Data asset management method and device based on knowledge graph, terminal equipment and storage medium

Abstract

The invention discloses a data asset management method and device based on a knowledge graph, terminal equipment and a storage medium, and belongs to the field of knowledge graphs. The method comprises the steps that all data sources serve as federated learning nodes, initial local models are constructed, and all initial local model parameters are aggregated through a federated aggregation algorithm; generating initial global model parameters, repeatedly carrying out iterative updating operation on the initial local model, generating a global knowledge graph, and obtaining change data when determining that a target data source with a data change event exists by monitoring the data change event of the data source; and according to the change data, updating the current global knowledge graph, and generating an updated target global knowledge graph. Therefore, by implementing the method and the device, the problem of data leakage when the multi-source knowledge graph is constructed at present can be solved.

Description

Technical Field

[0001] This invention relates to the field of knowledge graphs, and in particular to a data asset management method, apparatus, terminal device, and storage medium based on knowledge graphs. Background Technology

[0002] With the booming development of the digital economy, data has become a key production factor, and the management and application of data assets are becoming increasingly important. In recent years, knowledge graphs have been widely used in financial institutions to manage data assets. By constructing a structured network of entities, such as customers, accounts, products, transactions, and risk indicators, and the relationships between entities, such as holding, trading, guarantee, and association, efficient management of data assets can be achieved. The core logic is to transform scattered financial data into knowledge assets that are "associative, inferable, and reusable".

[0003] However, financial institutions' data is usually scattered across multiple data sources, such as core banking systems, credit systems, transaction systems, and risk control systems. In the process of building and updating data assets such as knowledge graphs, it is necessary to aggregate data from various data sources for processing. Therefore, the centralized data processing model of knowledge graphs can lead to excessive data concentration, resulting in data leakage problems in current knowledge graph-based data asset management methods. Summary of the Invention

[0004] This invention provides a data asset management method, apparatus, terminal device, and storage medium based on knowledge graphs. The method can solve the problem of data leakage in current data asset management methods based on knowledge graphs.

[0005] An embodiment of the present invention provides a data asset management method based on knowledge graphs, comprising:

[0006] Listen for data change events in each data source that stores data assets. When a target data source with a data change event is identified, obtain the change data of the target local knowledge graph of the target data source.

[0007] Based on the changed data, the current global knowledge graph is updated to generate the updated target global knowledge graph;

[0008] The construction of the global knowledge graph includes:

[0009] Obtain several data sources;

[0010] Each of the aforementioned data sources is used as a federated learning node, and an initial local model is constructed based on the pre-built local knowledge graph in each of the aforementioned data sources, and the initial local model parameters of each of the initial local models are determined; wherein, the local knowledge graph is a knowledge graph constructed based on the data assets in the corresponding data source, including: several entities and the relationships between the entities; the initial local model is used to generate the embedding vectors of each entity and each relationship in each local knowledge graph.

[0011] The federated aggregation algorithm is used to aggregate the initial local model parameters to generate the initial global model parameters;

[0012] Based on the initial global model parameters, the initial local model is iteratively updated repeatedly. In each iteration, a global knowledge graph to be evaluated is constructed based on the embedding vectors generated by the updated local model. When the accuracy of the global knowledge graph to be evaluated meets a preset accuracy threshold, the final generated global knowledge graph to be evaluated is used as the global knowledge graph.

[0013] Furthermore, the iterative update operation includes:

[0014] Obtain the first global model parameters; initially, the first global model parameters are the initial global model parameters.

[0015] The first global model parameters are distributed to each of the data sources so that the first local model in each of the data sources is updated according to the first global model parameters to generate a second local model, and the second local model is used to generate several first embedding vectors of the local knowledge graph; wherein, initially, the first local model is the initial local model;

[0016] Based on several first embedding vectors from each data source, a global knowledge graph to be evaluated is constructed, and it is determined whether the accuracy of the global knowledge graph to be evaluated meets a preset accuracy threshold.

[0017] If so, the global knowledge graph to be evaluated will be used as the global knowledge graph.

[0018] If not, the federated aggregation algorithm is used to aggregate the local model parameters of the second local model to generate the first global model parameters for the next round of iteration update operation.

[0019] Furthermore, the first embedding vector includes: a first sub-embedded vector of the entity and a second sub-embedded vector of the relation;

[0020] The step of constructing a global knowledge graph to be evaluated based on several first embedding vectors from various data sources includes:

[0021] Using a pre-defined similarity recognition model, the attribute similarity and structural similarity between each first sub-embedded vector and other first sub-embedded vectors are calculated.

[0022] Based on the attribute similarity and the structural similarity, determine the comprehensive similarity between every two first sub-embedding vectors;

[0023] When the overall similarity is greater than a preset similarity threshold, the two entities of the corresponding two first sub-embedding vectors are aligned to generate a global knowledge graph to be optimized.

[0024] A path query embedding algorithm is used to perform path queries on the second sub-embedded vector in the global knowledge graph to be optimized, determine the hidden relationships in the global knowledge graph to be optimized, and optimize the global knowledge graph to be optimized based on the hidden relationships to generate a global knowledge graph to be evaluated.

[0025] Furthermore, after using the final generated global knowledge graph to be evaluated as the global knowledge graph, it also includes:

[0026] Obtain the first embedding vector generated by each data source in the last round of iteration update operation;

[0027] A hash algorithm is used to generate ownership metadata for each first embedding vector;

[0028] Construct a blockchain network and smart contracts for verifying the blockchain network, and store the ownership metadata in the blockchain network.

[0029] Furthermore, before updating the current global knowledge graph based on the changed data, the method further includes:

[0030] The smart contract is used to perform security verification on the changed data;

[0031] Upon successful security verification, the current global knowledge graph is updated based on the changed data.

[0032] If security verification fails, generate an alert for abnormal data changes.

[0033] An embodiment of the present invention also provides a data asset management device based on a knowledge graph, comprising:

[0034] The data source monitoring module is used to monitor data change events of each data source that stores data assets. When a target data source with a data change event is determined to exist, the module obtains the change data of the target local knowledge graph of the target data source.

[0035] The knowledge graph construction and update module is used to update the current global knowledge graph based on the changed data and generate the updated target global knowledge graph.

[0036] The construction of the global knowledge graph includes:

[0037] Obtain several data sources;

[0038] Each of the aforementioned data sources is used as a federated learning node, and an initial local model is constructed based on the pre-built local knowledge graph in each of the aforementioned data sources, and the initial local model parameters of each of the initial local models are determined; wherein, the local knowledge graph is a knowledge graph constructed based on the data assets in the corresponding data source, including: several entities and the relationships between the entities; the initial local model is used to generate the embedding vectors of each entity and each relationship in each local knowledge graph.

[0039] The federated aggregation algorithm is used to aggregate the initial local model parameters to generate the initial global model parameters;

[0040] Based on the initial global model parameters, the initial local model is iteratively updated repeatedly. In each iteration, a global knowledge graph to be evaluated is constructed based on the embedding vectors generated by the updated local model. When the accuracy of the global knowledge graph to be evaluated meets a preset accuracy threshold, the final generated global knowledge graph to be evaluated is used as the global knowledge graph.

[0041] Furthermore, the iterative update operation includes:

[0042] Obtain the first global model parameters; initially, the first global model parameters are the initial global model parameters.

[0043] The first global model parameters are distributed to each of the data sources so that the first local model in each of the data sources is updated according to the first global model parameters to generate a second local model, and the second local model is used to generate several first embedding vectors of the local knowledge graph; wherein, initially, the first local model is the initial local model;

[0044] Based on several first embedding vectors from each data source, a global knowledge graph to be evaluated is constructed, and it is determined whether the accuracy of the global knowledge graph to be evaluated meets a preset accuracy threshold.

[0045] If so, the global knowledge graph to be evaluated will be used as the global knowledge graph.

[0046] If not, the federated aggregation algorithm is used to aggregate the local model parameters of the second local model to generate the first global model parameters for the next round of iteration update operation.

[0047] Furthermore, the first embedding vector includes: a first sub-embedded vector of the entity and a second sub-embedded vector of the relation;

[0048] The step of constructing a global knowledge graph to be evaluated based on several first embedding vectors from various data sources includes:

[0049] Using a pre-defined similarity recognition model, the attribute similarity and structural similarity between each first sub-embedded vector and other first sub-embedded vectors are calculated.

[0050] Based on the attribute similarity and the structural similarity, determine the comprehensive similarity between every two first sub-embedding vectors;

[0051] When the overall similarity is greater than a preset similarity threshold, the two entities of the corresponding two first sub-embedding vectors are aligned to generate a global knowledge graph to be optimized.

[0052] A path query embedding algorithm is used to perform path queries on the second sub-embedded vector in the global knowledge graph to be optimized, determine the hidden relationships in the global knowledge graph to be optimized, and optimize the global knowledge graph to be optimized based on the hidden relationships to generate a global knowledge graph to be evaluated.

[0053] This application also provides a terminal device, including:

[0054] One or more processors;

[0055] A memory, coupled to the processor, for storing one or more programs;

[0056] When the one or more programs are executed by the one or more processors, the one or more processors implement a knowledge graph-based data asset management method as described in the above embodiments of the invention.

[0057] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a knowledge graph-based data asset management method as described in the above embodiments.

[0058] The following benefits can be obtained by implementing the present invention:

[0059] This invention provides a data asset management method, apparatus, terminal device, and storage medium based on knowledge graphs. The method involves using each data source as a federated learning node, constructing an initial local model based on the pre-built local knowledge graph in each data source, and determining the initial local model parameters for each initial local model. A federated aggregation algorithm is then used to aggregate the initial local model parameters to generate initial global model parameters. It is understood that before parameter aggregation, the initial local models are trained based on the local knowledge graph of their respective data sources, resulting in different model parameters. At this stage, the embedding vectors generated by the local models cannot be fused to generate a global knowledge graph. By employing federated aggregation... The algorithm aggregates the initial local model parameters to generate initial global model parameters. Then, based on these initial global model parameters, iteratively updates the initial local model, allowing it to gradually unify the "rules" for extracting embedding vectors during the iterative update process. This leads to the construction of the global knowledge graph to be evaluated. As the "rules" for extracting embedding vectors become more consistent, the accuracy of constructing the global knowledge graph gradually improves, ultimately generating the global knowledge graph. After the global knowledge graph is constructed, it is updated by acquiring change data from the local knowledge graph during data asset management, rather than directly acquiring change data from the data assets, thus reducing the risk of data leakage. Therefore, in the data asset management process of constructing and updating the knowledge graph, this invention does not directly collect the raw data from various data sources, but rather collects model parameters and embedding vectors processed by the model. Thus, this invention solves the data leakage problem inherent in current knowledge graph-based data asset management methods. Attached Figure Description

[0060] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0061] Figure 1 This is a flowchart illustrating a knowledge graph-based data asset management method according to a certain embodiment of this application;

[0062] Figure 2 This is a schematic diagram of the structure of a knowledge graph-based data asset management device provided in a certain embodiment of this application;

[0063] Figure 3 This is a schematic diagram of the structure of a terminal device provided in a certain embodiment of this application. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0065] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0066] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0067] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0068] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0069] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0070] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0071] See Figure 1 To address the data leakage problem that exists when constructing multi-source knowledge graphs, this invention provides a data asset management method based on knowledge graphs, comprising:

[0072] S1. Listen for data change events in each data source that stores data assets. When a target data source with a data change event is determined to exist, obtain the change data of the target local knowledge graph of the target data source.

[0073] S2. Based on the changed data, update the current global knowledge graph to generate the updated target global knowledge graph;

[0074] Preferably, before updating the current global knowledge graph based on the changed data, the method further includes:

[0075] S21. The smart contract is used to perform security verification on the changed data;

[0076] S22. Upon passing the security verification, update the current global knowledge graph based on the changed data;

[0077] S23. If the security verification fails, generate an abnormal data change alarm.

[0078] In a preferred embodiment of the present invention, the data sources storing data assets in the financial institution include: core banking system, credit system, transaction system, risk control system, etc., and each data source covers structured data, semi-structured data in XML or JSON format, as well as unstructured data such as text files and images.

[0079] In this embodiment, a message queue system is built using Kafka. At the data source (data generation end), when data changes, the change event is encapsulated into a message and sent to the Kafka message queue. At the data processing end (the server hosting the global knowledge graph), a Kafka consumer listens to the queue in real time. Upon capturing the change message, it triggers the ownership verification and knowledge graph update process. Simultaneously, following an event tracing model, detailed information about each ownership change is recorded on the blockchain for easy subsequent traceability.

[0080] Furthermore, this application selects Neo4j as the global knowledge graph and the local knowledge graph storage engine, and designs appropriate node labels and relationship types based on the data structure characteristics of the knowledge graph. In the financial knowledge graph, customers, enterprises, and financial products are used as different node labels, and transaction relationships, holding relationships, and regulatory relationships are used as different relationship types.

[0081] Using Neo4j's Cypher query language, entity and relation data generated during multi-source data collaborative modeling, dynamic ownership verification, entity alignment, and relation reasoning are imported into the Neo4j database for storage. In subsequent knowledge graph updates, queries, and analyses, Cypher statements are written to achieve efficient data manipulation.

[0082] The construction of the global knowledge graph includes:

[0083] S24. Obtain several data sources;

[0084] S25. Each of the data sources is used as a federated learning node, and an initial local model is constructed based on the pre-built local knowledge graph in each of the data sources, and the initial local model parameters of each initial local model are determined; wherein, the local knowledge graph is a knowledge graph constructed based on the data assets in the corresponding data source, including: several entities and the relationships between the entities; the initial local model is used to generate the embedding vectors of each entity and each relationship in each local knowledge graph;

[0085] In a preferred embodiment of the present invention, federated learning nodes are deployed locally on each data source, and corresponding computing resources and network environments are configured. FedAvg or FedProx algorithms are selected as federated aggregation algorithms. Taking the financial institution field as an example, core banking systems, credit systems, transaction systems, risk control systems, etc. are used as data sources. Graph neural network models are built locally using TensorFlow or PyTorch frameworks to perform feature extraction and local embedding learning on the local knowledge graph.

[0086] S26. The federated aggregation algorithm is used to aggregate the initial local model parameters to generate the initial global model parameters.

[0087] In a preferred embodiment of the present invention, the federated learning framework uses the FedAvg or FedProx algorithm to aggregate model parameters, wherein the FedAvg algorithm formula is:

[0088]

[0089] Where, θ t+1 θ represents the global model parameters in round t+1, N represents the number of data sources participating in federated learning, reflecting the scale of data sources participating in collaborative modeling. i,t This represents the local model parameters of the i-th data source in round t, reflecting the model parameter status of each data source in a specific round;

[0090] The FedProx algorithm formula is:

[0091]

[0092] Where μ is the proximal term coefficient, used to balance the local model parameter update and the global model parameter consistency. The larger μ is, the closer the local model parameter update is to the global model parameter; conversely, the smaller μ is, the greater the influence of its own characteristics. θ t These are the global model parameters for round t, ||θ i,t -θ t || 2 This represents the squared distance between the local model parameters of the i-th data source and the global model parameters in round t, used to adjust the contribution of the local model parameters to the global model update.

[0093] S27. Based on the initial global model parameters, the initial local model is iteratively updated repeatedly. In each round of iterative update, a global knowledge graph to be evaluated is constructed based on the embedding vectors generated by the updated local model. When the accuracy of the global knowledge graph to be evaluated meets the preset accuracy threshold, the final generated global knowledge graph to be evaluated is used as the global knowledge graph.

[0094] Preferably, the iterative update operation includes:

[0095] S271. Obtain the first global model parameters; wherein, initially, the first global model parameters are the initial global model parameters;

[0096] S272. The first global model parameters are sent to each of the data sources so that the first local model in each of the data sources is updated according to the first global model parameters to generate a second local model, and the second local model is used to generate a number of first embedding vectors of the local knowledge graph; wherein, initially, the first local model is the initial local model.

[0097] S273. Based on several first embedding vectors from each data source, construct a global knowledge graph to be evaluated, and determine whether the accuracy of the global knowledge graph to be evaluated meets a preset accuracy threshold.

[0098] Preferably, the first embedding vector includes: a first sub-embedded vector of the entity and a second sub-embedded vector of the relation;

[0099] The step of constructing a global knowledge graph to be evaluated based on several first embedding vectors from various data sources includes:

[0100] S2731. Using a preset similarity recognition model, calculate the attribute similarity and structural similarity between each first sub-embedded vector and other first sub-embedded vectors;

[0101] S2732. Determine the comprehensive similarity between every two first sub-embedding vectors based on the attribute similarity and the structural similarity.

[0102] S2733. When the overall similarity is greater than the preset similarity threshold, the two entities of the corresponding two first sub-embedding vectors are aligned to generate a global knowledge graph to be optimized.

[0103] S2734. Using a path query embedding algorithm, perform path query on the second sub-embedded vector in the global knowledge graph to be optimized, determine the hidden relationships in the global knowledge graph to be optimized, and optimize the global knowledge graph to be optimized according to the hidden relationships to generate a global knowledge graph to be evaluated.

[0104] In a preferred embodiment of the present invention, the similarity recognition model mines the semantic similarity of cross-source entities through contrastive learning, and its training process includes:

[0105] Construct positive and negative sample pairs, perform comparative learning on the feature representations of entities, maximize the similarity between positive sample pairs, and minimize the similarity between negative sample pairs. The calculation formula is as follows:

[0106]

[0107] Where L represents the loss function, used to measure the difference between the model's prediction and the actual situation. The training objective is to minimize this loss function. N p N is the number of positive sample pairs. n Let s(·) be the number of negative sample pairs, and s(·) be the similarity calculation function, such as cosine similarity. and Two feature representations of the same entity constitute a positive sample pair. To and The feature representations of different entities constitute negative samples.

[0108] When integrating data assets from multiple financial institutions, records of the same customer across different institutions are treated as positive sample pairs, and records of different customers are treated as negative sample pairs. A selected similarity calculation function (such as cosine similarity) is used to calculate the similarity between sample pairs. Based on these sample pairs, a CL-GNNAlign model is trained using a deep learning framework (such as PyTorch). During training, the model parameters are continuously adjusted through backpropagation to minimize the loss function, enabling the model to accurately mine the attribute similarity and structural similarity of cross-source entities. Finally, the trained CL-GNNAlign model is applied to entities in multi-source data. Combining attribute similarity (such as the degree of matching of attributes like customer name and ID number) and structural similarity (such as the adjacency relationship of entities in the knowledge graph), it is determined whether cross-source entities are the same entity. Entities with similarity exceeding a set threshold are aligned and merged, and the entity information in the knowledge graph is updated.

[0109] Furthermore, the path query embedding algorithm is as follows: The relation embeddings on the path are combined to obtain the path query embedding vector Pq, assuming the path is composed of relation r. 1, r2...,r n The composition and relational embedding are e respectively. r1, e r2 ...,e rn ,but:

[0110]

[0111] Where Pq is the final path query embedding vector, which integrates relational information along the path for subsequent relational inference; σ is the activation function, such as Sigmoid or ReLU, which introduces non-linearity into the model, enabling it to learn complex relationships; n represents the number of relations along the path, reflecting the path length; and W... i It is a weight matrix used to adjust the importance of different relation embeddings in the combination process, e ri is the embedding vector of the i-th relation on the path, containing semantic information of the relation, and b is the bias vector, which provides additional learnable parameters for the model and improves the model's expressive power.

[0112] Understandably, for complex relationships in knowledge graphs that require reasoning, path query embedding vectors are calculated based on the relationship embedding along the path, following the path query embedding (PQE) calculation method. Then, reinforcement learning algorithms are used to continuously optimize the reasoning strategy based on environmental feedback, thereby achieving automatic reasoning of complex relationships and supplementing the missing hidden relationships in the knowledge graph.

[0113] S274. If so, then the global knowledge graph to be evaluated will be used as the global knowledge graph.

[0114] S275. If not, the federated aggregation algorithm is used to aggregate the local model parameters of the second local model to generate the first global model parameters for the next round of iteration update operation.

[0115] Preferably, after using the final generated global knowledge graph to be evaluated as the global knowledge graph, the process further includes:

[0116] S2741. Obtain the first embedding vector generated by each data source in the last round of iterative update operation;

[0117] S2742. Use a hash algorithm to generate ownership metadata for each first embedding vector;

[0118] S2743. Construct a blockchain network and a smart contract for verifying the blockchain network, and store the ownership metadata in the blockchain network.

[0119] In a preferred embodiment of the present invention, during the knowledge graph construction process, ownership metadata is added to each entity and relation (first embedding vector). For example, for a set of financial transaction data, information such as the data owner, creation time, and authorization scope are used as ownership metadata.

[0120] Furthermore, a network based on blockchain platforms such as Hyperledger Fabric or Ethereum can be built, and smart contracts can be developed to store ownership metadata on the blockchain. When there is a data access or change request, the smart contract verifies the ownership information based on the hash value on the blockchain.

[0121] In a smart contract development environment (such as the Solidity language environment), define the ownership authorization policy according to the prescribed format. For example, define that internal employees of a financial institution can access the transaction summary information of specific customers during working hours on weekdays. Deploy the written smart contract to the blockchain network. When there is a data access request, the smart contract verifies it according to the defined authorization policy. If the request conforms to the policy, access is allowed and the access behavior is recorded; if it does not conform, access is denied.

[0122] Blockchain records every data access behavior, including access time, user ID, access operation type, and accessed data asset ID. These access records are collected and combined with a federated learning model to analyze data usage patterns. For example, by analyzing a user's access frequency and the type of data accessed over a period of time, thresholds can be set to determine if abnormal access behavior exists. Once an anomaly is detected, an early warning is issued and appropriate measures are taken.

[0123] In summary, this invention provides a data asset management method based on knowledge graphs. It involves using each data source as a federated learning node, constructing an initial local model based on the pre-built local knowledge graphs in each data source, and determining the initial local model parameters for each initial local model. A federated aggregation algorithm is then used to aggregate the initial local model parameters to generate initial global model parameters. It is understood that before parameter aggregation, the initial local models are trained based on the local knowledge graphs of their respective data sources, resulting in different model parameters. At this stage, the embedding vectors generated by the local models cannot be fused to generate a global knowledge graph. By employing the federated aggregation algorithm, the parameters of each local model are aggregated to generate initial global model parameters. The initial local model parameters are aggregated to generate initial global model parameters. Then, based on these initial global model parameters, the initial local model is iteratively updated repeatedly. This allows the local model to gradually unify the "rules" for extracting embedding vectors during the iterative update process, thereby constructing the global knowledge graph to be evaluated. As the "rules" for extracting embedding vectors become unified, the accuracy of constructing the global knowledge graph to be evaluated gradually improves, ultimately generating the global knowledge graph. After the global knowledge graph is constructed, it is updated by obtaining change data from the local knowledge graph during data asset management, rather than directly obtaining change data from the data assets, thus reducing the risk of data leakage. Therefore, this invention does not directly aggregate the original data from various data sources during the data asset management process of constructing and updating the knowledge graph, thus solving the data leakage problem inherent in current knowledge graph-based data asset management methods.

[0124] See Figure 2 This is an embodiment of a knowledge graph-based data asset management device provided by the present invention, comprising:

[0125] The data source monitoring module is used to monitor data change events of each data source that stores data assets. When a target data source with a data change event is determined to exist, the module obtains the change data of the target local knowledge graph of the target data source.

[0126] The knowledge graph construction and update module is used to update the current global knowledge graph based on the changed data and generate the updated target global knowledge graph.

[0127] The construction of the global knowledge graph includes:

[0128] Obtain several data sources;

[0129] Each of the aforementioned data sources is used as a federated learning node, and an initial local model is constructed based on the pre-built local knowledge graph in each of the aforementioned data sources, and the initial local model parameters of each of the initial local models are determined; wherein, the local knowledge graph is a knowledge graph constructed based on the data assets in the corresponding data source, including: several entities and the relationships between the entities; the initial local model is used to generate the embedding vectors of each entity and each relationship in each local knowledge graph.

[0130] The federated aggregation algorithm is used to aggregate the initial local model parameters to generate the initial global model parameters;

[0131] Based on the initial global model parameters, the initial local model is iteratively updated repeatedly. In each iteration, a global knowledge graph to be evaluated is constructed based on the embedding vectors generated by the updated local model. When the accuracy of the global knowledge graph to be evaluated meets a preset accuracy threshold, the final generated global knowledge graph to be evaluated is used as the global knowledge graph.

[0132] Preferably, the iterative update operation includes:

[0133] Obtain the first global model parameters; initially, the first global model parameters are the initial global model parameters.

[0134] The first global model parameters are distributed to each of the data sources so that the first local model in each of the data sources is updated according to the first global model parameters to generate a second local model, and the second local model is used to generate several first embedding vectors of the local knowledge graph; wherein, initially, the first local model is the initial local model;

[0135] Based on several first embedding vectors from each data source, a global knowledge graph to be evaluated is constructed, and it is determined whether the accuracy of the global knowledge graph to be evaluated meets a preset accuracy threshold.

[0136] If so, the global knowledge graph to be evaluated will be used as the global knowledge graph.

[0137] If not, the federated aggregation algorithm is used to aggregate the local model parameters of the second local model to generate the first global model parameters for the next round of iteration update operation.

[0138] Preferably, the first embedding vector includes: a first sub-embedded vector of the entity and a second sub-embedded vector of the relation;

[0139] The step of constructing a global knowledge graph to be evaluated based on several first embedding vectors from various data sources includes:

[0140] Using a pre-defined similarity recognition model, the attribute similarity and structural similarity between each first sub-embedded vector and other first sub-embedded vectors are calculated.

[0141] Based on the attribute similarity and the structural similarity, determine the comprehensive similarity between every two first sub-embedding vectors;

[0142] When the overall similarity is greater than a preset similarity threshold, the two entities of the corresponding two first sub-embedding vectors are aligned to generate a global knowledge graph to be optimized.

[0143] A path query embedding algorithm is used to perform path queries on the second sub-embedded vector in the global knowledge graph to be optimized, determine the hidden relationships in the global knowledge graph to be optimized, and optimize the global knowledge graph to be optimized based on the hidden relationships to generate a global knowledge graph to be evaluated.

[0144] See Figure 3 One embodiment of this application also provides a terminal device, including:

[0145] One or more processors;

[0146] A memory, coupled to the processor, for storing one or more programs;

[0147] When the one or more programs are executed by the one or more processors, the one or more processors implement a knowledge graph-based data asset management method as described above.

[0148] The processor controls the overall operation of the terminal device to complete all or part of the steps of the knowledge graph-based data asset management method described above. The memory stores various types of data to support the operation of the terminal device. This data may include, for example, instructions for any application or method operating on the terminal device, as well as application-related data. The memory can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0149] In an exemplary embodiment, the terminal device may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to execute a knowledge graph-based data asset management method as described in any of the foregoing embodiments, and achieve the same technical effects as the methods described above.

[0150] In another exemplary embodiment, a computer-readable storage medium including a computer program is also provided. When executed by a processor, the computer program implements the steps of a knowledge graph-based data asset management method as described in any of the foregoing embodiments. For example, the computer-readable storage medium may be the aforementioned memory including the computer program, which may be executed by a processor of a terminal device to complete the knowledge graph-based data asset management method as described in any of the foregoing embodiments and achieve the same technical effects as the aforementioned method.

[0151] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A data asset management method based on knowledge graphs, characterized in that, include: Listen for data change events in each data source that stores data assets. When a target data source with a data change event is identified, obtain the change data of the target local knowledge graph of the target data source. Based on the changed data, the current global knowledge graph is updated to generate the updated target global knowledge graph; The construction of the global knowledge graph includes: Obtain several data sources; Each of the aforementioned data sources is used as a federated learning node, and an initial local model is constructed based on the pre-built local knowledge graph in each of the aforementioned data sources, and the initial local model parameters of each of the initial local models are determined; wherein, the local knowledge graph is a knowledge graph constructed based on the data assets in the corresponding data source, including: several entities and the relationships between the entities; the initial local model is used to generate the embedding vectors of each entity and each relationship in each local knowledge graph. The federated aggregation algorithm is used to aggregate the initial local model parameters to generate the initial global model parameters; Based on the initial global model parameters, the initial local model is iteratively updated repeatedly. In each iteration, a global knowledge graph to be evaluated is constructed based on the embedding vectors generated by the updated local model. When the accuracy of the global knowledge graph to be evaluated meets a preset accuracy threshold, the final generated global knowledge graph to be evaluated is used as the global knowledge graph.

2. The data asset management method based on knowledge graphs as described in claim 1, characterized in that, The iterative update operation includes: Obtain the first global model parameters; initially, the first global model parameters are the initial global model parameters. The first global model parameters are distributed to each of the data sources so that the first local model in each of the data sources is updated according to the first global model parameters to generate a second local model, and the second local model is used to generate several first embedding vectors of the local knowledge graph; wherein, initially, the first local model is the initial local model; Based on several first embedding vectors from each data source, a global knowledge graph to be evaluated is constructed, and it is determined whether the accuracy of the global knowledge graph to be evaluated meets a preset accuracy threshold. If so, the global knowledge graph to be evaluated will be used as the global knowledge graph. If not, the federated aggregation algorithm is used to aggregate the local model parameters of the second local model to generate the first global model parameters for the next round of iteration update operation.

3. The data asset management method based on knowledge graphs as described in claim 2, characterized in that, The first embedding vector includes: a first sub-embedded vector of the entity and a second sub-embedded vector of the relation; The step of constructing a global knowledge graph to be evaluated based on several first embedding vectors from various data sources includes: Using a pre-defined similarity recognition model, the attribute similarity and structural similarity between each first sub-embedded vector and other first sub-embedded vectors are calculated. Based on the attribute similarity and the structural similarity, determine the comprehensive similarity between every two first sub-embedding vectors; When the overall similarity is greater than a preset similarity threshold, the two entities of the corresponding two first sub-embedding vectors are aligned to generate a global knowledge graph to be optimized. A path query embedding algorithm is used to perform path queries on the second sub-embedded vector in the global knowledge graph to be optimized, determine the hidden relationships in the global knowledge graph to be optimized, and optimize the global knowledge graph to be optimized based on the hidden relationships to generate a global knowledge graph to be evaluated.

4. The data asset management method based on knowledge graphs as described in claim 3, characterized in that, After using the final generated global knowledge graph to be evaluated as the global knowledge graph, the following is also included: Obtain the first embedding vector generated by each data source in the last round of iteration update operation; A hash algorithm is used to generate ownership metadata for each first embedding vector; Construct a blockchain network and smart contracts for verifying the blockchain network, and store the ownership metadata in the blockchain network.

5. The data asset management method based on knowledge graphs as described in claim 4, characterized in that, Before updating the current global knowledge graph based on the changed data, the method further includes: The smart contract is used to perform security verification on the changed data; Upon successful security verification, the current global knowledge graph is updated based on the changed data. If security verification fails, generate an alert for abnormal data changes.

6. A data asset management device based on knowledge graphs, characterized in that, include: The data source monitoring module is used to monitor data change events of each data source that stores data assets. When a target data source with a data change event is determined to exist, the module obtains the change data of the target local knowledge graph of the target data source. The knowledge graph construction and update module is used to update the current global knowledge graph based on the changed data and generate the updated target global knowledge graph. The construction of the global knowledge graph includes: Obtain several data sources; Each of the aforementioned data sources is used as a federated learning node, and an initial local model is constructed based on the pre-built local knowledge graph in each of the aforementioned data sources, and the initial local model parameters of each of the initial local models are determined; wherein, the local knowledge graph is a knowledge graph constructed based on the data assets in the corresponding data source, including: several entities and the relationships between the entities; the initial local model is used to generate the embedding vectors of each entity and each relationship in each local knowledge graph. The federated aggregation algorithm is used to aggregate the initial local model parameters to generate the initial global model parameters; Based on the initial global model parameters, the initial local model is iteratively updated repeatedly. In each iteration, a global knowledge graph to be evaluated is constructed based on the embedding vectors generated by the updated local model. When the accuracy of the global knowledge graph to be evaluated meets a preset accuracy threshold, the final generated global knowledge graph to be evaluated is used as the global knowledge graph.

7. A knowledge graph-based data asset management device as described in claim 6, characterized in that, The iterative update operation includes: Obtain the first global model parameters; initially, the first global model parameters are the initial global model parameters. The first global model parameters are distributed to each of the data sources so that the first local model in each of the data sources is updated according to the first global model parameters to generate a second local model, and the second local model is used to generate several first embedding vectors of the local knowledge graph; wherein, initially, the first local model is the initial local model; Based on several first embedding vectors from each data source, a global knowledge graph to be evaluated is constructed, and it is determined whether the accuracy of the global knowledge graph to be evaluated meets a preset accuracy threshold. If so, the global knowledge graph to be evaluated will be used as the global knowledge graph. If not, the federated aggregation algorithm is used to aggregate the local model parameters of the second local model to generate the first global model parameters for the next round of iteration update operation.

8. A knowledge graph-based data asset management device as described in claim 7, characterized in that, The first embedding vector includes: a first sub-embedded vector of the entity and a second sub-embedded vector of the relation; The step of constructing a global knowledge graph to be evaluated based on several first embedding vectors from various data sources includes: Using a pre-defined similarity recognition model, the attribute similarity and structural similarity between each first sub-embedded vector and other first sub-embedded vectors are calculated. Based on the attribute similarity and the structural similarity, determine the comprehensive similarity between every two first sub-embedding vectors; When the overall similarity is greater than a preset similarity threshold, the two entities of the corresponding two first sub-embedding vectors are aligned to generate a global knowledge graph to be optimized. A path query embedding algorithm is used to perform path queries on the second sub-embedded vector in the global knowledge graph to be optimized, determine the hidden relationships in the global knowledge graph to be optimized, and optimize the global knowledge graph to be optimized based on the hidden relationships to generate a global knowledge graph to be evaluated.

9. A terminal device, characterized in that, include: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a knowledge graph-based data asset management method as described in any one of claims 1-5.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a knowledge graph-based data asset management method as described in any one of claims 1-5.

Images