Model updating method, device, system and equipment

Abstract

The embodiments of this specification provide a model updating method, apparatus, system and equipment, wherein the method comprises: receiving an update request for a first feature extraction model, and determining a first feature vector corresponding to the first graph structure data based on the first feature extraction model to be trained and the first graph structure data, wherein the first graph structure data is determined by the first feature information of the target user; sending the first feature vector to a server, and receiving model parameters sent by the server, wherein the model parameters are obtained by the server by training a first classification model to be trained based on the first feature vector and the second feature vector, and obtaining parameters of the trained first classification model; updating the first feature extraction model based on the model parameters, and classifying the local user data based on the updated first feature extraction model.

Description

Technical Field

[0001] This document relates to the field of data processing technology, and in particular to a model updating method, device, system and equipment. Background Art

[0002] With the rapid development of computer technology, the types and quantities of application services provided by enterprises to users are increasing. As a result, the amount of user data is increasing and the data structure is becoming more complex. For this reason, enterprises often use graph neural network models built based on knowledge graphs to process complex user data.

[0003] After building a graph neural network model, the enterprise can update the model parameters of the graph neural network model based on local user data. However, the amount of user private data stored locally in the enterprise is small and the data quality is poor. In order to protect the privacy and security of user data, data collaboration cannot be achieved between different enterprises or even between different business units within the same enterprise. Therefore, updating the model parameters of the graph neural network model based on local user data results in poor accuracy in model parameter updates. Therefore, a solution that can improve the accuracy of model parameter updates is needed. Summary of the invention

[0004] The purpose of the embodiments of this specification is to provide a model updating method, apparatus, system and device, so as to provide a solution that can improve the updating accuracy of model parameters.

[0005] In order to implement the above technical solution, the embodiments of this specification are implemented as follows:

[0006] In a first aspect, a model updating method includes: receiving an update request for a first feature extraction model, and determining a first feature vector corresponding to the first graph structure data based on the first feature extraction model to be trained and first graph structure data, wherein the first graph structure data is determined by first feature information of a target user; sending the first feature vector to a server, and receiving model parameters sent by the server, wherein the model parameters are obtained by the server by training a first classification model to be trained based on the first feature vector and a second feature vector, and parameters of the trained first classification model are obtained, wherein the second feature vector is a vector determined by a target client based on a pre-trained second feature extraction model and second graph structure data, wherein the second graph structure data is determined by second feature information of the target user, wherein the second feature extraction model is obtained by training based on historical graph structure data, wherein the historical graph structure data is constructed by a preset number of historical feature information, and the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model; based on the model parameters, the first feature extraction model is updated, and local user data is classified based on the updated first feature extraction model.

[0007] In a second aspect, an embodiment of the present specification provides a model updating method, comprising: receiving a first feature vector sent by a first client and a second feature vector sent by a target client, wherein the first feature vector is a vector determined by the first client based on a first feature extraction model to be trained and first graph structure data, and the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, wherein the second graph structure data is determined by second feature information of the target user, and the second feature extraction model is trained based on historical graph structure data, and the historical graph structure data is constructed by a preset number of historical feature information; based on the first feature vector and the second feature vector, a first classification model to be trained is trained to obtain a trained first classification model, wherein the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model; and the model parameters of the trained first classification model are sent to the first client, so that the first client updates the first feature extraction model based on the model parameters, and processes the local user data of the first client based on the updated first feature extraction model.

[0008] In a third aspect, an embodiment of the present specification provides a model updating system, comprising a server and multiple first clients, wherein: the first client is used to receive an update request for a first feature extraction model, and determine a first feature vector corresponding to the first graph structure data based on the first feature extraction model to be trained and the first graph structure data, wherein the first graph structure data is determined by the first feature information of the target user; the server is used to receive the first feature vectors sent by the multiple first clients and the second feature vector sent by the target client, wherein the second feature vector is a vector determined by the target client based on the pre-trained second feature extraction model and the second graph structure data, wherein the second graph structure data is determined by the second feature information of the target user, and the second feature extraction model is It is obtained based on training of historical graph structure data, and the historical graph structure data is constructed by a preset number of historical feature information; the server is used to train the first classification model to be trained based on the first feature vector and the second feature vector to obtain the trained first classification model, and the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model, and send the model parameters of the trained first classification model to the multiple first clients; the first client is used to update the first feature extraction model based on the model parameters, and classify the local user data based on the updated first feature extraction model.

[0009] In a fourth aspect, a model updating method provided in an embodiment of the present specification is applied to a federated learning client, comprising: receiving an update request for a node classification model, and training the node classification model based on first graph structure data to obtain a trained node classification model; sending the first model parameters of the trained node classification model to a federated learning server, and receiving the target model parameters of the node classification model determined by the federated learning server based on the second model parameters sent by the target client and the first model parameters sent by the federated learning client, wherein the second model parameters are model parameters obtained by the target client by training the node classification model based on a predetermined amount of historical graph structure data; updating the trained node classification model based on the target model parameters to obtain an updated node classification model, and classifying local user data based on the updated node classification model.

[0010] In a fifth aspect, a model updating method provided in an embodiment of the present specification is applied to a federated learning server, comprising: receiving a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client based on the first graph structure data to train a node classification model to be trained, and the second model parameter is a model parameter obtained by the target client based on a predetermined amount of historical graph structure data to train the node classification model; determining a target model parameter of the node classification model based on the first model parameter and the second model parameter; sending the target model parameter to the federated learning client so that the federated learning client updates the trained node classification model based on the target model parameter, and processes the local user data of the federated learning client based on the updated node classification model.

[0011] In a sixth aspect, a model updating method provided in an embodiment of the present specification is applied to a blockchain system, comprising: based on a smart contract pre-deployed in the blockchain system, receiving a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client for training a node classification model to be trained based on first graph structure data, and the second model parameter is a model parameter obtained by the target client for training the node classification model based on a predetermined amount of historical graph structure data; based on the smart contract, the first model parameter and the second model parameter, determining a target model parameter of the node classification model; sending the target model parameter to the federated learning client so that the federated learning client updates the trained node classification model based on the target model parameter, and processes the local user data of the federated learning client based on the updated node classification model.

[0012] In a seventh aspect, an embodiment of the present specification provides a model update system, comprising a federated learning server and multiple federated learning clients, wherein: the federated learning client is used to receive an update request for a node classification model, and train the node classification model based on first graph structure data to obtain a trained node classification model; the federated learning server is used to receive a first model parameter sent by the federated learning client and a second model parameter sent by a target client, wherein the second model parameter is a model parameter obtained by the target client by training the node classification model based on a predetermined amount of historical graph structure data; the federated learning server is used to determine a target model parameter of the node classification model based on the first model parameter and the second model parameter, and send the target model parameter to the federated learning client; the federated learning client is used to update the trained node classification model based on the target model parameter to obtain an updated node classification model, and classify local user data based on the updated node classification model.

[0013] In an eighth aspect, an embodiment of the present specification provides a model updating device, the device comprising: a first receiving module, used to receive an update request for a first feature extraction model, and based on the first feature extraction model to be trained and the first graph structure data, determine a first feature vector corresponding to the first graph structure data, the first graph structure data being determined by the first feature information of the target user; a vector sending module, used to send the first feature vector to a server, and receive model parameters sent by the server, the model parameters are obtained by the server based on the first feature vector and the second feature vector to train a first classification model to be trained, and the second feature vector is the parameters of the trained first classification model, The client terminal determines a vector based on a pre-trained second feature extraction model and second graph structure data, the second graph structure data is determined by the second feature information of the target user, the second feature extraction model is trained based on historical graph structure data, and the historical graph structure data is constructed by a preset number of historical feature information, the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model; a model updating module is used to update the first feature extraction model based on the model parameters, and classify the local user data based on the updated first feature extraction model.

[0014] In a ninth aspect, an embodiment of the present specification provides a model updating device, the device comprising: a vector receiving module, used to receive a first feature vector sent by a first client and a second feature vector sent by a target client, the first feature vector being a vector determined by the first client based on a first feature extraction model to be trained and first graph structure data, the second feature vector being a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, the second graph structure data being determined by second feature information of the target user, the second feature extraction model being trained based on historical graph structure data, the historical graph structure data being constructed by a preset number of historical feature information; a model training module, used to train a first classification model to be trained based on the first feature vector and the second feature vector to obtain a trained first classification model, the first classification model being used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model; a data sending module, used to send model parameters of the trained first classification model to the first client, so that the first client updates the first feature extraction model based on the model parameters, and processes the local user data of the first client based on the updated first feature extraction model.

[0015] In the tenth aspect, an embodiment of the present specification provides a model updating device, comprising: a first receiving module, used to receive an update request for a node classification model, and train the node classification model based on first graph structure data to obtain a trained node classification model; a data sending module, used to send the first model parameters of the trained node classification model to a federated learning server, and receive the target model parameters of the node classification model determined by the federated learning server based on the second model parameters sent by the target client and the first model parameters sent by the federated learning client, wherein the second model parameters are model parameters obtained by the target client by training the node classification model based on a predetermined amount of historical graph structure data; a model updating module, used to update the trained node classification model based on the target model parameters to obtain an updated node classification model, and classify local user data based on the updated node classification model.

[0016] In the eleventh aspect, an embodiment of the present specification provides a model updating device, comprising: a data receiving module, used to receive a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client based on the first graph structure data to train a node classification model to be trained, and the second model parameter is a model parameter obtained by the target client based on a predetermined amount of historical graph structure data to train the node classification model; a parameter determination module, used to determine a target model parameter of the node classification model based on the first model parameter and the second model parameter; a parameter sending module, used to send the target model parameter to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameter, and processes the local user data of the federated learning client based on the updated node classification model.

[0017] In the twelfth aspect, an embodiment of the present specification provides a model updating device, which is a device in a blockchain system, comprising: a data receiving module, used to receive a first model parameter sent by a federated learning client and a second model parameter sent by a target client based on a smart contract pre-deployed in the blockchain system, the first model parameter being a model parameter obtained by the federated learning client for training a node classification model to be trained based on first graph structure data, and the second model parameter being a model parameter obtained by the target client for training the node classification model based on a predetermined amount of historical graph structure data; a parameter determination module, used to determine a target model parameter of the node classification model based on the smart contract, the first model parameter and the second model parameter; a parameter sending module, used to send the target model parameter to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameter, and processes the local user data of the federated learning client based on the updated node classification model.

[0018] In a thirteenth aspect, an embodiment of the present specification provides a model updating device, the model updating device comprising: a processor; and a memory arranged to store computer executable instructions, the executable instructions, when executed, cause the processor to: receive an update request for a first feature extraction model, and determine a first feature vector corresponding to the first graph structure data based on the first feature extraction model to be trained and the first graph structure data, the first graph structure data being determined by the first feature information of the target user; send the first feature vector to a server, and receive model parameters sent by the server, the model parameters being trained by the server based on the first feature vector and the second feature vector to train a first classification model to be trained, and the trained first feature vector is obtained; Parameters of the classification model, the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, the second graph structure data is determined by the second feature information of the target user, the second feature extraction model is trained based on historical graph structure data, and the historical graph structure data is constructed by a preset number of historical feature information, the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model; based on the model parameters, the first feature extraction model is updated, and the local user data is classified based on the updated first feature extraction model.

[0019] In a fourteenth aspect, an embodiment of the present specification provides a model updating device, the model updating device comprising: a processor; and a memory arranged to store computer executable instructions, the executable instructions, when executed, causing the processor to: receive a first feature vector sent by a first client and a second feature vector sent by a target client, the first feature vector being a vector determined by the first client based on a first feature extraction model to be trained and first graph structure data, the second feature vector being a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, the second graph structure data being determined by second feature information of the target user, and the second feature extraction model being a vector determined by a pre-trained second feature extraction model and second graph structure data based on historical graph structure data The historical graph structure data is constructed by a preset number of historical feature information; based on the first feature vector and the second feature vector, the first classification model to be trained is trained to obtain the trained first classification model, and the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model; the model parameters of the trained first classification model are sent to the first client, so that the first client updates the first feature extraction model based on the model parameters, and processes the local user data of the first client based on the updated first feature extraction model.

[0020] In a fifteenth aspect, an embodiment of the present specification provides a model updating device, comprising: a processor; and a memory arranged to store computer executable instructions, wherein the executable instructions, when executed, cause the processor to: receive an update request for a node classification model, and train the node classification model based on first graph structure data to obtain a trained node classification model; send a first model parameter of the trained node classification model to a federated learning server, and receive a target model parameter of the node classification model determined by the federated learning server based on a second model parameter sent by a target client and a first model parameter sent by the federated learning client, wherein the second model parameter is a model parameter obtained by the target client by training the node classification model based on a predetermined amount of historical graph structure data; update the trained node classification model based on the target model parameter to obtain an updated node classification model, and classify local user data based on the updated node classification model.

[0021] In a sixteenth aspect, an embodiment of the present specification provides a model updating device, comprising: a processor; and a memory arranged to store computer executable instructions, wherein the executable instructions, when executed, cause the processor to: receive a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client through training a node classification model to be trained based on first graph structure data, and the second model parameter is a model parameter obtained by the target client through training the node classification model based on a predetermined amount of historical graph structure data; determine a target model parameter of the node classification model based on the first model parameter and the second model parameter; send the target model parameter to the federated learning client so that the federated learning client updates the trained node classification model based on the target model parameter, and processes the local user data of the federated learning client based on the updated node classification model.

[0022] In the seventeenth aspect, an embodiment of the present specification provides a model updating device, which is a device in a blockchain system, comprising: a processor; and a memory arranged to store computer executable instructions, wherein the executable instructions, when executed, cause the processor to: based on a smart contract pre-deployed in the blockchain system, receive a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client for training a node classification model to be trained based on first graph structure data, and the second model parameter is a model parameter obtained by the target client for training the node classification model based on a predetermined amount of historical graph structure data; based on the smart contract, the first model parameter and the second model parameter, determine a target model parameter of the node classification model; send the target model parameter to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameter, and processes the local user data of the federated learning client based on the updated node classification model. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0024] Figure 1AA flowchart of an embodiment of a model updating method of this specification;

[0025] Figure 1B A schematic diagram of a processing process of a model updating method in this specification;

[0026] Figure 2 A schematic diagram of a model updating method in this specification;

[0027] Figure 3 A schematic diagram of a processing process of another model updating method of this specification;

[0028] Figure 4A A flowchart of another model updating method embodiment of the present specification;

[0029] Figure 4B A schematic diagram of a processing process of another model updating method of this specification;

[0030] Figure 5 A schematic diagram of a processing process of another model updating method of this specification;

[0031] Figure 6 A schematic diagram of a model updating system in this specification;

[0032] Fig. 7A A flowchart of another model updating method embodiment of the present specification;

[0033] Figure 7B A schematic diagram of a processing process of another model updating method of this specification;

[0034] Figure 8 A schematic diagram of a processing process of another model updating method of this specification;

[0035] Fig.9A A flowchart of another model updating method embodiment of the present specification;

[0036] Fig. 9B A schematic diagram of a processing process of another model updating method of this specification;

[0037] Fig. 10A A flowchart of another model updating method embodiment of the present specification;

[0038] Fig. 10B A schematic diagram of a processing process of another model updating method of this specification;

[0039] Fig.11 This is a schematic diagram of the structure of a model updating device embodiment of this specification;

[0040] Fig.12 A schematic diagram of the structure of another model updating device embodiment of the present specification;

[0041] Fig.13 A schematic diagram of the structure of another model updating device embodiment of the present specification;

[0042] Fig.14 A schematic diagram of the structure of another model updating device embodiment of the present specification;

[0043] Fig.15 A schematic diagram of the structure of another model updating device embodiment of the present specification;

[0044] Fig.16 This is a schematic diagram of the structure of a model updating device in this manual. DETAILED DESCRIPTION

[0045] The embodiments of this specification provide a model updating method, device, system and equipment.

[0046] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0047] Embodiment 1

[0048] like Figure 1A and 1B As shown, the embodiment of this specification provides a model updating method, the execution subject of the method can be a client, and the client can be a server or a terminal device, wherein the server can be an independent server or a server cluster composed of multiple servers, and the terminal device can be a device such as a personal computer, or a mobile terminal device such as a mobile phone or a tablet computer. The method can specifically include the following steps:

[0049] In S102, an update request for a first feature extraction model is received, and a first feature vector corresponding to the first graph structure data is determined based on the first feature extraction model to be trained and the first graph structure data.

[0050] Among them, the first feature extraction model can be a model constructed based on a preset deep learning algorithm, which is used to perform feature extraction processing on the graph structure data to obtain a feature vector corresponding to the graph structure data. The first graph structure data can be determined by the first feature information of the target user. The first feature information may include the target user's attribute information (such as gender, age, region, etc.), behavior information (such as purchasing behavior, calling behavior, etc.), etc.

[0051] In practice, with the rapid development of computer technology, the types and quantities of application services provided by enterprises to users are increasing. As a result, the amount of user data is increasing, and the data structure is becoming more and more complex. For this reason, enterprises often use graph neural network models built based on knowledge graphs to process complex user data. After building a graph neural network model, the enterprise can update the model parameters of the graph neural network model based on local user data. However, the amount of user private data stored locally in the enterprise is small, the data quality is poor, and in order to protect the privacy and security of user data, data collaboration cannot be achieved between different enterprises or even between different business units within the same enterprise. Therefore, updating the model parameters of the graph neural network model based on local user data makes the model parameter update accuracy poor. Therefore, a solution that can improve the accuracy of model parameter updates is needed. To this end, the embodiments of this specification provide a technical solution that can solve the above problems. For details, please refer to the following content.

[0052] Taking the client used by an institution that can provide resource transfer services to users as an example, since different institutions have different user private data, and the user private data owned by each institution may have small data volume and poor data quality, in order to make the first feature extraction model used locally achieve better use results (that is, the feature vector extracted based on the first feature extraction model meets the preset classification processing requirements), the model parameters of the first feature extraction model used locally can be updated through data interaction with the server.

[0053] For example, taking the first feature extraction model as a model built based on a graph neural network (GNN), the first feature information such as attribute information and behavior information of the target user stored locally can be obtained, and the first graph structure data can be constructed based on the first feature information. When an update request for the first feature extraction model is received, the first graph structure data can be input into the GNN model to obtain a first feature vector corresponding to the first graph structure data. Among them, the update request for the first feature extraction model can be triggered based on a preset model update cycle, and the data utilization rate of the first feature information of the target user can be improved through the first graph structure data.

[0054] In S104, the first feature vector is sent to the server, and the model parameters sent by the server are received.

[0055] Among them, the model parameters can be obtained by the server based on the first feature vector and the second feature vector to train the first classification model to be trained, and the parameters of the trained first classification model can be obtained. The second feature vector can be a vector determined by the target client based on the pre-trained second feature extraction model and the second graph structure data. The second graph structure data can be determined by the second feature information of the target user. The second feature extraction model can be a model constructed based on a preset deep learning algorithm, which is used to perform feature extraction processing on the graph structure data to obtain a feature vector corresponding to the graph structure data. The second feature extraction model can be trained based on historical graph structure data, and the historical graph structure data is constructed by a preset number of historical feature information. The first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model. The model parameters can be the gradient, loss, etc. of the trained first classification model.

[0056] In implementation, the same user may have data exchanges with multiple service providers, and different service providers serve different numbers of users (i.e., the amount of data processed) and different data processing capabilities. Therefore, the client of a service provider with strong data processing capabilities and large data processing volume can be used as the target client, that is, the target client can be a client that stores a pre-trained second feature extraction model, and the historical graph structure data used to train the second feature extraction model is constructed from a preset number of historical feature information, that is, the second feature extraction model of the target client is a model with high extraction accuracy.

[0057] like Figure 2 As shown, the server can receive the first feature vector sent locally (i.e., client 1) and the second feature vector extracted by the target client based on the second graph structure data and a pre-trained second feature extraction model, train the first classification model to be trained, and obtain the parameters of the trained first classification model.

[0058] For example, assuming that the target user has a data exchange relationship with Bank 1 and Bank 2, and the number of users of Bank 1 is smaller than that of Bank 2, and the data processing capacity of Bank 1 is smaller than that of Bank 2, therefore, the second feature information of the target user stored in Bank 2 can be obtained based on the target client of Bank 2, and the second graph structure data can be constructed based on the second feature information, and then the second feature vector can be determined based on the pre-trained second feature extraction model and the second graph structure data. The client of Bank 1 can obtain the first feature information corresponding to the target user upon receiving an update request for the first feature extraction model, and construct the first graph structure data based on the first feature information, and then determine the first feature vector based on the first feature extraction model to be trained and the first graph structure data. The server can train the first classification model to be trained based on the received first feature vector and the second feature vector, obtain the model parameters of the trained first classification model, and return the model parameters to the client of Bank 1.

[0059] In this way, since the second feature extraction model of the target client is trained based on the historical graph structure data constructed by a predetermined number of historical feature information, the feature extraction effect of the second feature extraction model of the target client is good, that is, with the assistance of the second feature extraction model, the classification effect of the first classification model trained by the server based on the first feature vector and the second feature vector is good.

[0060] In S106, the first feature extraction model is updated based on the model parameters, and the local user data is classified based on the updated first feature extraction model.

[0061] In implementation, the first feature extraction model can be updated based on the model parameters, so that the model effect of the first feature extraction model can be improved, that is, the classification accuracy of local user data is high based on the feature vector extracted by the updated first feature extraction model.

[0062] In addition, to ensure security during data transmission, the server can generate a client identifier for the client, and generate first verification information corresponding to the client based on a preset verification information generation algorithm (such as a preset hash algorithm), model parameters and client identifier.

[0063] In this way, the client can generate second verification information based on the locally stored client identifier, the preset verification information generation algorithm and the received model parameters, and perform verification processing based on the second verification information and the received first verification information to determine whether the model parameters have been tampered with during transmission.

[0064] After the verification is passed, the first feature extraction model is updated based on the model parameters to obtain the updated first feature extraction model.

[0065] An embodiment of the present specification provides a model updating method, receiving an update request for a first feature extraction model, and determining a first feature vector corresponding to the first graph structure data based on the first feature extraction model to be trained and the first graph structure data, the first graph structure data being determined by the first feature information of the target user, sending the first feature vector to a server, and receiving model parameters sent by the server, the model parameters being obtained by the server by training a first classification model to be trained based on the first feature vector and the second feature vector, the second feature vector being a vector determined by the target client based on a pre-trained second feature extraction model and the second graph structure data, the second graph structure data being determined by the second feature information of the target user, the second feature extraction model being obtained by training based on the historical graph structure data, the historical graph structure data being constructed by a preset number of historical feature information, the first classification model being used to classify the feature information of the target user based on the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model, updating the first feature extraction model based on the model parameters, and classifying the local user data based on the updated first feature extraction model. In this way, since the second feature extraction model of the target client is trained based on the historical graph structure data constructed by a preset number of historical feature information, the training accuracy of the first classification model can be improved with the participation of the second feature extraction model of the target client, and then the first feature extraction model is updated based on the model parameters of the trained first classification model, which can improve the feature extraction accuracy of the updated first feature extraction model and improve the classification accuracy of subsequent classification of local data. At the same time, since the client only sends the first feature vector to the server, the privacy security of local user private data is guaranteed.

[0066] Embodiment 2

[0067] like Figure 3 As shown, the embodiment of this specification provides a model updating method, the execution subject of the method can be a client, and the client can be a server or a terminal device, wherein the server can be an independent server or a server cluster composed of multiple servers, and the terminal device can be a device such as a personal computer, or a mobile terminal device such as a mobile phone or a tablet computer. The method can specifically include the following steps:

[0068] In S102, an update request for a first feature extraction model is received, and a first feature vector corresponding to the first graph structure data is determined based on the first feature extraction model to be trained and the first graph structure data.

[0069] In S104, the first feature vector is sent to the server, and the model parameters sent by the server are received.

[0070] In S106, the first feature extraction model is updated based on the model parameters, and the local user data is classified based on the updated first feature extraction model.

[0071] In S302, fourth graph structure data is obtained.

[0072] The fourth graph structure data may be constructed from historical feature information stored locally. For example, the fourth graph structure data may be constructed from historical feature information of all users stored locally in the past month.

[0073] In S304, the target classification model is trained based on the fourth graph structure data to obtain a pre-trained target classification model.

[0074] Among them, the target classification model can be obtained by training a model constructed by a preset classification sub-model and the first feature extraction model after update processing based on the fourth graph structure data.

[0075] In implementation, the preset classification sub-model can be a classification model constructed by a preset deep learning algorithm. The fourth graph structure data can be input into the first feature extraction model after update processing to obtain a feature vector corresponding to the fourth graph structure data. The feature vector is then input into the preset classification sub-model to perform model training on the target classification model to obtain a trained target classification model.

[0076] In addition, the trained first classification model sent by the server can also be received, that is, the preset classification sub-model can be the trained first classification model sent by the server. That is, the target classification model constructed by the updated first feature extraction model and the trained first classification model can be personalized trained based on the locally stored fourth graph structure data to improve the matching degree between the trained target classification model and the local data processing requirements, and can also effectively reduce the dependence on data from other domains when scoring the model.

[0077] In S306 , a node classification request for the third graph structure data is received.

[0078] The third graph structure data may be any graph structure data to be classified.

[0079] In S308, the third graph structure data is input into a pre-trained target classification model to obtain a node classification result corresponding to the third graph structure data.

[0080] In implementation, since the target classification model is trained based on the fourth graph structure data stored locally, the pre-trained target classification model has a high degree of match with local data processing requirements. Therefore, the third graph structure data is classified and processed based on the pre-trained target classification model, and the node classification results obtained are highly accurate.

[0081] An embodiment of the present specification provides a model updating method, receiving an update request for a first feature extraction model, and determining a first feature vector corresponding to the first graph structure data based on the first feature extraction model to be trained and the first graph structure data, the first graph structure data being determined by the first feature information of the target user, sending the first feature vector to a server, and receiving model parameters sent by the server, the model parameters being obtained by the server by training a first classification model to be trained based on the first feature vector and the second feature vector, the second feature vector being a vector determined by the target client based on a pre-trained second feature extraction model and the second graph structure data, the second graph structure data being determined by the second feature information of the target user, the second feature extraction model being obtained by training based on the historical graph structure data, the historical graph structure data being constructed by a preset number of historical feature information, the first classification model being used to classify the feature information of the target user based on the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model, updating the first feature extraction model based on the model parameters, and classifying the local user data based on the updated first feature extraction model. In this way, since the second feature extraction model of the target client is trained based on the historical graph structure data constructed by a preset number of historical feature information, the training accuracy of the first classification model can be improved with the participation of the second feature extraction model of the target client, and then the first feature extraction model is updated based on the model parameters of the trained first classification model, which can improve the feature extraction accuracy of the updated first feature extraction model and improve the classification accuracy of subsequent classification of local data. At the same time, since the client only sends the first feature vector to the server, the privacy security of local user private data is guaranteed.

[0082] Embodiment 3

[0083] like Figure 4A and Figure 4BAs shown, the embodiment of this specification provides a model updating method, the execution subject of the method can be a server, and the server can be a server or a terminal device, wherein the server can be an independent server or a server cluster composed of multiple servers, and the terminal device can be a device such as a personal computer, or a mobile terminal device such as a mobile phone or a tablet computer. The method can specifically include the following steps:

[0084] In S402, a first feature vector sent by a first client and a second feature vector sent by a target client are received.

[0085] Among them, the first feature vector can be a vector determined by the first client based on the first feature extraction model to be trained and the first graph structure data, the second feature vector can be a vector determined by the target client based on the pre-trained second feature extraction model and the second graph structure data, the second graph structure data can be determined by the second feature information of the target user, the second feature extraction model can be trained based on historical graph structure data, and the historical graph structure data can be constructed by a preset number of historical feature information.

[0086] In implementation, the server may receive first feature vectors sent by one or more first clients and second feature vectors sent by one or more target clients. The embodiments of this specification do not specifically limit the number of first clients and target clients.

[0087] In S404, the first classification model to be trained is trained based on the first feature vector and the second feature vector to obtain a trained first classification model.

[0088] Among them, the first classification model can be a classification model constructed based on a preset machine learning algorithm, which is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model. For example, the first classification model can be a classification model constructed based on a neural network algorithm, which is used to classify the feature information of the target user according to the first feature vector and the second feature vector.

[0089] In implementation, the first feature vector and the second feature vector may be input into a first classification model for model training to obtain a trained first classification model.

[0090] In S406, the model parameters of the trained first classification model are sent to the first client, so that the first client updates the first feature extraction model based on the model parameters, and processes the local user data of the first client based on the updated first feature extraction model.

[0091] In implementation, since the second classification model of the target client is a pre-trained model, the server only sends the model parameters of the trained first classification model to the first client, so that the first client updates the first feature extraction model based on the model parameters.

[0092] In addition, the server can also receive a model parameter update request from the target client, that is, the server can send the model parameters of the trained first classification model to the target client in response to the model parameter update request, so that the target client updates the second feature extraction model based on the model parameters, and processes the local user data of the target client based on the updated second feature extraction model.

[0093] An embodiment of the present specification provides a model updating method, receiving a first feature vector sent by a first client and a second feature vector sent by a target client, wherein the first feature vector is a vector determined by the first client based on a first feature extraction model to be trained and first graph structure data, and the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, the second graph structure data is determined by second feature information of a target user, the second feature extraction model is trained based on historical graph structure data, and the historical graph structure data is constructed by a preset number of historical feature information, and a first classification model to be trained is trained based on the first feature vector and the second feature vector to obtain a trained first classification model, the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model, and the model parameters of the trained first classification model are sent to the first client, so that the first client updates the first feature extraction model based on the model parameters, and processes the local user data of the first client based on the updated first feature extraction model. In this way, since the second feature extraction model of the target client is trained based on the historical graph structure data constructed by a preset number of historical feature information, the training accuracy of the first classification model can be improved with the participation of the second feature extraction model of the target client. The first client can update the first feature extraction model based on the model parameters of the trained first classification model, which can improve the feature extraction accuracy of the updated first feature extraction model and improve the classification accuracy of the subsequent classification processing of the local data of the first client. At the same time, since the client (i.e., the first client and the target client) only sends the feature vector to the server, the privacy security of the private data of the local user of the client is guaranteed.

[0094] Embodiment 4

[0095] like Figure 5As shown, the embodiment of this specification provides a model updating method, the execution subject of the method can be a server, and the server can be a server or a terminal device, wherein the server can be an independent server or a server cluster composed of multiple servers, and the terminal device can be a device such as a personal computer, or a mobile terminal device such as a mobile phone or a tablet computer. The method can specifically include the following steps:

[0096] In S502, a target user is determined based on user identifiers sent by multiple first clients and target clients.

[0097] In implementation, the first client and the target client may send locally stored user identifiers to the server, and the server may match the user identifiers sent by the first client and the target client, and determine the user corresponding to the user identifier shared by the first client and the target client as the target user.

[0098] For example, assuming that the first client and the target client use the user's communication number as the user identifier, the first client and the target client can send the locally stored communication number to the server, and the server can determine the user corresponding to the communication number shared by the target client and the first client as the target user.

[0099] In addition, to protect the privacy and security of user private data, the user identifier received by the server can be an identifier processed based on a preset encryption algorithm, that is, the first client and the target client can encrypt the locally stored user identifier based on a pre-agreed encryption algorithm, and send the encrypted user identifier to the server. The server can determine the target user based on the encrypted user identifier.

[0100] The above-mentioned method for determining the target user is an optional and feasible determination method. In actual application scenarios, there may be a variety of different determination methods, which may vary according to different actual application scenarios. The embodiments of this specification do not specifically limit this.

[0101] In S504, the user identification of the target user is sent to the first client, so that the first client determines the first characteristic information of the target user based on the user identification of the target user, and constructs the first graph structure data based on the first characteristic information of the target user.

[0102] In S506, the user identification of the target user is sent to the target client, so that the target client determines the second characteristic information of the target user based on the user identification of the target user, and constructs the second graph structure data based on the second characteristic information of the target user.

[0103] In implementation, in order to ensure security during data transmission, the server can generate corresponding client identifiers for the target client and the first client respectively, and based on a preset verification information generation algorithm (such as a preset hash algorithm), the user identifier of the target user and the client identifier, generate target verification information corresponding to the client, so that the client (i.e., the first client and the target client) can perform security verification on the received user identifier of the target user based on the target verification information.

[0104] In S402, a first feature vector sent by a first client and a second feature vector sent by a target client are received.

[0105] In S508, the first eigenvector and the second eigenvector are merged to obtain a third eigenvector.

[0106] In implementation, the server may merge the first feature vector and the second feature vector to obtain a third feature vector. In addition, there may be multiple methods for determining the third feature vector. For example, when the dimensions of the first feature vector and the second feature vector are the same, the mean of the first feature vector and the second feature vector may be determined as the third feature vector. Alternatively, the first feature vector and the second feature vector may be processed by a preset regression algorithm to obtain the third feature vector. The method for determining the third feature vector may vary depending on the actual application scenario, and the embodiments of this specification do not specifically limit this.

[0107] In addition, to protect the privacy and security of user private data, the first feature vector may be a vector obtained by the first client encrypting the target vector based on a preset privacy encryption algorithm, and the target vector may be a vector obtained by the first client performing feature extraction on the first graph structure data through a first feature extraction model to be trained.

[0108] Among them, the preset privacy encryption algorithm can be an encryption algorithm such as a homomorphic encryption (HE) algorithm, a differential privacy (DP) encryption algorithm, etc.

[0109] In S510, the first classification model to be trained is trained based on the third feature vector to obtain a trained first classification model.

[0110] In S406, the model parameters of the trained first classification model are sent to the first client, so that the first client updates the first feature extraction model based on the model parameters, and processes the local user data of the first client based on the updated first feature extraction model.

[0111] The embodiment of the present specification provides a model updating method. The embodiment of the present specification provides a model updating method, receiving a first feature vector sent by a first client and a second feature vector sent by a target client, the first feature vector is a vector determined by the first client based on a first feature extraction model to be trained and first graph structure data, the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, the second graph structure data is determined by second feature information of the target user, the second feature extraction model is trained based on historical graph structure data, the historical graph structure data is constructed by a preset number of historical feature information, based on the first feature vector and the second feature vector, a first classification model to be trained is trained to obtain a trained first classification model, the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model, the model parameters of the trained first classification model are sent to the first client, so that the first client updates the first feature extraction model based on the model parameters, and processes the local user data of the first client based on the updated first feature extraction model. In this way, since the second feature extraction model of the target client is trained based on the historical graph structure data constructed by a preset number of historical feature information, the training accuracy of the first classification model can be improved with the participation of the second feature extraction model of the target client. The first client can update the first feature extraction model based on the model parameters of the trained first classification model, which can improve the feature extraction accuracy of the updated first feature extraction model and improve the classification accuracy of the subsequent classification processing of the local data of the first client. At the same time, since the client (i.e., the first client and the target client) only sends the feature vector to the server, the privacy security of the private data of the local user of the client is guaranteed.

[0112] Embodiment 5

[0113] The embodiment of the present specification provides a model updating system, the model updating system comprising: a server and a plurality of first clients, wherein:

[0114] The first client is used to receive an update request for a first feature extraction model, and determine a first feature vector corresponding to the first graph structure data based on the first feature extraction model to be trained and the first graph structure data, where the first graph structure data is determined by first feature information of a target user.

[0115] The server is used to receive first feature vectors sent by multiple first clients and second feature vectors sent by target clients, where the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, the second graph structure data is determined by the second feature information of the target user, the second feature extraction model is trained based on historical graph structure data, and the historical graph structure data is constructed by a preset number of historical feature information.

[0116] The server is used to train the first classification model to be trained based on the first feature vector and the second feature vector to obtain the trained first classification model. The first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model, and send the model parameters of the trained first classification model to multiple first clients.

[0117] The first client is used to update the first feature extraction model based on the model parameters, and classify the local user data based on the updated first feature extraction model.

[0118] For example, the model update system may have a server and multiple first clients. Taking risk user identification in a resource transfer scenario as an example, assuming that the first client and the target client respectively store characteristic information of several users (such as resource transfer behavior information, etc.), the first client and the target client may send the user identifiers of locally stored users to the server, and the server may determine the target user based on the received user identifiers. For example, assuming that the user identifiers sent by the first client 1 include identifier 1, identifier 2, and identifier 3, the user identifiers sent by the first client 2 include identifier 1, identifier 2, and identifier 4, and the user identifiers sent by the target client include identifier 2 and identifier 4, then the user corresponding to identifier 2 shared by these three clients may be determined as the target user.

[0119] The server may send the user identification of the target user to the first client and the target client respectively, receive the first feature vector sent by the first client and the second feature vector sent by the target client, and then determine a third feature vector corresponding to the target user based on the first feature vector and the second feature vector.

[0120] The server can obtain the risk category of the target user based on the user identifier of the target user, and train the first classification model to be trained based on the third feature vector corresponding to the target user to obtain the trained first classification model, and return the gradient, loss and other model parameters of the trained first classification model to the first client. In this way, the server can calculate the loss based on the user label (i.e., risk category) with a smaller amount of data, reduce the cost of model training, save processing resources, improve processing efficiency, and perform back propagation based on the loss. When calculating back propagation, only the first client updates the model parameters, which ensures the model effectiveness of the target client and improves training efficiency.

[0121] like Figure 6 As shown, the first client can update the first feature extraction model based on the received model parameters, and classify the local user data based on the updated first feature extraction model to determine whether there are risky users. For example, the first client can perform feature extraction on the locally stored third graph structure data based on the updated first feature extraction model, and input the extracted feature vector corresponding to the third graph structure data into the pre-trained second classification model to obtain the user category of the user corresponding to the third graph structure data, and determine whether the user is a risky user based on the user category.

[0122] Alternatively, the first client may also input the third graph structure data into a pre-trained target classification model to obtain a node classification result corresponding to the third graph structure data, and determine whether there is a risky user based on the node classification result, wherein the target classification model may be obtained by training a model constructed by a preset classification sub-model and an updated first feature extraction model based on the fourth graph structure data, and the preset classification sub-model may be the trained first classification model sent by the server.

[0123] In addition, the above is an example of the model updating system identifying risky users in a resource scenario. In addition, the model updating system can also be applied to various application scenarios such as user relationship identification.

[0124] The embodiment of the present specification provides a model updating system. Since the second feature extraction model of the target client is trained based on historical graph structure data constructed by a preset number of historical feature information, the training accuracy of the first classification model can be improved with the participation of the second feature extraction model of the target client. The first client can update the first feature extraction model based on the model parameters of the trained first classification model, which can improve the feature extraction accuracy of the updated first feature extraction model and improve the classification accuracy of the subsequent classification processing of the local data of the first client. At the same time, since the client (i.e., the first client and the target client) only sends the feature vector to the server, the privacy security of the private data of the local user of the client is guaranteed.

[0125] Embodiment 6

[0126] like Fig. 7A and 7B As shown, the embodiment of this specification provides a model updating method, the execution subject of the method can be a federated learning client, and the federated learning client can be a server or a terminal device, wherein the server can be an independent server or a server cluster composed of multiple servers, and the terminal device can be a device such as a personal computer, or a mobile terminal device such as a mobile phone or a tablet computer. The method can specifically include the following steps:

[0127] In S702, an update request for a node classification model is received, and the node classification model is trained based on the first graph structure data to obtain a trained node classification model.

[0128] Among them, the node classification model can be a common model of multiple federated learning clients. In the federated learning scenario, there can be 1 federated learning server and C federated learning clients (C>=2). The node classification model used by the C federated learning clients is the same. The first graph structure data can be constructed based on the user's feature information.

[0129] In implementation, taking the federated learning client as an example of a client used by an institution that can provide resource transfer services to users, since different institutions have different user private data, and each institution's user private data may have small data volume and poor data quality, in order to achieve better use of the node classification model used locally, multiple institutions can update the model parameters of the node classification model used locally through federated learning. For example, the same user may have data transactions with multiple different banks, and each bank stores different user private data. In order to describe the user more accurately and provide better services to the user in the future, multiple banks can update the model parameters of the node classification model through federated learning, so as to process the user's private data through the updated node classification model.

[0130] The federated learning client of each institution participating in federated learning can receive update instructions for the model parameters of the node classification model issued by the federated learning server. That is, when the federated learning client receives an update request for the node classification model, it can train the node classification model based on the first graph structure data to obtain a trained node classification model.

[0131] In addition, the node classification model can be any graph neural network model. For example, the node classification model can be a graph neural network model composed of graph neural networks (GNN), or it can be a graph neural network model composed of GNN and multi-layer perceptron (MLP). In actual application scenarios, there can be multiple methods for constructing a node classification model, which may vary depending on the actual application scenario. The embodiments of this specification do not specifically limit this.

[0132] In S704, the first model parameters of the trained node classification model are sent to the federated learning server, and the target model parameters of the node classification model determined by the federated learning server based on the second model parameters sent by the target client and the first model parameters sent by the federated learning client are received.

[0133] The second model parameters may be model parameters obtained by training the node classification model by the target client based on a predetermined amount of historical graph structure data.

[0134] In implementation, since the node classification model of the target client is trained based on a predetermined amount of historical graph structure data, the node classification effect of the node classification model of the target client is good. That is, with the assistance of the second model parameters, the target model parameters of the node classification model determined by the server can improve the classification effect of the node classification model.

[0135] In S706, the trained node classification model is updated based on the target model parameters to obtain an updated node classification model, and the local user data is classified based on the updated node classification model.

[0136] During implementation, in order to ensure the security of data transmission, the federated learning server can generate a client identifier for each federated learning client participating in the federated learning, and generate verification information corresponding to each federated learning client based on a preset verification information generation algorithm (such as a preset hash algorithm), target model parameters and client identifier.

[0137] In this way, the federated learning client can generate corresponding verification information based on the locally stored client identifier, the preset verification information generation algorithm and the target model parameters, and perform verification processing based on the generated verification information and the received verification information to determine whether the target model parameters have been tampered with during the transmission process.

[0138] After the verification is passed, the trained node classification model is updated based on the target model parameters to obtain an updated node classification model, and the local user data is classified based on the updated node classification model.

[0139] The embodiment of the specification provides a model update method, receiving an update request for a node classification model, and training the node classification model based on the first graph structure data to obtain the trained node classification model, sending the first model parameter of the trained node classification model to the federated learning server, and receiving the target model parameter of the node classification model determined by the federated learning server based on the second model parameter sent by the target client and the first model parameter sent by the federated learning client, the second model parameter is the model parameter obtained by the target client training the node classification model based on a predetermined number of historical graph structure data, updating the trained node classification model based on the target model parameter to obtain the updated node classification model, and classifying the local user data based on the updated node classification model. In this way, the accuracy of the model parameter update of the node classification model can be improved through the participation of multiple federated learning clients, and since the second model parameter is the model parameter obtained by the target client mechanically training the node classification model based on a predetermined number of historical graph structure data, the accuracy of the model parameter update of the node classification model can be improved with the assistance of the second model parameter, and at the same time, since the client (i.e., the federated learning client and the target client) only sends the model parameter to the federated learning server, the privacy security of the private data of the local user of the client is guaranteed.

[0140] Embodiment 7

[0141] like Figure 8 As shown, the embodiment of this specification provides a model updating method, the execution subject of the method can be a federated learning client, and the federated learning client can be a server or a terminal device, wherein the server can be an independent server or a server cluster composed of multiple servers, and the terminal device can be a device such as a personal computer, or a mobile terminal device such as a mobile phone or a tablet computer. The method can specifically include the following steps:

[0142] In S702, an update request for a node classification model is received, and the node classification model is trained based on the first graph structure data to obtain a trained node classification model.

[0143] In S704, the first model parameters of the trained node classification model are sent to the federated learning server, and the target model parameters of the node classification model determined by the federated learning server based on the second model parameters sent by the target client and the first model parameters sent by the federated learning client are received.

[0144] In S706, the trained node classification model is updated based on the target model parameters to obtain an updated node classification model, and local user data is classified based on the updated node classification model.

[0145] In S802, a node classification request for third graph structure data is received.

[0146] In S804, the third graph structure data is input into the updated node classification model to obtain a node classification result corresponding to the third graph structure data.

[0147] In implementation, in actual applications, the above S804 can be processed in various ways. An optional implementation is provided below. For details, please refer to the following steps 1 to 3:

[0148] Step 1: Obtain the fourth graph structure data.

[0149] Step 2: Based on the fourth graph structure data, the updated node classification model is trained to obtain a trained target node classification model.

[0150] Step three: input the third graph structure data into the trained target node classification model to obtain a node classification result corresponding to the third graph structure data.

[0151] During implementation, the updated node classification model is retrained locally on the federated learning client based on the fourth graph structure data to obtain a trained target node classification model that has a high degree of match with the local node classification requirements.

[0152] The embodiment of the specification provides a model update method, receiving an update request for a node classification model, and training the node classification model based on the first graph structure data to obtain the trained node classification model, sending the first model parameter of the trained node classification model to the federated learning server, and receiving the target model parameter of the node classification model determined by the federated learning server based on the second model parameter sent by the target client and the first model parameter sent by the federated learning client, the second model parameter is the model parameter obtained by the target client training the node classification model based on a predetermined number of historical graph structure data, updating the trained node classification model based on the target model parameter to obtain the updated node classification model, and classifying the local user data based on the updated node classification model. In this way, the accuracy of the model parameter update of the node classification model can be improved through the participation of multiple federated learning clients, and since the second model parameter is the model parameter obtained by the target client mechanically training the node classification model based on a predetermined number of historical graph structure data, the accuracy of the model parameter update of the node classification model can be improved with the assistance of the second model parameter, and at the same time, since the client (i.e., the federated learning client and the target client) only sends the model parameter to the federated learning server, the privacy security of the private data of the local user of the client is guaranteed.

[0153] Embodiment 8

[0154] like Fig.9A and Fig. 9B As shown, the embodiment of this specification provides a model updating method, the execution subject of the method can be a federated learning server, and the federated learning server can be a server or a terminal device, wherein the server can be an independent server or a server cluster composed of multiple servers, and the terminal device can be a device such as a personal computer, or a mobile terminal device such as a mobile phone or a tablet computer. The method can specifically include the following steps:

[0155] In S902, a first model parameter sent by a federated learning client and a second model parameter sent by a target client are received.

[0156] Among them, the first model parameters can be model parameters obtained by the federated learning client training the node classification model to be trained based on the first graph structure data, and the second model parameters can be model parameters obtained by the target client training the node classification model based on a predetermined amount of historical graph structure data.

[0157] In S904, target model parameters of the node classification model are determined based on the first model parameters and the second model parameters.

[0158] In implementation, the federated learning server may use the first model parameters sent by multiple federated learning clients and the average of the second model parameters sent by the target client as the target model parameters of the node classification model.

[0159] In addition, there may be multiple target clients, and there may be multiple methods for determining the target model parameters. The embodiments of this specification do not specifically limit the number of target clients and the method for determining the target model parameters.

[0160] In S906, the target model parameters are sent to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameters, and processes the local user data of the federated learning client based on the updated node classification model.

[0161] In addition, the federated learning server can also determine the target model parameters of the node classification model based on the preset weights, first model parameters, and second model parameters corresponding to each federated learning client and the target client.

[0162] Since different federated learning clients and target clients have different amounts of user private data, preset weights may be assigned to different federated learning clients and target clients to determine target model parameters of the node classification model.

[0163] The above-mentioned method for determining the target model parameters is an optional and feasible updating method. In actual application scenarios, there may be a variety of different determination methods, which may vary according to the actual application scenarios. The embodiments of this specification do not make specific limitations on this.

[0164] In addition, the federated learning server can also select multiple federated learning clients from the federated learning clients participating in the federated learning based on multiple preset update cycles, and select a predetermined number of target clients from multiple target clients to update the model parameters. For example, there can be C federated learning clients participating in the federated learning, and G target clients. Assuming that there are E preset update cycles in total, in the first update cycle, F federated learning clients (F<=C) can be randomly selected from the C federated learning clients, and H target clients (H<=G) can be randomly selected from the G target clients. The federated learning server can determine the target parameters of the node classification model based on the first model parameters sent by the F federated learning clients and the second model parameters sent by the H target clients. The federated learning client can update the trained node classification model based on the target parameters to obtain the updated node classification model.

[0165] In the second update cycle (E>=2), A federated learning clients (A<=C) can be randomly selected from C federated learning clients, and I target clients (I<=G) can be randomly selected from G target clients. The federated learning server can determine the target parameters of the node classification model based on the first model parameters sent by the A federated learning clients and the second model parameters sent by the I target clients (the second model parameters can be the model parameters of the node classification model obtained by training the updated node classification model based on the first graph structure data). The federated learning client can update the updated node classification model again based on the target parameters to obtain the updated node classification model. And so on, until the end of the Eth update cycle.

[0166] The embodiment of the present specification provides a model updating method, receiving a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client training a node classification model to be trained based on the first graph structure data, and the second model parameter is a model parameter obtained by the target client training the node classification model based on a predetermined number of historical graph structure data, determining a target model parameter of the node classification model based on the first model parameter and the second model parameter, and sending the target model parameter to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameter, and processes the local user data of the federated learning client based on the updated node classification model. In this way, the accuracy of the model parameter update of the node classification model can be improved through the participation of multiple federated learning clients, and since the second model parameter is a model parameter obtained by the target client mechanically training the node classification model based on a predetermined number of historical graph structure data, the accuracy of the model parameter update of the node classification model can be improved with the assistance of the second model parameter, and at the same time, since the client (i.e., the federated learning client and the target client) only sends the model parameter to the federated learning server, the privacy security of the private data of the local user of the client is guaranteed.

[0167] Embodiment 9

[0168] The embodiment of this specification provides a model updating system, which includes: a federated learning server and multiple federated learning clients, wherein:

[0169] The federated learning client is used to receive an update request for the node classification model, and train the node classification model based on the first graph structure data to obtain a trained node classification model.

[0170] The federated learning server is used to receive the first model parameters sent by the federated learning client and the second model parameters sent by the target client, where the second model parameters are model parameters obtained by the target client by training the node classification model based on a predetermined amount of historical graph structure data.

[0171] The federated learning server is used to determine the target model parameters of the node classification model based on the first model parameters and the second model parameters, and send the target model parameters to the federated learning client.

[0172] The federated learning client is used to update the trained node classification model based on the target model parameters to obtain the updated node classification model, and classify the local user data based on the updated node classification model.

[0173] For example, taking risk user identification in a resource transfer scenario as an example, assuming that the federated learning client and the target client respectively store characteristic information of several users (such as user attribute information and resource transfer behavior information), the federated learning client can construct the first graph structure data based on the locally stored user characteristic information, and the target client can also construct historical graph structure data based on the locally stored user characteristic information.

[0174] The federated learning client may train the node classification model based on the first graph structure data upon receiving an update request for the node classification model, obtain the trained node classification model, and determine the model parameters of the trained node classification model as the first model parameters. The target client may train the node classification model locally based on a predetermined amount of historical graph structure data, and determine the model parameters of the trained node classification model as the second model parameters.

[0175] The federated learning server can determine the target model parameters of the node classification model based on the first model parameters and the second model parameters, and return the target model parameters to the federated learning client.

[0176] The federated learning client can update the trained node classification model based on the target model parameters to obtain an updated node classification model, and when receiving a node classification request for the third graph structure data, input the third graph structure data into the updated node classification model to obtain a node classification result corresponding to the third graph structure data, and determine whether there is a risky user based on the node classification result. Alternatively, the federated learning client can retrain the updated node classification model based on the fourth graph structure data to obtain a trained target node classification model, and input the third graph structure data into the trained target node classification model to obtain a node classification result corresponding to the third graph structure data.

[0177] The embodiments of this specification provide a model updating system, which can improve the accuracy of updating the model parameters of the node classification model through the participation of multiple federated learning clients, and because the second model parameters are model parameters obtained by the target client through mechanical energy training of the node classification model based on a predetermined amount of historical graph structure data, the accuracy of updating the model parameters of the node classification model can be improved with the assistance of the second model parameters. At the same time, because the client (i.e., the federated learning client and the target client) only sends the model parameters to the federated learning server, the privacy security of the private data of the local user of the client is guaranteed.

[0178] Embodiment 10

[0179] like Fig. 10A and Fig. 10B As shown, the embodiment of this specification provides a model updating method, and the execution subject of the method can be a blockchain system. The method can specifically include the following steps:

[0180] In S1002, based on a smart contract pre-deployed in the blockchain system, a first model parameter sent by a federated learning client and a second model parameter sent by a target client are received.

[0181] Among them, the first model parameters can be model parameters obtained by the federated learning client through training the node classification model to be trained based on the first graph structure data, and the second model parameters can be model parameters obtained by the target client through training the node classification model based on a predetermined amount of historical graph structure data. The node classification model can be a common model of multiple federated learning clients and the target client. The smart contract can be generated based on pre-set rules, wherein the pre-set rules may include, for example, determining whether the reception time of the model parameter update instruction is within the valid update time, whether the federated learning server has the model parameter update authority, etc.

[0182] In implementation, a corresponding trusted application can be developed based on the functions of the blockchain. The trusted application can be encapsulated into an independent application and installed in a local terminal device or server by each participant in federated learning (such as a federated learning client) and the target client. Alternatively, the participants in federated learning and the target client can also jointly agree on a certain application as a trusted application and send relevant information (such as model parameters, etc.) to the blockchain system through the trusted application.

[0183] The blockchain system may obtain the first model parameter and the second model parameter of the node classification model based on the pre-deployed smart contract. For example, the blockchain system may determine whether the current time reaches the preset model parameter update time for the node classification model based on the pre-deployed smart contract, and when the preset model parameter update time is reached, the first model parameter and the second model parameter corresponding to the node classification model may be obtained.

[0184] In S1004, target model parameters of the node classification model are determined based on the smart contract, the first model parameters, and the second model parameters.

[0185] In implementation, the blockchain system may use the first model parameters sent by multiple federated learning clients and the average of the second model parameters sent by the target client as the target model parameters of the node classification model.

[0186] In S1006, the target model parameters are sent to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameters, and processes the local user data of the federated learning client based on the updated node classification model.

[0187] An embodiment of the present specification provides a model updating method, which is based on a smart contract pre-deployed in a blockchain system, receives a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client for training a node classification model to be trained based on first graph structure data, and the second model parameter is a model parameter obtained by the target client for training a node classification model based on a predetermined amount of historical graph structure data, determines a target model parameter of the node classification model based on the smart contract, the first model parameter, and the second model parameter, and sends the target model parameter to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameter, and processes local user data of the federated learning client based on the updated node classification model. In this way, the accuracy of updating the model parameters of the node classification model can be improved through the participation of multiple federated learning clients, and since the second model parameter is the model parameter obtained by the target client through mechanical energy training of the node classification model based on a predetermined amount of historical graph structure data, the accuracy of updating the model parameters of the node classification model can be improved with the assistance of the second model parameter. At the same time, since the client (i.e., the federated learning client and the target client) only sends the model parameters to the federated learning server, the privacy security of the private data of the local user of the client is guaranteed, and by storing the first model parameter, the second model parameter and the target model parameter through the blockchain system, the security of the data can be guaranteed and the traceability of the data can be improved.

[0188] Embodiment 11

[0189] The above is a model updating method provided in the embodiment of this specification. Based on the same idea, the embodiment of this specification also provides a model updating device, such as Fig.11 shown.

[0190] The model updating device comprises: a first receiving module 1101, a vector sending module 1102 and a model updating module 1103, wherein:

[0191] A first receiving module 1101 is used to receive an update request for a first feature extraction model, and determine a first feature vector corresponding to the first graph structure data based on the first feature extraction model to be trained and the first graph structure data, where the first graph structure data is determined by first feature information of a target user;

[0192] The vector sending module 1102 is used to send the first feature vector to the server and receive the model parameters sent by the server, the model parameters are the parameters of the trained first classification model obtained by the server based on the first feature vector and the second feature vector to train the first classification model to be trained, the second feature vector is a vector determined by the target client based on the pre-trained second feature extraction model and the second graph structure data, the second graph structure data is determined by the second feature information of the target user, the second feature extraction model is trained based on the historical graph structure data, the historical graph structure data is constructed by a preset number of historical feature information, the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model;

[0193] The model updating module 1103 is used to update the first feature extraction model based on the model parameters, and classify the local user data based on the updated first feature extraction model.

[0194] In the embodiment of this specification, the device further includes:

[0195] A second receiving module, used to receive a node classification request for the third graph structure data;

[0196] A result acquisition module is used to input the third graph structure data into a pre-trained target classification model to obtain a node classification result corresponding to the third graph structure data. The target classification model is obtained by training a model constructed by a preset classification sub-model and the updated first feature extraction model based on the fourth graph structure data.

[0197] In the embodiment of this specification, the device further includes:

[0198] A data acquisition module, used to acquire the fourth graph structure data;

[0199] A model training module is used to train the target classification model based on the fourth graph structure data to obtain the pre-trained target classification model.

[0200] In the embodiment of this specification, the preset classification sub-model is the trained first classification model sent by the server.

[0201] Based on the same idea, the embodiment of this specification also provides a model updating device, such as Fig.12 shown.

[0202] The model updating device comprises: a vector receiving module 1201, a model training module 1202 and a data sending module 1203, wherein:

[0203] A vector receiving module 1201 is used to receive a first feature vector sent by a first client and a second feature vector sent by a target client, wherein the first feature vector is a vector determined by the first client based on a first feature extraction model to be trained and first graph structure data, and the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, wherein the second graph structure data is determined by second feature information of the target user, and the second feature extraction model is obtained by training based on historical graph structure data, and the historical graph structure data is constructed by a preset number of historical feature information;

[0204] A model training module 1202 is used to train a first classification model to be trained based on the first feature vector and the second feature vector to obtain a trained first classification model, wherein the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second feature extraction model;

[0205] The data sending module 1203 is used to send the model parameters of the trained first classification model to the first client, so that the first client updates the first feature extraction model based on the model parameters, and processes the local user data of the first client based on the updated first feature extraction model.

[0206] In the embodiment of this specification, the device further includes:

[0207] A user determination module, configured to determine the target user based on user identifiers sent by the first clients and the target clients;

[0208] A first sending module, configured to send the user identification of the target user to the first client, so that the first client determines the first characteristic information of the target user based on the user identification of the target user, and constructs the first graph structure data based on the first characteristic information of the target user;

[0209] The second sending module is used to send the user identification of the target user to the target client, so that the target client determines the second feature information of the target user based on the user identification of the target user, and constructs the second graph structure data based on the second feature information of the target user.

[0210] In the embodiment of this specification, the model training module 1202 is used to:

[0211] Merging the first eigenvector and the second eigenvector to obtain the third eigenvector;

[0212] Based on the third feature vector, the first classification model to be trained is trained to obtain the trained first classification model.

[0213] In an embodiment of the present specification, the first feature vector is a vector obtained by the first client encrypting a target vector based on a preset privacy encryption algorithm, and the target vector is a vector obtained by the first client performing feature extraction on the first graph structure data through the first feature extraction model to be trained.

[0214] In the embodiments of this specification, the preset privacy encryption algorithm is a homomorphic encryption algorithm or a differential privacy encryption algorithm.

[0215] The embodiment of the present specification provides a model updating device. Since the second feature extraction model of the target client is trained based on historical graph structure data constructed by a preset number of historical feature information, the training accuracy of the first classification model can be improved with the participation of the second feature extraction model of the target client. The first client can update the first feature extraction model based on the model parameters of the trained first classification model, which can improve the feature extraction accuracy of the updated first feature extraction model and improve the classification accuracy of the subsequent classification processing of the local data of the first client. At the same time, since the client (i.e., the first client and the target client) only sends the feature vector to the server, the privacy security of the private data of the local user of the client is guaranteed.

[0216] Embodiment 12

[0217] Based on the same idea, the embodiment of this specification also provides a model updating device, such as Fig.13 shown.

[0218] The model updating device comprises: a first receiving module 1301, a data sending module 1302 and a model updating module 1303, wherein:

[0219] The first receiving module 1301 is used to receive an update request for a node classification model, and train the node classification model based on the first graph structure data to obtain a trained node classification model;

[0220] A data sending module 1302 is used to send the first model parameters of the trained node classification model to a federated learning server, and receive the target model parameters of the node classification model determined by the federated learning server based on the second model parameters sent by the target client and the first model parameters sent by the federated learning client, wherein the second model parameters are model parameters obtained by the target client training the node classification model based on a predetermined amount of historical graph structure data;

[0221] The model updating module 1303 is used to update the trained node classification model based on the target model parameters to obtain an updated node classification model, and classify the local user data based on the updated node classification model.

[0222] In the embodiment of this specification, the device further includes:

[0223] A second receiving module, used to receive a node classification request for the third graph structure data;

[0224] The result determination module is used to input the third graph structure data into the updated node classification model to obtain a node classification result corresponding to the third graph structure data.

[0225] In the embodiment of this specification, the result determination module is used to:

[0226] Obtaining fourth graph structure data;

[0227] Based on the fourth graph structure data, training the updated node classification model to obtain a trained target node classification model;

[0228] The third graph structure data is input into the trained target node classification model to obtain a node classification result corresponding to the third graph structure data.

[0229] Based on the same idea, the embodiment of this specification also provides a model updating device, such as Fig.14 shown.

[0230] The model updating device comprises: a data receiving module 1401, a parameter determining module 1402 and a parameter sending module 14103, wherein:

[0231] The data receiving module 1401 is used to receive a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client training a node classification model to be trained based on the first graph structure data, and the second model parameter is a model parameter obtained by the target client training the node classification model based on a predetermined amount of historical graph structure data;

[0232] A parameter determination module 1402, configured to determine target model parameters of the node classification model based on the first model parameters and the second model parameters;

[0233] The parameter sending module 1403 is used to send the target model parameters to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameters, and processes the local user data of the federated learning client based on the updated node classification model.

[0234] The embodiments of this specification provide a model updating device, which can improve the accuracy of updating the model parameters of the node classification model through the participation of multiple federated learning clients, and because the second model parameters are model parameters obtained by the target client through mechanical energy training of the node classification model based on a predetermined amount of historical graph structure data, therefore, with the assistance of the second model parameters, the accuracy of updating the model parameters of the node classification model can be improved. At the same time, because the client (i.e., the federated learning client and the target client) only sends the model parameters to the federated learning server, the privacy security of the private data of the local user of the client is guaranteed.

[0235] Embodiment 13

[0236] Based on the same idea, the embodiment of this specification also provides a model updating device, such as Fig.15 shown.

[0237] The model updating device is a device in the blockchain system, comprising: a data receiving module 1501, a parameter determination module 1502 and a parameter sending module 1503, wherein:

[0238] A data receiving module 1501 is used to receive, based on a smart contract pre-deployed in the blockchain system, a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client for training a node classification model to be trained based on first graph structure data, and the second model parameter is a model parameter obtained by the target client for training the node classification model based on a predetermined amount of historical graph structure data;

[0239] A parameter determination module 1502, configured to determine a target model parameter of the node classification model based on the smart contract, the first model parameter, and the second model parameter;

[0240] The parameter sending module 1503 is used to send the target model parameters to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameters, and processes the local user data of the federated learning client based on the updated node classification model.

[0241] An embodiment of the present specification provides a model updating device, which is based on a smart contract pre-deployed in a blockchain system, receives a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client for training a node classification model to be trained based on first graph structure data, and the second model parameter is a model parameter obtained by the target client for training the node classification model based on a predetermined amount of historical graph structure data, determines a target model parameter of the node classification model based on the smart contract, the first model parameter, and the second model parameter, and sends the target model parameter to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameter, and processes local user data of the federated learning client based on the updated node classification model. In this way, the accuracy of updating the model parameters of the node classification model can be improved through the participation of multiple federated learning clients, and since the second model parameter is the model parameter obtained by the target client through mechanical energy training of the node classification model based on a predetermined amount of historical graph structure data, the accuracy of updating the model parameters of the node classification model can be improved with the assistance of the second model parameter. At the same time, since the client (i.e., the federated learning client and the target client) only sends the model parameters to the federated learning server, the privacy security of the private data of the local user of the client is guaranteed, and by storing the first model parameter, the second model parameter and the target model parameter through the blockchain system, the security of the data can be guaranteed and the traceability of the data can be improved.

[0242] Embodiment 14

[0243] Based on the same idea, the embodiment of this specification also provides a model updating device, such as Fig.16 shown.

[0244] The model update device may have relatively large differences due to different configurations or performances, and may include one or more processors 1601 and memory 1602, and the memory 1602 may store one or more storage applications or data. Among them, the memory 1602 may be a short-term storage or a persistent storage. The application stored in the memory 1602 may include one or more modules (not shown in the figure), and each module may include a series of computer executable instructions in the model update device. Furthermore, the processor 1601 may be configured to communicate with the memory 1602, and execute a series of computer executable instructions in the memory 1602 on the model update device. The model update device may also include one or more power supplies 1603, one or more wired or wireless network interfaces 1604, one or more input and output interfaces 1605, and one or more keyboards 1606.

[0245] Specifically, in this embodiment, the model updating device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the model updating device, and the one or more programs are configured to be executed by one or more processors, including computer executable instructions for performing the following:

[0246] receiving an update request for a first feature extraction model, and determining a first feature vector corresponding to the first graph structure data based on the first feature extraction model to be trained and first graph structure data, wherein the first graph structure data is determined by first feature information of a target user;

[0247] The first feature vector is sent to a server, and model parameters sent by the server are received, wherein the model parameters are parameters of the trained first classification model obtained by the server based on the first feature vector and the second feature vector to train the first classification model to be trained, wherein the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, wherein the second graph structure data is determined by second feature information of the target user, and the second feature extraction model is obtained by training based on historical graph structure data, wherein the historical graph structure data is constructed by a preset number of historical feature information, and the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second extraction model;

[0248] The first feature extraction model is updated based on the model parameters, and the local user data is classified based on the updated first feature extraction model.

[0249] Optionally, the method further comprises:

[0250] receiving a node classification request for third graph structure data;

[0251] The third graph structure data is input into a pre-trained target classification model to obtain a node classification result corresponding to the third graph structure data. The target classification model is obtained by training a model constructed by a preset classification sub-model and the updated first feature extraction model based on the fourth graph structure data.

[0252] Optionally, before inputting the third graph structure data into a pre-trained target node classification model to obtain a node classification result corresponding to the third graph structure data, the method further includes:

[0253] Acquire the fourth graph structure data;

[0254] Based on the fourth graph structure data, the target classification model is trained to obtain the pre-trained target classification model.

[0255] Optionally, the preset classification sub-model is the trained first classification model sent by the server.

[0256] In addition, the model updating device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the model updating device, and the one or more programs configured to be executed by one or more processors include computer executable instructions for performing the following:

[0257] Receive a first feature vector sent by a first client and a second feature vector sent by a target client, wherein the first feature vector is a vector determined by the first client based on a first feature extraction model to be trained and first graph structure data, and the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, wherein the second graph structure data is determined by second feature information of the target user, and the second feature extraction model is trained based on historical graph structure data, and the historical graph structure data is constructed by a preset number of historical feature information;

[0258] Based on the first feature vector and the second feature vector, a first classification model to be trained is trained to obtain a trained first classification model, wherein the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second feature extraction model;

[0259] The model parameters of the trained first classification model are sent to the first client, so that the first client updates the first feature extraction model based on the model parameters, and processes the local user data of the first client based on the updated first feature extraction model.

[0260] Optionally, before receiving the first feature vector sent by the first client and the second feature vector sent by the target client, the method further includes:

[0261] Determining the target user based on the user identifiers sent by the first clients and the target clients;

[0262] Sending the user identification of the target user to the first client, so that the first client determines the first characteristic information of the target user based on the user identification of the target user, and constructs the first graph structure data based on the first characteristic information of the target user;

[0263] The user identification of the target user is sent to the target client, so that the target client determines the second feature information of the target user based on the user identification of the target user, and constructs the second graph structure data based on the second feature information of the target user.

[0264] Optionally, the training of the first classification model to be trained based on the first feature vector and the second feature vector to obtain the trained first classification model includes:

[0265] Merging the first eigenvector and the second eigenvector to obtain the third eigenvector;

[0266] Based on the third feature vector, the first classification model to be trained is trained to obtain the trained first classification model.

[0267] Optionally, the first feature vector is a vector obtained by the first client encrypting a target vector based on a preset privacy encryption algorithm, and the target vector is a vector obtained by the first client performing feature extraction on the first graph structure data through the first feature extraction model to be trained.

[0268] Optionally, the preset privacy encryption algorithm is a homomorphic encryption algorithm or a differential privacy encryption algorithm.

[0269] In addition, the model updating device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the model updating device, and the one or more programs configured to be executed by one or more processors include computer executable instructions for performing the following:

[0270] receiving an update request for a node classification model, and training the node classification model based on the first graph structure data to obtain a trained node classification model;

[0271] Sending the first model parameters of the trained node classification model to a federated learning server, and receiving the target model parameters of the node classification model determined by the federated learning server based on the second model parameters sent by the target client and the first model parameters sent by the federated learning client, wherein the second model parameters are model parameters obtained by the target client training the node classification model based on a predetermined amount of historical graph structure data;

[0272] The trained node classification model is updated based on the target model parameters to obtain an updated node classification model, and local user data is classified based on the updated node classification model.

[0273] Optionally, the method further comprises:

[0274] receiving a node classification request for third graph structure data;

[0275] The third graph structure data is input into the updated node classification model to obtain a node classification result corresponding to the third graph structure data.

[0276] Optionally, inputting the third graph structure data into the updated node classification model to obtain a node classification result corresponding to the third graph structure data includes:

[0277] Obtaining fourth graph structure data;

[0278] Based on the fourth graph structure data, training the updated node classification model to obtain a trained target node classification model;

[0279] The third graph structure data is input into the trained target node classification model to obtain a node classification result corresponding to the third graph structure data.

[0280] In addition, the model updating device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the model updating device, and the one or more programs configured to be executed by one or more processors include computer executable instructions for performing the following:

[0281] Receive a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client training a node classification model to be trained based on first graph structure data, and the second model parameter is a model parameter obtained by the target client training the node classification model based on a predetermined amount of historical graph structure data;

[0282] Determining target model parameters of the node classification model based on the first model parameters and the second model parameters;

[0283] The target model parameters are sent to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameters, and processes the local user data of the federated learning client based on the updated node classification model.

[0284] In addition, the model updating device may be a device in the blockchain system, including a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the model updating device, and the one or more programs are configured to be executed by one or more processors, including computer executable instructions for performing the following:

[0285] Based on a smart contract pre-deployed in the blockchain system, receiving a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client training a node classification model to be trained based on first graph structure data, and the second model parameter is a model parameter obtained by the target client training the node classification model based on a predetermined amount of historical graph structure data;

[0286] Determining target model parameters of the node classification model based on the smart contract, the first model parameters, and the second model parameters;

[0287] The target model parameters are sent to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameters, and processes the local user data of the federated learning client based on the updated node classification model.

[0288] The embodiment of the present specification provides a model updating device. Since the second feature extraction model of the target client is trained based on historical graph structure data constructed by a preset number of historical feature information, the training accuracy of the first classification model can be improved with the participation of the second feature extraction model of the target client. Then, based on the model parameters of the trained first classification model, the first feature extraction model is updated, which can improve the feature extraction accuracy of the updated first feature extraction model and improve the classification accuracy of subsequent classification of local data. At the same time, since the client only sends the first feature vector to the server, the privacy security of local user private data is guaranteed.

[0289] Embodiment 15

[0290] The embodiment of this specification also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, each process of the above-mentioned model updating method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0291] The embodiment of the present specification provides a computer-readable storage medium. Since the second feature extraction model of the target client is trained based on historical graph structure data constructed by a preset number of historical feature information, the training accuracy of the first classification model can be improved with the participation of the second feature extraction model of the target client. Then, based on the model parameters of the trained first classification model, the first feature extraction model is updated, which can improve the feature extraction accuracy of the updated first feature extraction model and improve the classification accuracy of subsequent classification of local data. At the same time, since the client only sends the first feature vector to the server, the privacy security of local user private data is guaranteed.

[0292] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0293] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0294] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0295] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0296] For the convenience of description, the above devices are described in terms of functions and are divided into various units. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0297] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, one or more embodiments of this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0298] The embodiments of this specification are described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in one process or multiple processes in the flowchart and / or one box or multiple boxes in the block diagram.

[0299] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0300] These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0301] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0302] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0303] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0304] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0305] It should be understood by those skilled in the art that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, one or more embodiments of this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0306] One or more embodiments of the present specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0307] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0308] The above description is only an embodiment of the present specification and is not intended to limit the present specification. For those skilled in the art, the present specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification shall be included in the scope of the claims of the present specification.

Claims

1. A model updating method, comprising: receiving an update request for a first feature extraction model, and determining a first feature vector corresponding to the first graph structure data based on the first feature extraction model to be trained and first graph structure data, wherein the first graph structure data is determined by first feature information of a target user; The first feature vector is sent to a server, and model parameters sent by the server are received, wherein the model parameters are parameters of the trained first classification model obtained by the server based on the first feature vector and the second feature vector to train the first classification model to be trained, wherein the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, wherein the second graph structure data is determined by second feature information of the target user, and the second feature extraction model is obtained by training based on historical graph structure data, wherein the historical graph structure data is constructed by a preset number of historical feature information, and the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second feature extraction model; The first feature extraction model is updated based on the model parameters, and the local user data is classified based on the updated first feature extraction model.

2. The method according to claim 1, further comprising: receiving a node classification request for third graph structure data; The third graph structure data is input into a pre-trained target classification model to obtain a node classification result corresponding to the third graph structure data. The target classification model is obtained by training a model constructed by a preset classification sub-model and the updated first feature extraction model based on the fourth graph structure data.

3. According to the method of claim 2, the preset classification sub-model is the trained first classification model sent by the server.

4. A model updating method, comprising: Receive a first feature vector sent by a first client and a second feature vector sent by a target client, wherein the first feature vector is a vector determined by the first client based on a first feature extraction model to be trained and first graph structure data, and the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, wherein the second graph structure data is determined by second feature information of a target user, and the second feature extraction model is trained based on historical graph structure data, and the historical graph structure data is constructed by a preset number of historical feature information; Based on the first feature vector and the second feature vector, a first classification model to be trained is trained to obtain a trained first classification model, wherein the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second feature extraction model; The model parameters of the trained first classification model are sent to the first client, so that the first client updates the first feature extraction model based on the model parameters, and processes the local user data of the first client based on the updated first feature extraction model.

5. The method according to claim 4, before receiving the first feature vector sent by the first client and the second feature vector sent by the target client, the method further comprises: Determining the target user based on the user identifiers sent by the first clients and the target clients; Sending the user identification of the target user to the first client, so that the first client determines the first characteristic information of the target user based on the user identification of the target user, and constructs the first graph structure data based on the first characteristic information of the target user; The user identification of the target user is sent to the target client, so that the target client determines the second feature information of the target user based on the user identification of the target user, and constructs the second graph structure data based on the second feature information of the target user.

6. The method according to claim 5, wherein the first classification model to be trained is trained based on the first feature vector and the second feature vector to obtain the trained first classification model, comprising: Merging the first eigenvector and the second eigenvector to obtain a third eigenvector; Based on the third feature vector, the first classification model to be trained is trained to obtain the trained first classification model.

7. A model updating method, applied to a federated learning client, comprising: Receiving an update request for a node classification model, and training the node classification model based on the first graph structure data to obtain a trained node classification model, wherein the node classification model is a common model of a plurality of the federated learning clients; Sending the first model parameters of the trained node classification model to a federated learning server, and receiving the target model parameters of the node classification model determined by the federated learning server based on the second model parameters sent by the target client and the first model parameters sent by the federated learning client, wherein the second model parameters are model parameters obtained by the target client training the node classification model based on a predetermined amount of historical graph structure data; The trained node classification model is updated based on the target model parameters to obtain an updated node classification model, and local user data is classified based on the updated node classification model.

8. The method according to claim 7, further comprising: receiving a node classification request for third graph structure data; The third graph structure data is input into the updated node classification model to obtain a node classification result corresponding to the third graph structure data.

9. A model updating method, applied to a federated learning server, comprising: Receive a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client by training a node classification model to be trained based on first graph structure data, and the second model parameter is a model parameter obtained by the target client by training the node classification model based on a predetermined amount of historical graph structure data, and the node classification model is a common model of multiple federated learning clients; Determining target model parameters of the node classification model based on the first model parameters and the second model parameters; The target model parameters are sent to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameters, and processes the local user data of the federated learning client based on the updated node classification model.

10. A model updating method, applied to a blockchain system, comprising: Based on a smart contract pre-deployed in the blockchain system, receiving a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client training a node classification model to be trained based on first graph structure data, and the second model parameter is a model parameter obtained by the target client training the node classification model based on a predetermined amount of historical graph structure data; Determining target model parameters of the node classification model based on the smart contract, the first model parameters, and the second model parameters; The target model parameters are sent to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameters, and processes the local user data of the federated learning client based on the updated node classification model.

11. A model updating system, comprising a federated learning server and a plurality of federated learning clients, wherein: The federated learning client is used to receive an update request for a node classification model, and train the node classification model based on the first graph structure data to obtain a trained node classification model; The federated learning server is used to receive a first model parameter sent by the federated learning client and a second model parameter sent by the target client, wherein the second model parameter is a model parameter obtained by the target client by training the node classification model based on a predetermined amount of historical graph structure data; The federated learning server is used to determine the target model parameters of the node classification model based on the first model parameters and the second model parameters, and send the target model parameters to the federated learning client; The federated learning client is used to update the trained node classification model based on the target model parameters to obtain an updated node classification model, and classify local user data based on the updated node classification model.

12. A model updating device, comprising: a first receiving module, configured to receive an update request for a first feature extraction model, and determine a first feature vector corresponding to the first graph structure data based on the first feature extraction model to be trained and the first graph structure data, wherein the first graph structure data is determined by first feature information of a target user; A vector sending module, used to send the first feature vector to a server, and receive model parameters sent by the server, the model parameters are parameters of the trained first classification model obtained by the server based on the first feature vector and the second feature vector to train the first classification model to be trained, the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, the second graph structure data is determined by the second feature information of the target user, the second feature extraction model is trained based on historical graph structure data, the historical graph structure data is constructed by a preset number of historical feature information, the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second feature extraction model; The model updating module is used to update the first feature extraction model based on the model parameters, and classify the local user data based on the updated first feature extraction model.

13. A model updating device, comprising: A vector receiving module, used to receive a first feature vector sent by a first client and a second feature vector sent by a target client, wherein the first feature vector is a vector determined by the first client based on a first feature extraction model to be trained and first graph structure data, and the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, wherein the second graph structure data is determined by second feature information of a target user, and the second feature extraction model is obtained by training based on historical graph structure data, and the historical graph structure data is constructed by a preset number of historical feature information; A model training module, used for training a first classification model to be trained based on the first feature vector and the second feature vector to obtain a trained first classification model, wherein the first classification model is used for classifying the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second feature extraction model; A data sending module is used to send the model parameters of the trained first classification model to the first client, so that the first client updates the first feature extraction model based on the model parameters, and processes the local user data of the first client based on the updated first feature extraction model.

14. A model updating device, comprising: A first receiving module, configured to receive an update request for a node classification model, and train the node classification model based on the first graph structure data to obtain a trained node classification model, wherein the node classification model is a common model of multiple federated learning clients; A data sending module, used to send the first model parameters of the trained node classification model to a federated learning server, and receive the target model parameters of the node classification model determined by the federated learning server based on the second model parameters sent by the target client and the first model parameters sent by the federated learning client, wherein the second model parameters are model parameters obtained by the target client training the node classification model based on a predetermined amount of historical graph structure data; The model updating module is used to update the trained node classification model based on the target model parameters to obtain an updated node classification model, and classify the local user data based on the updated node classification model.

15. A model updating device, the device being a device in a blockchain system, comprising: A data receiving module, used to receive a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client through training a node classification model to be trained based on first graph structure data, and the second model parameter is a model parameter obtained by the target client through training the node classification model based on a predetermined amount of historical graph structure data, and the node classification model is a common model of multiple federated learning clients; A parameter determination module, used for determining target model parameters of the node classification model based on the first model parameters and the second model parameters; A parameter sending module is used to send the target model parameters to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameters, and processes the local user data of the federated learning client based on the updated node classification model.

16. A model updating device, comprising: processor; as well as a memory arranged to store computer executable instructions which, when executed, cause the processor to: receiving an update request for a first feature extraction model, and determining a first feature vector corresponding to the first graph structure data based on the first feature extraction model to be trained and first graph structure data, wherein the first graph structure data is determined by first feature information of a target user; The first feature vector is sent to a server, and model parameters sent by the server are received, wherein the model parameters are parameters of the trained first classification model obtained by the server based on the first feature vector and the second feature vector to train the first classification model to be trained, wherein the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, wherein the second graph structure data is determined by second feature information of the target user, and the second feature extraction model is obtained by training based on historical graph structure data, wherein the historical graph structure data is constructed by a preset number of historical feature information, and the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second feature extraction model; The first feature extraction model is updated based on the model parameters, and the local user data is classified based on the updated first feature extraction model.

17. A model updating device, comprising: processor; as well as a memory arranged to store computer executable instructions which, when executed, cause the processor to: Receive a first feature vector sent by a first client and a second feature vector sent by a target client, wherein the first feature vector is a vector determined by the first client based on a first feature extraction model to be trained and first graph structure data, and the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, wherein the second graph structure data is determined by second feature information of a target user, and the second feature extraction model is trained based on historical graph structure data, and the historical graph structure data is constructed by a preset number of historical feature information; Based on the first feature vector and the second feature vector, a first classification model to be trained is trained to obtain a trained first classification model, wherein the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second feature extraction model; The model parameters of the trained first classification model are sent to the first client, so that the first client updates the first feature extraction model based on the model parameters, and processes the local user data of the first client based on the updated first feature extraction model.

18. A model updating device, comprising: processor; as well as a memory arranged to store computer executable instructions which, when executed, cause the processor to: receiving an update request for a node classification model, and training the node classification model based on the first graph structure data to obtain a trained node classification model, wherein the node classification model is a common model of multiple federated learning clients; Sending the first model parameters of the trained node classification model to a federated learning server, and receiving the target model parameters of the node classification model determined by the federated learning server based on the second model parameters sent by the target client and the first model parameters sent by the federated learning client, wherein the second model parameters are model parameters obtained by the target client training the node classification model based on a predetermined amount of historical graph structure data; The trained node classification model is updated based on the target model parameters to obtain an updated node classification model, and local user data is classified based on the updated node classification model.

19. A model updating device, comprising: processor; as well as a memory arranged to store computer executable instructions which, when executed, cause the processor to: Receive a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client by training a node classification model to be trained based on first graph structure data, and the second model parameter is a model parameter obtained by the target client by training the node classification model based on a predetermined amount of historical graph structure data, and the node classification model is a common model of multiple federated learning clients; Determining target model parameters of the node classification model based on the first model parameters and the second model parameters; The target model parameters are sent to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameters, and processes the local user data of the federated learning client based on the updated node classification model.

20. A storage medium for storing computer executable instructions, wherein the executable instructions implement the following process when executed: receiving an update request for a first feature extraction model, and determining a first feature vector corresponding to the first graph structure data based on the first feature extraction model to be trained and first graph structure data, wherein the first graph structure data is determined by first feature information of a target user; The first feature vector is sent to a server, and model parameters sent by the server are received, wherein the model parameters are parameters of the trained first classification model obtained by the server based on the first feature vector and the second feature vector to train the first classification model to be trained, wherein the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, wherein the second graph structure data is determined by second feature information of the target user, and the second feature extraction model is obtained by training based on historical graph structure data, wherein the historical graph structure data is constructed by a preset number of historical feature information, and the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second feature extraction model; The first feature extraction model is updated based on the model parameters, and the local user data is classified based on the updated first feature extraction model.

21. A storage medium for storing computer executable instructions, wherein the executable instructions implement the following process when executed: Receive a first feature vector sent by a first client and a second feature vector sent by a target client, wherein the first feature vector is a vector determined by the first client based on a first feature extraction model to be trained and first graph structure data, and the second feature vector is a vector determined by the target client based on a pre-trained second feature extraction model and second graph structure data, wherein the second graph structure data is determined by second feature information of a target user, and the second feature extraction model is trained based on historical graph structure data, and the historical graph structure data is constructed by a preset number of historical feature information; Based on the first feature vector and the second feature vector, a first classification model to be trained is trained to obtain a trained first classification model, wherein the first classification model is used to classify the feature information of the target user according to the first feature vector extracted by the first feature extraction model and the second feature vector extracted by the second feature extraction model; The model parameters of the trained first classification model are sent to the first client, so that the first client updates the first feature extraction model based on the model parameters, and processes the local user data of the first client based on the updated first feature extraction model.

22. A storage medium for storing computer executable instructions, wherein the executable instructions implement the following process when executed: receiving an update request for a node classification model, and training the node classification model based on the first graph structure data to obtain a trained node classification model, wherein the node classification model is a common model of multiple federated learning clients; Sending the first model parameters of the trained node classification model to a federated learning server, and receiving the target model parameters of the node classification model determined by the federated learning server based on the second model parameters sent by the target client and the first model parameters sent by the federated learning client, wherein the second model parameters are model parameters obtained by the target client training the node classification model based on a predetermined amount of historical graph structure data; The trained node classification model is updated based on the target model parameters to obtain an updated node classification model, and local user data is classified based on the updated node classification model.

23. A storage medium for storing computer executable instructions, wherein the executable instructions implement the following process when executed: Receive a first model parameter sent by a federated learning client and a second model parameter sent by a target client, wherein the first model parameter is a model parameter obtained by the federated learning client by training a node classification model to be trained based on first graph structure data, and the second model parameter is a model parameter obtained by the target client by training the node classification model based on a predetermined amount of historical graph structure data, and the node classification model is a common model of multiple federated learning clients; Determining target model parameters of the node classification model based on the first model parameters and the second model parameters; The target model parameters are sent to the federated learning client, so that the federated learning client updates the trained node classification model based on the target model parameters, and processes the local user data of the federated learning client based on the updated node classification model.

Images