a large model. The system includes: a construction module for constructing an ontology hierarchical structure of the target domain; a preprocessing module for receiving multi-source heterogeneous business data streams in real time and preprocessing the received data on the local end; and a cloud semantic bridging layer for performing entity alignment and standardization on the preprocessed data and dividing it into data partitions corresponding to the ontology hierarchical structure.

[0016] The reasoning module is used for a collaborative reasoning architecture based on a cloud-based large-scale reasoning model and local mapping. It performs multiple rounds of ontology knowledge extraction and supplementation on the distributed business data at each level to generate a candidate knowledge set to be inserted into the ontology. The data processor is used to perform multi-dimensional conflict detection between the candidate knowledge set and the current ontology, retaining the legitimate candidate knowledge. Based on the propagation range of the topological impact after inserting legitimate candidate knowledge according to the pre-simulated ontology hierarchical structure, it calculates the ontology update prediction impact value. If the ontology update prediction impact value is less than the hierarchical safety threshold, the candidate knowledge is inserted; otherwise, the insertion of candidate knowledge is prohibited. The verification module is used to optimize the model through multi-layer semantics and topological structure, verify the consistency of the ontology in real time, and retain the verified ontology as the optimized hierarchical ontology.

[0017] As described above, the ontology modeling and dynamic evolution system based on a large model includes a data processor that is also used to calculate the ontology version difference entropy based on the optimized hierarchical ontology, generate an ontology version evolution chain, compare the current version with historical versions, and trigger an evolution quality assessment when the difference entropy exceeds the evolution control threshold.

[0018] The beneficial effects achieved by this application are as follows: (1) This application utilizes a large model to automatically extract core entities, attribute constraints, and semantic topological relationships from domain literature, autonomously constructs multi-level concept nodes, attribute nodes, and related relationship edges, and quantitatively sets semantic weights and topological influence values. It completely eliminates the heavy reliance on domain experts, significantly shortens the modeling cycle, and ensures the uniformity and standardization of the ontology hierarchical structure and semantic relationships.

[0019] (2) This application completes data cleaning, deduplication, compression and other preprocessing on the local end, combines the semantic bridging layer in the cloud to achieve entity alignment and data standardization, and then relies on semantic matching and hierarchical routing algorithms to complete automatic data partitioning and distribution. It can seamlessly connect with various streaming business data, open up the connection channel between heterogeneous data and the ontology system, has a high degree of data processing automation, and effectively reduces the cloud transmission and computing pressure.

[0020] (3) This application adopts a collaborative reasoning architecture of cloud-based large reasoning model and local lightweight model. It can adaptively call single-path / dual-path reasoning mode according to the complexity of the task to complete multiple rounds of knowledge extraction and supplementation. The completeness and accuracy of the candidate knowledge set are significantly improved, taking into account both the requirements of complex knowledge deep reasoning and local low-latency response.

[0021] (4) This application performs multi-dimensional conflict detection before knowledge insertion to filter out illegal knowledge; at the same time, it quantitatively calculates the propagation range and predicted impact value of the topological influence after knowledge insertion, and combines security threshold to control update behavior, thereby avoiding the destruction of the ontology topology from the source and greatly reducing the probability of update failure.

[0022] (5) This application uses a multi-layer semantic and topological collaborative optimization model to carry out consistency verification from four dimensions: semantic disambiguation, relation direction, association rules, and logical axioms. It can accurately identify problems such as ambiguous nodes, logical loops, and rule conflicts, correct ontology defects in real time, and ensure the continuous compliance of ontology semantics, logic, and hierarchy.

[0023] (6) This application achieves full-process traceability of evolution and closed-loop quality control. It introduces ontology version difference entropy calculation to construct a complete version evolution chain, which can quickly compare differences between different versions; when the difference entropy exceeds the standard, it automatically triggers quality assessment and supports abnormal version rollback and local correction. It realizes traceability of the ontology evolution process, early warning of anomalies, and repairability of problems, forming a complete quality control closed loop. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0025] Figure 1 This is a flowchart illustrating an ontology modeling and dynamic evolution method based on a large model, according to an embodiment of this application.

[0026] Figure 2 This is a schematic diagram of an ontology modeling and dynamic evolution system based on a large model, according to an embodiment of this application. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0028] Example 1 like Figure 1 As shown, this application provides an ontology modeling and dynamic evolution method based on a large model, which includes the following steps: Step S1: Construct the ontology hierarchy of the target domain.

[0029] Each ontology has its own conceptual hierarchy, attribute constraints, and relational topology. Based on the ontology's conceptual hierarchy, attribute constraints, and relational topology, the ontology hierarchy structure of the target domain ontology is constructed, including multi-level conceptual nodes, attribute nodes, and relational edges, and semantic weights and topological influence values ​​are set for each relational edge.

[0030] Individual target fields can include intelligent manufacturing equipment, medical care, power operation and maintenance, or urban transportation. The entity itself can be intelligent manufacturing equipment, medical equipment, power equipment, or urban transportation facilities.

[0031] Step S1 includes: Step S110: Extract core entities from standard literature in the target domain using a large model, extract entity attribute constraint data, and semantic topological relationships between entities.

[0032] The entity attribute constraint data includes: the entity's inherent attributes, dynamic attributes, numerical attributes, or state attributes. The entity's inherent attributes include its name, category, and classification level.

[0033] Among them, the semantic topological relationships between entities include: subordinate, associated, belonging, causal and other relationships.

[0034] Step S120: Construct the ontology hierarchy of the target domain based on the core entities, entity attribute data, and semantic relationships between entities.

[0035] Step S120 includes: Step S121: Construct multi-level concept nodes of the ontology hierarchy.

[0036] The ontology hierarchy includes multi-level concept nodes, each comprising multiple levels. This can be 3 levels, 4 levels, etc. Taking a 3-level structure as an example, the multi-level concept nodes include: a core concept top-level layer, an intermediate layer, and a concrete instance layer. Each layer includes one or more concept nodes.

[0037] Understandably, each domain class, term, entity, or instance in the ontology can be mapped one-to-one to a concept node in the ontology hierarchy.

[0038] Step S122: Based on the attribute constraints of the ontology, construct the attribute nodes attached to the multi-level concept nodes.

[0039] The attribute constraints of an ontology include data attributes (numerical, text, boolean, state), object attributes (feature associations between concepts), attribute domain, value range, and constraints (value range, type constraints, cardinality constraints), etc. Specifically, based on the attribute constraints of each ontology, each independent attribute constraint is attached as an attribute node to the corresponding concept node.

[0040] Step S123: Based on the internal semantic relationships of the ontology, construct the association edges between concept nodes, between concept nodes and attribute nodes, and between attribute nodes.

[0041] Specifically, iterate through all semantic relationships between concepts, between concepts and attributes, and between attributes in the ontology, and model each semantic relationship as a directed association edge.

[0042] Step S124: Based on the semantic similarity and correlation between concepts in the ontology, set semantic weight and topological influence value for each correlation edge.

[0043] As an example, the ontology is intelligent manufacturing industrial equipment, and its hierarchical topology includes: a core top layer, a middle layer, and a specific instance layer.

[0044] Multi-level concept nodes include: Core Top-Level: Industrial Intelligent Manufacturing Equipment; Intermediate layer: processing equipment, conveying equipment, and testing equipment; Specific instance layer: CNC machine tools, milling machines, belt conveyors, roller conveyors, dimensional measuring instruments, and appearance measuring instruments.

[0045] Data attributes, performance constraints, and structural parameters are obtained from the industrial equipment itself and used as attribute nodes attached to multi-level conceptual nodes.

[0046] Attribute nodes for CNC machine tools: rated speed, machining accuracy, load power, and failure rate; attribute nodes for belt conveyors: conveying speed, bandwidth, load capacity, and operating temperature; attribute nodes for dimensional measuring instruments: measuring range, measurement error, and sampling frequency.

[0047] As an example, the following nodes are connected by edges: CNC machine tools are inherited from machining equipment; belt conveyors are inherited from conveying equipment; CNC machine tools have the attribute of machining accuracy; belt conveyors belong to intelligent production line units.

[0048] The formula for calculating semantic weight is as follows: ; in, Indicates semantic weight; The weight representing the influence of a fixed association strength on the semantic relationship type between the two nodes connected by an edge. This represents the fixed association strength between two nodes connected by an edge representing the semantic relationship type. Different semantic relationship types correspond to different fixed association strengths: the fixed association strength for inheritance is 0.95; the fixed association strength for membership is 0.9; and the fixed association strength for attribute relationships is 0.88. This represents the hierarchical distance between two nodes connected by an edge, indicating the association strength. If the two nodes are at the same level, If two nodes are spaced one level apart, then... Two nodes separated by two or more layers. ; The weight representing the influence of the hierarchical distance association strength between two nodes. This indicates the degree of overlap in attributes between the two nodes connected by an edge representing an association relationship. The weight representing the influence of the degree of attribute overlap between the two nodes connected by the edge of the association relationship.

[0049] ; in, This indicates the number of overlapping attributes between the two nodes connected by an edge representing an association relationship. This indicates the number of attributes of the first node connected by the edge of the association relationship; This indicates the number of attributes of the second node connected by the edge of the association relationship.

[0050] in, ; As an example: .

[0051] The formula for calculating the topological influence value is as follows: ; in; Indicates the topological influence value; This represents the hub connectivity of the current associated edges; This represents the cross-layer index between two nodes connected by an edge indicating a relationship. This represents the neighborhood coverage of the two nodes connected by the edge indicating an association relationship. These represent the proportion of the association edge traversed by the shortest path in the ontology hierarchical semantic topology, the cross-layer index between the two nodes connected by the association edge, and the weight of the neighborhood coverage of the two nodes connected by the association edge, respectively. .

[0052] Where the current association edge connects two multi-level concept nodes, then If the current association edge connects to branch nodes within the same multi-level concept node, then If the edge of the association connects two isolated edge nodes, then For example, edge-isolated connections: a small backup fan is an auxiliary accessory.

[0053] If two nodes are connected at the same level, If two nodes are connected in a middle layer to the root layer or in a middle layer to the bottom layer, then If two nodes are separated by two or more layers, .

[0054] in, ; in, This indicates the number of neighboring nodes of the first node connected by the edge of the association relationship; This indicates the number of neighboring nodes of the second node connected by the edge of the association relationship; This represents the total number of nodes in the ontology-level semantic topology. For example, the neighboring nodes of a CNC machine tool include machining equipment, machining accuracy, and milling machine; the neighboring nodes of a dimensional inspection instrument include inspection equipment, measurement error, and appearance inspection instrument.

[0055] Step S2: Receive multi-source heterogeneous business data streams in real time, preprocess the received data on the local end, perform entity alignment and standardization on the preprocessed data through the cloud semantic bridging layer, and divide it into data partitions corresponding to the ontology hierarchical structure.

[0056] Specifically, it collects multi-source heterogeneous business data streams from business systems, IoT, and text in real time, automatically matches and distributes them to corresponding data partitions according to the ontology hierarchy, realizes real-time perception of incremental data, and provides a data source for the dynamic evolution of the ontology. Multi-source heterogeneous business data streams refer to a collection of business data originating from multiple independent business systems, devices, or platforms, which have significant differences in data structure, storage format, semantic representation, and transmission protocol, and are generated and transmitted in a real-time streaming and continuous manner.

[0057] Step S2 includes: Step S210: Receive multi-source heterogeneous business data streams in real time, perform preliminary cleaning, deduplication, invalid data filtering and compression on the received data streams, and upload them to the cloud.

[0058] The received data stream is initially cleaned, deduplicated, and invalid data is filtered and compressed before being transmitted to the cloud. Only valid data is uploaded to the cloud, reducing the data processing pressure on the cloud.

[0059] Step S220: Standardize and semantically align the received data through the cloud semantic bridging layer.

[0060] In the semantic bridging layer, based on the ontology hierarchy, mapping rules are established between each data source field and concept nodes and attribute nodes. The mapping rules include direct mapping, transformation mapping and aggregation mapping.

[0061] Standardization and semantic alignment include: name unification, relation unification, attribute unit unification, etc.

[0062] Step S230: Using semantic matching and hierarchical routing algorithms deployed in the cloud, semantic similarity calculation, hierarchical path search and target concept node determination are performed. The data that has been standardized and semantically aligned is automatically distributed to the data partition corresponding to the ontology hierarchical structure that matches it.

[0063] Specifically, by deploying a semantic matching and hierarchical routing algorithm in the cloud, semantic similarity calculation, hierarchical path search, and target concept node determination are performed. Based on the semantic type, concept, attribute features, and relationships of the data, it automatically determines which layer and concept node of the ontology it belongs to without manual intervention. The semantic matching and hierarchical routing algorithm is implemented using existing technologies. This application deploys it in the cloud and integrates it with the ontology hierarchical structure to achieve efficient hierarchical routing of multi-source heterogeneous business data streams.

[0064] It should be explained that each concept node at the ontology level corresponds to a data partition, namely a data storage and processing partition. Only after the data is placed in the corresponding data partition can it be efficiently used by the subsequent knowledge extraction and evolution update modules.

[0065] It should be noted that the ontology hierarchy, semantic bridging layer, semantic matching and hierarchical routing algorithms, and data partitioning are all deployed in the cloud.

[0066] Step S3: Based on the collaborative reasoning architecture of cloud-based reasoning big model and local mapping, perform multiple rounds of ontology knowledge extraction and supplementation on the distributed business data at each level to generate a candidate knowledge set to be inserted into the ontology.

[0067] The collaborative reasoning architecture of cloud-based inference big model and local mapping includes a cloud-based inference big model deployed in the cloud and a local lightweight inference model deployed locally.

[0068] Depending on the task complexity, real-time requirements, and hierarchical depth, the cloud-based inference model and the local lightweight inference model are adaptively invoked in single-path or dual-path mode. Single-path invocation means invoking only one of the cloud-based inference model and the local lightweight inference model, while dual-path invocation means invoking both the cloud-based inference model and the local lightweight inference model simultaneously.

[0069] Step S4: Perform multi-dimensional conflict detection between the candidate knowledge set and the current ontology, and retain the legitimate candidate knowledge.

[0070] Multi-dimensional conflict detection includes: concept conflict, attribute conflict, relationship conflict, hierarchy conflict, logical consistency conflict, etc.

[0071] Among them, if any conflict exists between the candidate knowledge in the candidate knowledge set and the ontology during multi-dimensional conflict detection, it is considered an invalid candidate knowledge.

[0072] Step S5: Calculate the predicted impact value of the ontology update based on the propagation range of the topological impact after inserting legitimate candidate knowledge into the pre-simulated ontology hierarchical structure. If the predicted impact value of the ontology update is less than the hierarchical safety threshold, then insert candidate knowledge; otherwise, prohibit the insertion of candidate knowledge.

[0073] Step S5 includes: Step S510: Based on the topological influence value of the associated edges, calculate the predicted topological influence value of the associated edges after inserting valid candidate knowledge into the ontology hierarchical structure.

[0074] The formula for calculating the predicted topological influence value of the associated edges after inserting valid candidate knowledge into the ontology hierarchy is as follows: ; in, Indicates the first Predicted topological impact values ​​of each associated edge; Indicates the first The original topological influence values ​​of each associated edge; Indicates and the Each association edge directly shares the total number of all association edges (i.e., neighborhood association edges) of the same concept node; Indicates the first The nth relation edge and its neighborhood's nth relation edge The coupling coefficient of each associated edge; different coupling coefficients correspond to different association types and strengths of association between two edges; if a change in the weight of one associated edge directly affects another associated edge, it needs to be updated synchronously. If a change in the weight of one association edge indirectly affects a change in the weight of another neighboring association edge, the weight change must be propagated proportionally. If two related edges do not affect each other and do not require linked updates, then . For the depth attenuation rate, preferably, set... ; Indicates the first The hierarchy depth of the associated edge; with the first The maximum depth of the concept nodes connected by the i-th relation edge is used as the th The depth of each layer containing a relational edge; the top-level concept node has a depth of 1, the middle-level node has a depth of 2, the bottom-level node has a depth of 3, and so on. The larger the number, the greater the depth, the greater the decay, and the smaller the topological influence value. Indicates the correction factor; This represents the confidence level of the legitimate candidate knowledge awaiting insertion. This is a positive correction coefficient. The larger the value, the stronger the influence of knowledge confidence on the topological decay term; the smaller the value, the weaker the regulatory effect of confidence. The value ranges from 0 to 1. This applies to high-risk core knowledge. General business knowledge, Low-risk example knowledge, .

[0075] in, ; in, This represents the matching value between the current candidate knowledge and the standard ontology knowledge; This indicates the number of times that a valid candidate knowledge entry of this type was judged as conflicting or incorrect and returned within a predetermined time period T. This represents the total number of times this type of valid candidate knowledge has been inserted into the past within a predetermined time period T. This is achieved through calculation. The stability and error rate of this type of candidate knowledge in past ontology update evolution are measured.

[0076] in, ; in, This indicates the degree of matching between the current candidate knowledge and the ontology entity; This indicates the degree of matching between the current candidate knowledge and the ontology; This indicates the degree of matching between the current candidate knowledge and the ontology attributes; This represents the weight of the influence of the matching degree between the current candidate knowledge and the ontology entity on the matching value between the current candidate knowledge and the standard ontology knowledge. This represents the weight of the influence of the matching degree between the current candidate knowledge and the ontology on the matching value between the current candidate knowledge and the standard ontology knowledge; This represents the weight of the influence of the matching degree between the current candidate knowledge and the ontology attribute on the matching value between the current candidate knowledge and the standard ontology knowledge.

[0077] in, ; The semantic vectors representing entities in the candidate knowledge (generated by BERT / SBERT); This represents the semantic vector corresponding to the standard entity in the ontology. represents the cosine similarity function, used to calculate the directional similarity between two semantic vectors, and its value ranges from [-1, 1]. Wherein, if the relation in the candidate knowledge completely matches the type of the corresponding standard relation in the ontology, then ;otherwise ;in A semantic vector representing the relations in candidate knowledge; A semantic vector representing the standard relation in the ontology.

[0078] in, ; A semantic vector representing the attributes in the candidate knowledge; This represents the semantic vector corresponding to the standard attributes in the ontology.

[0079] Step S520: Compare the ontology update impact value with the hierarchical safety threshold. If the ontology update prediction impact value is less than the hierarchical safety threshold, insert candidate knowledge; otherwise, prohibit the insertion of candidate knowledge.

[0080] Step S6: Through the multi-layer semantic and topological structure collaborative optimization model, the consistency of the ontology is verified in real time, and the verified ontologies are retained as the optimized hierarchical ontologies.

[0081] Specifically, real-time ontology consistency verification includes: real-time verification of ontology logical consistency, hierarchical integrity, and semantic correctness, and repair of potential problems after evolution. This enables continuous, stable, and self-correcting dynamic evolution of the ontology.

[0082] Step S6 includes: Step S610: Through a multi-layer semantic and topological structure collaborative optimization model, multi-dimensional semantic matching verification is performed on the ontology, retaining the verified ontology and removing the verified ontology.

[0083] Specifically, multi-dimensional semantic matching verification includes: comparing the similarity of the entity, relation, and attribute three-dimensional vectors of the ontology with the corresponding standard entities, relations, and attributes of the ontology; using cosine similarity to calculate semantic consistency; and identifying synonymous, ambiguous, and contradictory semantic content.

[0084] Step S620: Perform consistency verification on the ontology.

[0085] The consistency verification of the ontology includes the following steps: Step S621: Perform semantic disambiguation on the text description of the newly added concept node to verify whether the newly added concept node has any ambiguity overlap with the existing concept node. If not, the current verification passes; otherwise, the current verification fails.

[0086] Step S622: Verify the rationality of the direction of the newly added association edge and check for logical loops. If the direction of the newly added association edge is reasonable, the current verification passes; otherwise, the current verification fails.

[0087] Step S623: Verify that the connection of the newly added association edge conforms to the association rules defined in the ontology. If it does, the current verification passes; otherwise, the current verification fails.

[0088] Step S624: Verify the consistency between the newly added candidate knowledge and the original ontology logical axioms. If they are consistent, the current verification passes; otherwise, the current verification fails.

[0089] Step S625: If the verification in steps S621-S624 all pass, the ontology consistency verification is qualified; otherwise, the ontology consistency verification fails.

[0090] Step S7: Based on the optimized hierarchical ontology, calculate the ontology version difference entropy, generate the ontology version evolution chain, compare the current version with the historical versions, and trigger the evolution quality assessment when the difference entropy exceeds the evolution control threshold.

[0091] Evolutionary quality assessment: A comprehensive quality score is calculated based on six quality dimensions: integrity, consistency, accuracy, timeliness, traceability, and scalability. When the comprehensive quality score is lower than the quality baseline, ontology rollback or local correction is performed based on the version evolution chain to achieve closed-loop quality control of ontology evolution.

[0092] Assuming the ontology is from version Evolved to The formula for calculating the entropy of version differences in the ontology is: ; in, This represents the entropy of the current ontology version difference; This indicates the total number of nodes added, deleted, or modified in this update; Indicates the first The average of the predicted topological influence values ​​of the edges connecting the nodes; This indicates the total number of newly added, deleted, or modified relationship edges in this update; Indicates the first Predicted topological impact value of each associated edge.

[0093] The first initial ontology is denoted as The initial difference entropy of the ontology version is denoted as As the starting point of the evolutionary chain, after each round of ontology update, the entropy difference between the current version and the previous version is calculated. This version unit is encapsulated and sequentially attached to the end of the evolution chain. Based on the ontology version difference entropy obtained from each update, a version unit is constructed by combining the version number, timestamp, changes, and quality status. These version units are then chained together in chronological order to form the ontology version evolution chain. This evolution chain allows for arbitrary retrieval of historical versions, enabling cross-version difference comparisons. Abnormal versions are marked based on exceeding the difference entropy limit, supporting evolution quality control and version backtracking.

[0094] Example 2 like Figure 2 As shown, this application provides an ontology modeling and dynamic evolution system 100 based on a large model, the system comprising: Module 10 is used to build the ontology hierarchy of the target domain.

[0095] The preprocessing module 20 is used to receive multi-source heterogeneous business data streams in real time and preprocess the received data locally.

[0096] The cloud-based semantic bridging layer 30 is used to perform entity alignment and standardization on the preprocessed data and divide it into data partitions corresponding to the ontology hierarchy structure.

[0097] The reasoning module 40 is used for a collaborative reasoning architecture based on a cloud-based large reasoning model and local mapping. It performs multiple rounds of ontology knowledge extraction and supplementation on the distributed business data at each level to generate a candidate knowledge set to be inserted into the ontology.

[0098] The data processor 50 is used to perform multi-dimensional conflict detection between the candidate knowledge set and the current ontology, and retain the legitimate candidate knowledge. Based on the propagation range of the topological influence after inserting legitimate candidate knowledge according to the pre-simulated ontology hierarchical structure, the ontology update prediction influence value is calculated. If the ontology update prediction influence value is less than the hierarchical safety threshold, the candidate knowledge is inserted; otherwise, the insertion of candidate knowledge is prohibited.

[0099] The verification module 60 is used to verify the consistency of the ontology in real time through multi-layer semantics and topology collaborative optimization of the model, and retain the verified ontology as the optimized hierarchical ontology.

[0100] The data processor 50 is also used to calculate the ontology version difference entropy based on the optimized hierarchical ontology, generate the ontology version evolution chain, compare the current version with the historical version, and trigger the evolution quality assessment when the difference entropy exceeds the evolution control threshold.

[0101] This application also provides a computer storage medium storing computer instructions, which, when invoked, execute the address mapping method of the large-capacity solid-state drive. The computer storage medium includes one or more program instructions, which are executed by a processor to provide an ontology modeling and dynamic evolution method based on a large model.

[0102] The embodiments disclosed in this invention provide a computer-readable storage medium storing computer program instructions. When the computer program instructions are executed on a computer, the computer performs the above-described ontology modeling and dynamic evolution method based on a large model.

[0103] This invention provides a processor for processing the above-described ontology modeling and dynamic evolution method based on a large model.

[0104] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0105] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.

[0106] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0107] The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EEPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM).

[0108] The beneficial effects achieved by this application are as follows: (1) This application utilizes a large model to automatically extract core entities, attribute constraints, and semantic topological relationships from domain literature, autonomously constructs multi-level concept nodes, attribute nodes, and related relationship edges, and quantitatively sets semantic weights and topological influence values. It completely eliminates the heavy reliance on domain experts, significantly shortens the modeling cycle, and ensures the uniformity and standardization of the ontology hierarchical structure and semantic relationships.

[0109] (2) This application completes data cleaning, deduplication, compression and other preprocessing on the local end, combines the semantic bridging layer in the cloud to achieve entity alignment and data standardization, and then relies on semantic matching and hierarchical routing algorithms to complete automatic data partitioning and distribution. It can seamlessly connect with various streaming business data, open up the connection channel between heterogeneous data and the ontology system, has a high degree of data processing automation, and effectively reduces the cloud transmission and computing pressure.

[0110] (3) This application adopts a collaborative reasoning architecture of cloud-based large reasoning model and local lightweight model. It can adaptively call single-path / dual-path reasoning mode according to the complexity of the task to complete multiple rounds of knowledge extraction and supplementation. The completeness and accuracy of the candidate knowledge set are significantly improved, taking into account both the requirements of complex knowledge deep reasoning and local low-latency response.

[0111] (4) This application performs multi-dimensional conflict detection before knowledge insertion to filter out illegal knowledge; at the same time, it quantitatively calculates the propagation range and predicted impact value of the topological influence after knowledge insertion, and combines security threshold to control update behavior, thereby avoiding the destruction of the ontology topology from the source and greatly reducing the probability of update failure.

[0112] (5) This application uses a multi-layer semantic and topological collaborative optimization model to carry out consistency verification from four dimensions: semantic disambiguation, relation direction, association rules, and logical axioms. It can accurately identify problems such as ambiguous nodes, logical loops, and rule conflicts, correct ontology defects in real time, and ensure the continuous compliance of ontology semantics, logic, and hierarchy.

[0113] (6) This application achieves full-process traceability of evolution and closed-loop quality control. It introduces ontology version difference entropy calculation to construct a complete version evolution chain, which can quickly compare differences between different versions; when the difference entropy exceeds the standard, it automatically triggers quality assessment and supports abnormal version rollback and local correction. It realizes traceability of the ontology evolution process, early warning of anomalies, and repairability of problems, forming a complete quality control closed loop.

[0114] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0115] In the description of this application, the word "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0116] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for ontology modeling and dynamic evolution based on a large model, characterized in that, The method includes the following steps: Construct the ontology hierarchy of the target domain; It receives multi-source heterogeneous business data streams in real time, preprocesses the received data on the local end, performs entity alignment and standardization on the preprocessed data through the cloud semantic bridging layer, and divides it into the data partitions corresponding to the ontology hierarchical structure. Based on a collaborative reasoning architecture that combines a cloud-based large-scale reasoning model with local mapping, the system performs multiple rounds of ontology knowledge extraction and supplementation on the distributed business data at each level to generate a candidate knowledge set for the ontology to be inserted. Perform multi-dimensional conflict detection between the candidate knowledge set and the current ontology, and retain the legitimate candidate knowledge; Based on the propagation range of topological influence after inserting legitimate candidate knowledge into the pre-simulated ontology hierarchical structure, calculate the ontology update prediction impact value. If the ontology update prediction impact value is less than the hierarchical safety threshold, then insert candidate knowledge; otherwise, prohibit the insertion of candidate knowledge. By using a multi-layer semantic and topological structure collaborative optimization model, the consistency of the ontology is verified in real time, and the verified ontology is retained as the optimized hierarchical ontology.

2. The ontology modeling and dynamic evolution method based on a large model according to claim 1, characterized in that, The method also includes the following steps: based on the optimized hierarchical ontology, calculate the ontology version difference entropy, generate the ontology version evolution chain, compare the current version with the historical version, and trigger the evolution quality assessment when the difference entropy exceeds the evolution control threshold.

3. The ontology modeling and dynamic evolution method based on a large model according to claim 1, characterized in that, Constructing the ontology hierarchy of the target domain includes: Core entities are extracted from standard literature in the target domain using a large model, and entity attribute constraint data and semantic topological relationships between entities are extracted. Based on core entities, entity attribute data, and semantic relationships between entities, construct an ontology hierarchy for the target domain.

4. The ontology modeling and dynamic evolution method based on a large model according to claim 1, characterized in that, It receives multi-source heterogeneous business data streams in real time, preprocesses the received data locally, and then performs entity alignment and standardization on the preprocessed data through a cloud-based semantic bridging layer, dividing it into data partitions corresponding to the ontology hierarchy, including: It receives multi-source heterogeneous business data streams in real time, performs preliminary cleaning, deduplication, invalid data filtering and compression on the received data streams, and uploads them to the cloud. The received data is standardized and semantically aligned through a cloud-based semantic bridging layer. By deploying semantic matching and hierarchical routing algorithms in the cloud, semantic similarity calculation, hierarchical path search and target concept node determination are performed. The data that has been standardized and semantically aligned is automatically distributed to the data partition corresponding to the ontology hierarchical structure that matches it.

5. The ontology modeling and dynamic evolution method based on a large model according to claim 1, characterized in that, Based on the propagation range of topological influence after inserting legitimate candidate knowledge into the pre-simulated ontology hierarchical structure, the predicted impact value of the ontology update is calculated. If the predicted impact value of the ontology update is less than the hierarchical safety threshold, candidate knowledge is inserted; otherwise, the insertion of candidate knowledge is prohibited, including: Based on the topological impact value of the association edges, calculate the predicted topological impact value of the association edges after inserting valid candidate knowledge into the ontology hierarchy structure; Compare the ontology update impact value with the hierarchical safety threshold. If the ontology update predicted impact value is less than the hierarchical safety threshold, then insert candidate knowledge; otherwise, prohibit the insertion of candidate knowledge.

6. The ontology modeling and dynamic evolution method based on a large model according to claim 1, characterized in that, Through a multi-layer semantic and topological structure collaborative optimization model, ontology consistency is verified in real time, and the verified ontologies are retained as the optimized hierarchical ontologies, including: By using a multi-layer semantic and topological structure collaborative optimization model, the ontology is subjected to multi-dimensional semantic matching verification, retaining the verified ontology and removing the verified ontology. Perform a consistency check on the ontology.

7. The ontology modeling and dynamic evolution method based on a large model according to claim 6, characterized in that, Performing consistency checks on the ontology includes the following steps: Semantic disambiguation is performed on the text description of the newly added concept node to verify whether the newly added concept node has any ambiguity overlap with the existing concept node. If not, the current verification passes; otherwise, the current verification fails. Verify the rationality of the direction of the newly added association edge and check for the existence of logical loops. If the direction of the newly added association edge is reasonable, the current verification passes; otherwise, the current verification fails. Verify that the connection of the newly added associated edges conforms to the association rules defined in the ontology. If it does, the current verification passes; otherwise, the current verification fails. Verify the consistency between the newly added candidate knowledge and the original ontology logic axioms. If they are consistent, the current verification passes; otherwise, the current verification fails. If all the above verifications pass, the ontology consistency check is successful; otherwise, the ontology consistency check fails.

8. The ontology modeling and dynamic evolution method based on a large model according to claim 3, characterized in that, Based on core entities, entity attribute data, and semantic relationships between entities, the ontology hierarchy of the target domain is constructed as follows: Constructing multi-level concept nodes within the ontology's hierarchical structure; Based on the attribute constraints of the ontology, construct attribute nodes attached to multi-level concept nodes; Based on the internal semantic relationships of the ontology, construct the association edges between concept nodes, between concept nodes and attribute nodes, and between attribute nodes; Based on the semantic similarity and correlation between concepts in the ontology, semantic weights and topological influence values ​​are assigned to each edge of the association relationship.

9. An ontology modeling and dynamic evolution system based on a large model, characterized in that, The system includes: Modules are used to build the ontology hierarchy of the target domain; The preprocessing module is used to receive multi-source heterogeneous business data streams in real time and preprocess the received data locally. The cloud-based semantic bridging layer is used to perform entity alignment and standardization on the preprocessed data and divide it into data partitions corresponding to the ontology hierarchy. The reasoning module is used for a collaborative reasoning architecture based on a cloud-based large-scale reasoning model and local mapping. It performs multiple rounds of ontology knowledge extraction and supplementation on the distributed business data at each level to generate a candidate knowledge set to be inserted into the ontology. The data processor is used to perform multi-dimensional conflict detection between the candidate knowledge set and the current ontology, and retain the legitimate candidate knowledge. Based on the propagation range of the topological influence after inserting legitimate candidate knowledge according to the pre-simulated ontology hierarchical structure, it calculates the ontology update prediction impact value. If the ontology update prediction impact value is less than the hierarchical safety threshold, the candidate knowledge is inserted; otherwise, the insertion of candidate knowledge is prohibited. The verification module is used to optimize the model through multi-layer semantics and topology, verify the consistency of the ontology in real time, and retain the verified ontology as the optimized hierarchical ontology.

10. The ontology modeling and dynamic evolution system based on a large model according to claim 9, characterized in that, The data processor is also used to calculate the ontology version difference entropy based on the optimized hierarchical ontology, generate the ontology version evolution chain, compare the current version with the historical version, and trigger the evolution quality assessment when the difference entropy exceeds the evolution control threshold.

Images