Privacy protection method and system for data cross-domain sharing and computer equipment

Abstract

The invention relates to the technical field of privacy protection, and particularly discloses a privacy protection method and system for data cross-domain sharing and computer equipment, and the method comprises the steps: constructing a global GNN model based on federated learning, distributing the constructed global GNN model to local clients, constructing sub-graphs of the local clients, and distributing the sub-graphs of the local clients to the local clients; configuring a local GNN model of the local client based on the global GNN model; training local client data based on a local sub-graph and a local GNN model to obtain a local model and generate node embedding, uploading the node embedding of the local model to a central server by each local client, performing weighted aggregation operation on the node embedding based on node similarity, and generating global node embedding; according to the method, parameter weights based on a federated learning framework are updated, and the privacy problem of data elements in cross-domain circulation and sharing in a multi-source heterogeneous service fusion scene is solved by introducing a sub-graph, a graph neural network and a federated learning technology.

Description

Technical Field

[0001] The present invention relates to the technical field of privacy protection, and specifically to a privacy protection method, system, and computer device for cross-domain data sharing. Background Art

[0002] In the current new stage of data element value creation, cross-domain data circulation has brought new opportunities for digital transformation. However, the circulation and sharing of data have also brought more and more challenges such as data security and privacy protection, data sovereignty and jurisdiction, and compliance of data transmission and storage.

[0003] Currently, there are three aspects of risks in the process of different institutions sharing with external data sources: First, a large amount of personal user information is involved, and strict regulatory requirements are imposed; second, the data assets and trade secrets accumulated by the institution's own business are easily leaked; third, due to the replicability and easy dissemination of data itself, once shared, it cannot be traced, and it is difficult to confirm the ownership of data assets, severely restricting commercialization.

[0004] As a key technology for data security and privacy protection, privacy computing has made remarkable progress in recent years. At the technical level, core technologies such as homomorphic encryption, secure multi-party computing, and differential privacy have been continuously mature, promoting the implementation and application of privacy computing in fields such as finance, healthcare, and government affairs. Although the current privacy computing technology has made remarkable progress, data sharing applications still face various challenges such as difficulties in interconnecting different platforms and low data openness. At the same time, the challenge of balancing privacy protection and data utility still needs to be further explored, especially in the scenarios of non-independent and identically distributed (Non-IID) data and high-dimensional data. Therefore, how to achieve secure sharing and circulation of data across borders, institutions, and departments on the premise of ensuring compliance of all parties in data circulation has become an urgent problem to be solved in the current data element field. Summary of the Invention

[0005] The purpose of the present invention is to provide a privacy protection method and system for cross-domain data sharing to solve the problems raised in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A privacy protection method for cross-domain data sharing, the method includes:

[0007] Construct a global GNN model based on federated learning, distribute the constructed global GNN model to each local client, construct a subgraph of the local client, and configure the local GNN model of the local client based on the global GNN model;

[0008] Train the local client data based on the local subgraph and the local GNN model, obtain the local model and generate node embeddings, and each local client uploads the node embeddings of the local model to the central server;

[0009] Construct a global spatial graph based on the node embeddings of the local models uploaded by each local client;

[0010] Calculate the similarity of the node embeddings, perform a weighted aggregation operation on the node embeddings based on the node similarity, and generate global node embeddings;

[0011] Update the parameter weights under the federated learning framework.

[0012] As a further solution of the present invention, the steps of constructing the subgraphs of each local client and configuring the local GNN model of the local client based on the global GNN model specifically include:

[0013] Obtain the node information of the local client and the data circulation relationship between the nodes;

[0014] Dynamically evolve and aggregate the data circulation relationship between the nodes to generate a subgraph.

[0015] As a further solution of the present invention, the step of dynamically evolving and aggregating the data circulation relationship between the nodes to generate a subgraph is specifically manifested as: at For the client And the layer directory = 0,..., L−1, a The GNN of layer is expressed as:

[0016] ; ;

[0017] Among them, Represents The node features of the client, Is the layer index, The node features of the client at the Layer are expressed as ; Is an aggregation function, which can vary according to different GNN variants; Represents the node Neighborhood set; Is the message generation function, and its input is the hidden state of the current node , the hidden state of the neighbor node And edge features ; Is the state update function, used to receive the aggregated features .

[0018] As a further solution of the present invention, during the graph-level representation, after L-layer message propagation, in the readout phase, the feature vector of the entire graph is calculated from the representations of all nodes according to the hidden state of the last layer of the message-passing neural network:

[0019] .

[0020] As a further solution of the present invention, the step of updating the parameter weights based on the federated learning framework specifically includes:

[0021] Set W = {M θ , U φ , R δ} as the overall learnable weight set in the graph neural network of client k, where M θ represents the parameter of the message-passing process, U φ represents the parameter in the node update process, and R δ represents the parameter in the readout phase. On this basis, adjust the weight W to minimize the objective function F(W), expressed in the following form:

[0022] ;

[0023] ;

[0024] where the function represents the local objective function of the client, which is used to measure the local empirical risk of the graph dataset with data samples, while represents the loss function of the global graph neural network.

[0025] As a further solution of the present invention, in the step of updating the parameter weights based on the federated learning framework, at time, then represents the node set of the subgraph, and each node corresponds to a cluster, is the set of edge weights, where the weight represents the normalized cosine similarity between nodes and , and satisfies ;

[0026] For ∀k ∈ K, the central node is updated in a weighted sum manner:

[0027] ;

[0028] where is a function for knowledge propagation, represents the number of times of the repeated propagation process, is the node after rounds of propagation and aggregation to obtain the updated model parameters, is the node and the weight between, when at this time, , indicating the model parameter state of the node under the initial propagation round, and the new model set is .

[0029] The present invention also provides a privacy protection system for cross - domain data sharing, which is used to implement the privacy protection method for cross - domain data sharing described above. The system includes:

[0030] A configuration module, which is used to construct a global GNN model based on federated learning, distribute the constructed global GNN model to each local client, construct a sub - graph of the local client, and configure the local GNN model of the local client based on the global GNN model;

[0031] An upload module, which is used to train the local client data based on the local sub - graph and the local GNN model, obtain the local model and generate node embeddings, and each local client uploads the node embeddings of the local model to the central server;

[0032] A global graph construction module, which is used to construct a global spatial graph based on the node embeddings of the local models uploaded by each local client;

[0033] An aggregation module, which is used to calculate the similarity of the node embeddings, perform a weighted aggregation operation on the node embeddings based on the node similarity, and generate global node embeddings;

[0034] An update module, which is used to update the parameter weights under the federated learning framework.

[0035] The present invention also provides a computer device, including a memory and a processor. When the computer program stored in the memory is executed by the processor, the processor executes the steps of the privacy protection method for cross - domain data sharing according to any one of claims 1 to 7.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows: The present method provides a privacy protection method for data cross - domain sharing. By introducing sub - graphs, graph neural networks, and federated learning technologies, it solves the privacy problems in the cross - domain circulation and sharing of data elements in the scenario of multi - source heterogeneous service integration, breaks the barriers of data islands, and meets the requirements in aspects such as data privacy and security in cross - border data circulation and sharing, improving the efficiency of data processing and analysis, promoting data cooperation and sharing, enhancing model performance and generalization ability, and meeting compliance and regulatory requirements. At the same time, it integrates multi - source heterogeneous data to achieve more efficient, accurate, and secure data processing and analysis, and promotes cross - domain and cross - border data cooperation and innovative applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings in the following descriptions are only some embodiments of the present invention.

[0038] Figure 1 It is a flowchart of a privacy protection method for data cross - domain sharing provided by an embodiment of the present invention.

[0039] Figure 2 It is a model architecture diagram provided by an embodiment of the present invention.

[0040] Figure 3 It is a schematic diagram of the working principle of federated learning provided by an embodiment of the present invention.

[0041] Figure 4 It is a process diagram of sub - graph construction provided by an embodiment of the present invention.

[0042] Figure 5 It is a block diagram of the composition structure of a privacy protection system for data cross - domain sharing provided by an embodiment of the present invention.

[0043] Figure 6 It is a block diagram of the composition structure of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] In order to make the technical problems, technical solutions, and beneficial effects to be solved by the present invention clearer, the following further details the present invention in conjunction with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0045] As Figure 1 Figure 2 、 Figure 3 shown, in an embodiment of the present invention, a privacy protection method for data cross - domain sharing, the method includes steps S100 to step S500:

[0046] Step S100: Build a global GNN model based on federated learning, distribute the built global GNN model to each local client, construct a subgraph of the local client, and configure the local GNN model of the local client based on the global GNN model;

[0047] Step S200: Train the local client data based on the local subgraph and the local GNN model, obtain the local model and generate node embeddings, and each local client uploads the node embeddings of the local model to the central server;

[0048] Step S300: Build a global spatial graph based on the node embeddings of the local models uploaded by each local client;

[0049] Step S400: Calculate the similarity of the node embeddings, perform a weighted aggregation operation on the node embeddings based on the node similarity, and generate global node embeddings;

[0050] Step S500: Update the parameter weights under the federated learning framework.

[0051] In this embodiment, the central server initializes the global GNN model parameters, defines the global graph structure and aggregation strategy, etc., so as to distribute the global model (GNN) to each client;

[0052] When the local client receives the global model from the central server, it uses the local subgraph and the local GNN model for training, and uses the trained local GNN to generate node embeddings, thereby obtaining the local model;

[0053] Each client uploads the node embeddings of the local model and updates them to the central server;

[0054] After the central server receives the node embeddings of each client, it constructs a global graph structure according to the node embeddings of each client, calculates the similarity of the node embeddings, etc., performs a weighted aggregation operation on the node embeddings using the node similarity, and generates global node embeddings; repeat the above steps.

[0055] In cross-border and cross-domain business scenarios of data elements, in addition to protecting user privacy, the randomness and subgraph heterogeneity characteristics shown in data flow and sharing are also the key areas we focus on. Because each node in the subgraph is deeply affected by its adjacent nodes This further affects the dynamic changes of the subgraph's structural characteristics and data statistical characteristics, and any changes in the subgraph's structure and characteristics will cause the global graph structure to continuously evolve over time. Therefore, the structural states of both the subgraph and the global graph are continuously evolving over time.

[0056] To this end, the role of subgraphs and GNNs in cross-border scenarios is to reveal the hidden structures in cross-domain sharing and learn the implicit features of subgraphs and between subgraphs (global graph). Privacy issues are ensured through federated learning, guaranteeing that each subgraph uploads only the results or parameters of training for each subgraph (client) to the central server, rather than directly uploading local data.

[0057] Such as Figure 2 , Figure 3 , Figure 4 shown, as a preferred embodiment of the present invention, the steps of constructing subgraphs for each local client and configuring the local GNN model of the local client based on the global GNN model specifically include:

[0058] Obtain the node information of the local client and the data flow relationship between the nodes;

[0059] Dynamically evolve and aggregate the data flow relationship between the nodes to generate a subgraph.

[0060] In this embodiment, the subsidiaries or countries where the data demanders and providers are located are abstracted as local subgraphs. The users within each branch are regarded as the nodes of the local subgraph, and the data flow relationship between the users is used as the edges of the local subgraph. The present invention will use GNN to automatically capture the key information between each node in the local subgraph, such as the structural features of the nodes: that is, the features related to the graph topology structure, namely information such as node degree, neighbor relationship, and connection pattern, as well as the statistical features of the nodes: such as the attribute features related to the nodes (user age, commodity category, etc.), and the attribute information of the edges (such as the interaction intensity between the user and the commodity).

[0061] Within the subgraph, information is transmitted between the data sender and the provider, and continuous aggregation operations are performed through the edge routers or servers of the subgraph, thereby completing the construction of the subgraph relationship. Each subgraph represents a client in federated learning, and its subgraph construction process is as follows:

[0062] Each subgraph represents a client in federated learning; the users within each branch are regarded as the nodes of the local subgraph, and the data flow relationship between the users is used as the edges of the local subgraph;

[0063] In each client, as the data continuously flows, the local model automatically captures the key information between each node in the local subgraph (information such as node location and edge attributes) through GNN;

[0064] Within the subgraph, as information is transmitted or flows between the data sender and the provider, continuous aggregation operations are performed through the edge routers or servers of the subgraph, thereby completing the construction of the local subgraph relationship.

[0065] Among them, each sub - graph represents a client of federated learning. In each client, as data continuously flows, the edge server or edge router continuously and dynamically completes the local construction of the sub - graph through the agg() operation. When the sub - graph reaches equilibrium at a certain moment, it obtains the node features of the sub - graph and the statistical features of the flowing data from the edge server to achieve data supervision and privacy protection within the sub - graph.

[0066] In this way, it can effectively capture important information in the local sub - graph and perform information transfer and aggregation operations between sub - graphs, thus constructing a complete sub - graph relationship. This federated learning method based on sub - graphs can ensure data privacy protection while achieving efficient model training and information sharing.

[0067] Collaborative training through federated learning ensures data privacy and solves the data silo problem. Each client has a GNN model to learn the graph representation and make corresponding predictions, etc. Multiple clients cooperate through a central server to improve their respective GNN models without disclosing the local graph datasets of the clients.

[0068] As a preferred embodiment of the present invention, the GNN consists of message propagation and neighborhood aggregation. Each node iteratively collects the information propagated by its neighbors and aggregates it with its own information to update its representation. For the dynamic evolution process of the global graph, at time, for client and layer directory = 0,..., L−1, the GNN representation of a layer is:

[0069] ; ;

[0070] Among them, represents the node features of the client, is the layer index, from which it can be obtained that: the node features of the th layer of the client are represented as ; is an aggregation function, which can vary according to different GNN variants; represents the neighborhood set of node ; is the message generation function, whose inputs are the hidden state of the current node , the hidden state of the neighbor node and the edge feature ; Is a status update function used to receive aggregated features .

[0071] When performing graph-level representation, after L layers of message propagation, in the readout phase, based on the hidden state of the last layer of the message-passing neural network, the feature vector of the entire graph is calculated from the representations of all nodes in the following manner, and various tasks can be accurately performed for downstream.

[0072] .

[0073] Among them, There are various implementation methods, including average pooling and sum pooling, etc. Its essential role is to aggregate the embedding information of all nodes on the graph, and then generate a unified embedding vector. This vector can comprehensively reflect the overall characteristics of the graph, thus supporting various tasks such as graph classification and regression.

[0074] Based on the above subgraph construction method, the edge server relationships between branches or between subgraphs are abstracted as nodes of the spatial graph structure, and the data circulation paths are abstracted as the edges of the spatial graph. In this spatial network structure, each node contains a subgraph structure, and these subgraphs are scattered and deployed on different edge servers. Due to the protection of different domain privacy or regulatory restrictions, the data on these servers cannot be centrally trained, but can be collaboratively trained through federated learning. Information is transmitted between the subgraph and the spatial graph through the global server and the boundary server of each subgraph.

[0075] Each client (node) has a private dataset and uses the GNN model for local training and prediction. With the assistance of the edge server, each client can cooperate with each other without directly sharing the dataset. Thus, it not only effectively protects data privacy, but also realizes model training and cooperation in federated learning.

[0076] This spatial graph structure can better understand the flow of data across domains and borders, and connect each subgraph with the spatial graph to achieve effective information transmission. This method provides a new perspective for cross-domain and cross-border federated learning, helps to better manage and optimize data flow, while protecting privacy and security.

[0077] As a preferred embodiment of the present invention, according to the above steps, through node features, a global network structure can be constructed with edge servers as nodes and the relationships of data flow between servers as edges. At the same time, to solve the data privacy problems of each subgraph, the present invention will construct a parameter weight update strategy based on the federated learning framework. For this purpose, set W = {M θ , U φ , R δ} is the set of overall learnable weights in the graph neural network for client k, where M θ Parameters representing the message passing process, U φ Parameters representing the node update process, R δ Parameters representing the readout phase. Based on this, the objective function of the global GNN, that is, adjusting the weights W to minimize the objective function F(W), is expressed in the following form:

[0078] ;

[0079] ;

[0080] Among them, the function represents the local objective function of the client, which is used to measure the local empirical risk of the graph dataset with data samples, while represents the loss function of the global graph neural network;

[0081] As mentioned before, at moment, for the constructed graph it is the case that represents the node set of the subgraph, and each node corresponds to a cluster, is the set of edge weights, where the weights represent the normalized cosine similarity between nodes and and satisfy ;

[0082] However, in the dynamic process of promoting the flow of data elements and establishing connections, given the constructed graph , for ∀k ∈ K, the central nodes are updated in a weighted sum manner to ensure that the dynamic adjustment of the graph structure is synchronized with the flow of data.

[0083] ;

[0084] Among them, is a function for knowledge dissemination, which uses the updated model parameters to achieve the circulation and sharing of information. represents the number of times of the repeated dissemination process, which determines the depth and breadth of knowledge dissemination in the graph. is the updated model parameters of node after rounds of dissemination and aggregation, which incorporates the information and wisdom from other nodes. And is the node and The weights between them reflect the importance and influence of these two nodes in the information propagation process. In particular, when at this time, represents the model parameter status of node in the initial propagation round. After the above propagation process, we obtain a new set of model collections named , which not only contains the information of the clients within the cluster but also integrates the weighted information of the nodes outside the cluster, enabling the knowledge in the entire graph structure to be comprehensively and accurately propagated and shared.

[0085] For the scenarios of cross - domain and cross - border circulation and sharing of data elements, a method for data circulation and sharing of heterogeneous federated graph neural networks is provided to meet the compliance requirements of "data does not move while the model moves" in the cross - border trusted data space. It realizes model training and data processing without sharing the original data, thereby protecting user privacy and data security.

[0086] In the process of promoting cross - border circulation and sharing of data elements, a major issue - personalized sub - graph collaboration problem is particularly introduced. This problem aims to flexibly and specifically collaborate to optimize relevant local models within the same branch according to the diversity characteristics of business data to meet the needs of different business scenarios.

[0087] For complex and variable multi - source heterogeneous data, not only is the heterogeneous data between different institutions or data sources integrated into a unified feature space to achieve efficient comparison and in - depth association, but further, the temporal changes and evolution process of dynamic graph - structured data are accurately captured, and the internal logic and value of the data are comprehensively analyzed from the micro - structure, providing new ideas for data analysis and decision - making.

[0088] As Figure 5 shown, the present invention also provides a privacy protection system for cross - domain data sharing, used to implement the privacy protection method for cross - domain data sharing. The system is characterized in that it includes:

[0089] A configuration module 100, configured to build a global GNN model based on federated learning, distribute the built global GNN model to each local client, build a sub - graph of the local client, and configure the local GNN model of the local client based on the global GNN model;

[0090] An upload module 200, configured to train the local client data based on the local sub - graph and the local GNN model, obtain a local model and generate node embeddings, and each local client uploads the node embeddings of the local model to the central server;

[0091] The global graph construction module 300 is used to construct a global spatial graph based on the local model node embeddings uploaded by each local client;

[0092] The aggregation module 400 is used to calculate the similarity of node embeddings, perform a weighted aggregation operation on the node embeddings based on the node similarity, and generate global node embeddings;

[0093] The update module 500 is used to update the parameter weights under the federated learning framework.

[0094] Figure 6 The internal structure diagram of a computer device in an embodiment is shown. The computer device includes a processor, a memory, a network interface, an input device, and a display screen connected through a system bus. Among them, the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and can also store a computer program. When the computer program is executed by the processor, the processor can implement a privacy protection method for cross-domain data sharing. The internal memory can also store a computer program. When the computer program is executed by the processor, the processor can execute a privacy protection method for cross-domain data sharing. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the outer shell of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0095] Those skilled in the art can understand that Figure 6 the structure shown in

[0096] is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements. Figure 6 In an embodiment, the privacy protection system for cross-domain data sharing provided by this application can be implemented in the form of a computer program, and the computer program can run on a computer device as shown in

[0097] In an embodiment, a computer device is proposed. The computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented:

[0098] Step S100: Build a global GNN model based on federated learning, distribute the built global GNN model to each local client, construct a subgraph of the local client, and configure the local GNN model of the local client based on the global GNN model;

[0099] Step S200: Train the local client data based on the local subgraph and the local GNN model to obtain a local model and generate node embeddings, and each local client uploads the node embeddings of the local model to the central server;

[0100] Step S300: Build a global spatial graph based on the node embeddings of the local models uploaded by each local client;

[0101] Step S400: Calculate the similarity of the node embeddings, perform a weighted aggregation operation on the node embeddings based on the node similarity, and generate global node embeddings;

[0102] Step S500: Update the parameter weights under the federated learning framework.

[0103] It should be understood that although the steps in the flowcharts of the embodiments of the present invention are shown in sequence according to the arrows, these steps do not necessarily have to be executed in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in each embodiment may include multiple sub-steps or multiple stages. These sub-steps or stages do not necessarily have to be executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages does not necessarily have to be sequential, but can be executed alternately or alternately with at least a part of other steps or sub-steps or stages of other steps.

[0104] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0105] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0106] The above-described embodiments merely represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent for the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention should be subject to the appended claims.

[0107] The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A privacy protection method for cross-domain data sharing, characterized in that, The method includes: Constructing a global GNN model based on federated learning, distributing the constructed global GNN model to each local client, constructing a subgraph of the local client, and configuring the local GNN model of the local client based on the global GNN model; Training the local client data based on the local subgraph and the local GNN model to obtain a local model and generate node embeddings, and each local client uploads the node embeddings of the local model to the central server; Constructing a global spatial graph based on the node embeddings of the local models uploaded by each local client; Calculating the similarity of the node embeddings, performing a weighted aggregation operation on the node embeddings based on the node similarity, and generating global node embeddings; Updating the parameter weights under the federated learning framework.

2. A privacy protection method for cross - domain data sharing according to claim 1, characterized in that, The steps of constructing the subgraphs of each local client and configuring the local GNN models of the local clients based on the global GNN model specifically include: Obtaining the node information of the local client and the data circulation relationship between the nodes; Dynamically evolving and aggregating the data circulation relationship between the nodes to generate a subgraph.

3. A privacy protection method for cross-domain data sharing according to claim 2, characterized in that The step of dynamically evolving and aggregating the data flow relationship between nodes to generate a sub - graph is specifically manifested as follows: At the moment, for the client and the layer directory of layer = 0,..., L−1, the GNN of one layer is expressed as: ； ； Among them, denotes the node features of the client, is the layer index, and the node features of the th layer of the client are denoted as ; is an aggregation function that can vary according to different GNN variants; denotes the neighborhood set of node ; is a message generation function whose inputs are the hidden state of the current node , the hidden state of the neighbor node and the edge feature ; is a state update function used to receive the aggregated features .

4. A privacy protection method for cross-domain data sharing according to claim 3, characterized in that, In the graph-level representation, after L-layer message propagation, in the readout stage, the feature vector of the entire graph is calculated from the representations of all nodes according to the hidden state of the last layer of the message passing neural network: 。 5. A privacy protection method for cross-domain data sharing according to claim 1, characterized in that, The steps of updating the parameter weights under the federated learning framework specifically include: Set \(W = \{M\ θ , U\ φ , R\ δ \}\) as the overall learnable weight set in the graph neural network of client \(k\), where \(M\ θ \) represents the parameters for the message passing process, \(U\ φ \) represents the parameters in the node update process, \(R\ δ \) represents the parameters in the readout phase. Based on this, the weights \(W\) are adjusted to minimize the objective function \(F(W)\), expressed in the following form: ； ； Among them, the function represents the local target function of the client, which is used to measure the local empirical risk of the graph dataset with a number of data samples, while represents the loss function of the global graph neural network.​ 6. A privacy protection method for cross-domain data sharing according to claim 5, characterized in that, In the step of updating the parameter weights based on the federated learning framework, at time, then represents the node set of the subgraph, and each node corresponds to a cluster. is the set of edge weights, where the weight represents the and normalized cosine similarity between nodes, and satisfies ; For ∀k ∈ K, the central node is updated in a weighted sum manner: ； Among them, is a function for knowledge dissemination, represents the number of times of the repeated dissemination process, is the node after rounds of dissemination and aggregation to obtain the updated model parameters, is the weight between the node and When then , indicating the model parameter state of the node at the initial dissemination round, and the new model set is .

7. A privacy protection system for cross - domain data sharing, which is used to implement the privacy protection method for cross - domain data sharing according to any one of claims 1 - 7, characterized in that, The system includes: A configuration module for constructing a global GNN model based on federated learning, distributing the constructed global GNN model to each local client, constructing a subgraph of the local client, and configuring the local GNN model of the local client based on the global GNN model; An upload module for training the local client data based on the local subgraph and the local GNN model to obtain a local model and generate node embeddings, and each local client uploads the node embeddings of the local model to the central server; A global graph construction module for constructing a global spatial graph based on the node embeddings of the local models uploaded by each local client; An aggregation module for calculating the similarity of the node embeddings, performing a weighted aggregation operation on the node embeddings based on the node similarity, and generating global node embeddings; An update module for updating the parameter weights under the federated learning framework.

8. A computer device, characterized in that, It includes a memory and a processor. When the computer program stored in the memory is executed by the processor, the processor executes the steps of the privacy protection method for data cross-domain sharing according to any one of claims 1 to 7.

Images