Graph neural network processing method and device, equipment and storage medium

Abstract

The invention relates to the technical field of graph neural network processing, and provides a graph neural network processing method and device, equipment and a storage medium, and the method comprises the steps: obtaining a graph neural network, the graph neural network comprises at least one sub-graph neural network, and the sub-graph neural network comprises user data features of at least one user; determining a target sub-graph neural network according to a data deletion request of a target user; the target sub-graph neural network comprises user data features of a target user; and according to other user data features except the user data feature indicated by the data deletion request in the target sub-graph neural network, adjusting network parameters of the target sub-graph neural network so as to improve network parameter adjustment convenience and network parameter adjustment efficiency of the target sub-graph neural network in the graph neural network. Therefore, the data memory deletion convenience and the data memory deletion efficiency of the graph neural network are improved.

Description

Technical Field

[0001] The present application relates to the field of graph neural network processing technology, and in particular to a graph neural network processing method, device, equipment and storage medium. Background Art

[0002] At present, with the development of artificial intelligence (AI) models, AI models can provide users with traversal and high-quality services. For example, in financial scenarios, AI models can be used to answer financial questions raised by users. In social scenarios, AI models can be used to recommend social content to users, such as videos, articles, playlists, etc.

[0003] During the training process of the AI ​​model, the AI ​​model will obtain the user's user data, such as the user's financial operation data, the user's social operation data, etc., to improve the intelligence of the AI ​​model. Accordingly, the AI ​​model can retain data memory corresponding to the user data, such as data features corresponding to the user data. In the related art, when the user deletes the corresponding user data, it is generally necessary to retrain the AI ​​model based on the currently retained user data, which requires a lot of computing power, resulting in poor convenience and efficiency of data memory deletion for the AI ​​model. Summary of the invention

[0004] The main purpose of this application is to provide a method, device, equipment and storage medium for processing a graph neural network, aiming to improve the convenience and efficiency of adjusting the network parameters of a target subgraph neural network in a graph neural network, so as to improve the convenience and efficiency of deleting data memory of the graph neural network.

[0005] In a first aspect, the present application provides a method for processing a graph neural network, comprising the following steps:

[0006] Acquire a graph neural network, wherein the graph neural network includes at least one sub-graph neural network, and the sub-graph neural network includes user data features of at least one user;

[0007] Determine a target subgraph neural network according to a data deletion request of a target user; the target subgraph neural network includes user data features of the target user;

[0008] Adjust the network parameters of the target subgraph neural network according to other user data features in the target subgraph neural network except the user data features indicated by the data deletion request.

[0009] In a second aspect, the present application further provides a processing device for a graph neural network, the processing device comprising:

[0010] An acquisition module, used to acquire a graph neural network, wherein the graph neural network includes at least one sub-graph neural network, and the sub-graph neural network includes user data features of at least one user;

[0011] A subgraph determination module, used to determine a target subgraph neural network according to a data deletion request of a target user; the target subgraph neural network includes user data features of the target user;

[0012] A subgraph adjustment module is used to adjust the network parameters of the target subgraph neural network according to other user data features in the target subgraph neural network except the user data features indicated by the data deletion request.

[0013] In a third aspect, the present application further provides a computer device, the computer device comprising a memory and a processor;

[0014] The memory is used to store computer programs;

[0015] The processor is used to execute the computer program and implement the steps of the above-mentioned graph neural network processing method when executing the computer program.

[0016] In a fourth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the graph neural network processing method as described above are implemented.

[0017] The present application provides a method, apparatus, device and storage medium for processing a graph neural network. The method for processing a graph neural network includes: obtaining a graph neural network, the graph neural network includes at least one sub-graph neural network, the sub-graph neural network includes user data features of at least one user; determining a target sub-graph neural network based on a data deletion request of a target user; the target sub-graph neural network includes user data features of the target user; adjusting network parameters of the target sub-graph neural network based on other user data features in the target sub-graph neural network except for the user data features indicated by the data deletion request, so as to improve the convenience and efficiency of adjusting network parameters of the target sub-graph neural network in the graph neural network, thereby improving the convenience and efficiency of data memory deletion of the graph neural network.

[0018] For example, the AI ​​model may include a graph neural network. If a target user initiates a data deletion request, the graph neural network may determine a target subgraph neural network from at least one subgraph neural network in response to the data deletion request, and the target subgraph neural network includes user data features of the target user. Accordingly, the graph neural network may delete the user data features indicated by the data deletion request in the target subgraph neural network, and the graph neural network may adjust the network parameters of the target subgraph neural network based on other user data in the target subgraph neural network except the user data features indicated by the data deletion request. Based on this, when a data deletion request is obtained, the graph neural network may delete the user data features indicated by the data deletion request, and the graph neural network may adjust the network parameters of the target subgraph neural network based on other user data features except the user data features indicated by the data deletion request without retraining the graph neural network, which is conducive to improving the convenience of adjusting the network parameters of the target subgraph neural network in the graph neural network and the efficiency of adjusting the network parameters, thereby improving the convenience of deleting data memory of the graph neural network and the efficiency of deleting data memory. For example, the processing method of the graph neural network can be used in financial scenarios. After the graph neural network answers the finance-related questions raised by the user, if the target user initiates a data deletion request to the graph neural network for the user data features involved in the question-and-answer process of financial questions, the graph neural network can respond to the data deletion request, delete the corresponding data from the target subgraph neural network where the user data features involved in the question-and-answer process of financial questions are located, and adjust the network parameters of the target subgraph neural network based on other user data features in the target subgraph neural network. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 It is a flowchart of a method for processing a graph neural network provided in an embodiment of the present application;

[0021] Figure 2 is a schematic block diagram of a graph neural network involved in an embodiment of the present application;

[0022] Figure 3 is a schematic block diagram of a graph neural network processing device provided in an embodiment of the present application;

[0023] Figure 4It is a schematic block diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0025] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0026] The embodiments of the present application provide a method, device, equipment and storage medium for processing a graph neural network. Among them, the processing method of the graph neural network can be applied to a computer device, which can be a mobile phone, a tablet computer, a laptop computer, a desktop computer and other devices. It can also be applied to a server, which can be a separate server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0027] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0028] See also Figure 1 , Figure 1 It is a flowchart of a method for processing a graph neural network provided in an embodiment of the present application. It should be noted that the method for processing a graph neural network provided in an embodiment of the present application can be used in a computer device, and of course can also be used in a server. For example, a server may be provided with a graph neural network. The server may obtain a data deletion request of a target user from a computer device, and process the data deletion request accordingly according to the processing method of the graph neural network to adjust the network parameters of a target sub-graph neural network in the graph neural network. Exemplarily, the server may, for example, send the network parameters of the target sub-graph neural network to a computer device; of course, this is not limited to this, and is not limited here.

[0029] In specific implementation, computer devices include but are not limited to: any one of mobile phones, tablet computers, laptop computers, and desktop computers; the server can be a single server or a server cluster, or a cloud server that provides cloud computing services.

[0030] like Figure 1 As shown, the graph neural network processing method includes steps S101 to S103.

[0031] Step S101: Obtain a graph neural network, where the graph neural network includes at least one sub-graph neural network, and the sub-graph neural network includes user data features of at least one user.

[0032] For example, a graph neural network can be divided into at least one sub-graph neural network. The sub-graph neural networks do not overlap each other. Accordingly, different sub-graph neural networks can correspond to different users, and the users corresponding to different sub-graph neural networks do not overlap.

[0033] In the case where the graph neural network includes at least one sub-graph neural network, the sub-graph neural network can be trained based on the user data of at least one user corresponding to the sub-graph neural network. For example, for financial scenarios, when a user queries financial issues through the graph neural network, the graph neural network can obtain the user data of the user when querying financial issues, and the graph neural network can train the sub-graph neural network based on the obtained user data. For social scenarios, when a user performs corresponding social operations through the graph neural network, the graph neural network can obtain the user data of the user when performing social operations, and the graph neural network can train the sub-graph neural network based on the obtained user data. Of course, it is not limited to this and is not limited here.

[0034] In the process of training the subgraph neural network, the subgraph neural network can perform feature extraction processing on the user data to obtain the user data features of the user, and the subgraph neural network can include the user data features of at least one user. Accordingly, by integrating all the trained subgraph neural networks, the graph neural network can be determined. For example, the graph neural network can be represented as a graph network Image Network It includes k subgraph neural networks, and the subgraph neural network can be referred to as subgraphs. It can be expressed as: in, Used to indicate graph networks Included sub Figure 1To subgraph k. A graph neural network can be called a graph model, and a subgraph neural network can be called a subgraph model. Then the graph model can be represented as graph model F. When the graph model includes k subgraph models, graph model F can also be represented as graph model F. i Graph Model F i It can be expressed as F i ={F1,F2,…,F k}

[0035] In some embodiments, based on at least one user, a node of a graph neural network is determined; the node corresponds to the user one-to-one, and the node includes user data features of the user; the node is randomly assigned to an initial subgraph neural network of the graph neural network; based on the initial subgraph neural network corresponding to the node and the initial subgraph neural network corresponding to the node's neighboring nodes, a subgraph reallocation process is performed on the node to obtain a subgraph neural network corresponding to the node; based on each subgraph neural network, a graph neural network is determined.

[0036] The nodes of the graph neural network can be determined based on the users corresponding to the graph neural network. For example, since the sub-graph neural network can include the user data features of at least one user, the sub-graph neural network can have a corresponding relationship with at least one user. During the training process of the sub-graph neural network, the graph neural network can, for example, change the corresponding relationship between the sub-graph neural network and at least one user. Figure 1 And son Figure 2 For example, the graph neural network can assign user data feature 1 of user 1 and user data feature 2 of user 2 to sub Figure 1 , and assigning user data feature 3 of user 3 and user data feature 4 of user 4 to sub Figure 2 If the graph neural network needs to redistribute user features 3 to the child Figure 1 , then the graph neural network needs to be Figure 2 User data feature 3 is obtained by screening out from all included user data features, and then user data feature 3 can be reallocated.

[0037] In order to improve the convenience of allocating user data features, the graph neural network can be provided with corresponding nodes, and the nodes correspond to users one by one, so as to establish a corresponding relationship between the nodes and the users, and thus establish a corresponding relationship between the nodes and the user data features of the users. For example, when there are N users using the graph neural network, the graph neural network can be provided with N nodes. Each node corresponds to one of the users, and each node can include the user data features of the corresponding user.

[0038] Graph neural networks include graph networks For example. It can also be expressed as:

[0039]

[0040] Among them, V is used to indicate the graph network A is used to indicate the node set corresponding to the node of the graph network. The neighboring node matrix corresponding to the node. X is used to indicate the graph network The feature matrix corresponding to the user data features included in each node.

[0041] For example, in the graph network In the case of including N nodes, the node set V may be a set integrated by the N nodes.

[0042] For example, graph networks Each node has a corresponding neighboring node, then the neighboring nodes of all nodes can form a neighboring node matrix. In the case of N nodes, A∈{0,1} N×N .

[0043] For example, graph networks Each node has the user data features of its own user, so the user data features of all nodes can form a feature matrix. In the case of N nodes, Among them, d X Used to indicate the feature dimension corresponding to the user data feature included in the node.

[0044] The graph neural network includes subgraphs Figure 1 And son Figure 2 ,son Figure 1 Assigned with user data feature 1 and user data feature 2, sub Figure 2 For example, user data feature 3 and user data feature 4 are assigned. In the graph neural network, user feature 3 needs to be reallocated to sub Figure 1 In the case of , if user 3 corresponds to node 3, the graph neural network can reallocate node 3 to child Figure 1 , the redistribution of user data feature 3 can be completed without Figure 2 User data feature 3 is selected from all included user data features.

[0045] Based on this, the graph neural network can determine the nodes of the graph neural network according to at least one user, so as to improve the convenience of allocating user data features to the user based on the correspondence between the nodes and the users.

[0046] When the nodes of the graph neural network are determined, the graph neural network can establish an association relationship between the nodes and the initial subgraph neural network. The initial subgraph neural network can be used to indicate the subgraph neural network before the nodes are subjected to subgraph redistribution processing.

[0047] For example, a graph neural network can randomly assign nodes to the initial sub-graph neural network. Graph Network Includes k initial subgraphs, and the graph network Take N nodes as an example, the graph network N nodes can be randomly assigned to k initial subgraphs.

[0048] After the graph neural network randomly assigns nodes to the initial subgraph neural network, the node number balance between different initial subgraph neural networks is generally low. Based on this, in order to improve the node number balance between different initial subgraph neural networks in the graph neural network, the graph neural network needs to perform subgraph redistribution processing on the nodes in different initial subgraph neural networks to determine the subgraph neural network.

[0049] For example, the graph neural network can perform subgraph redistribution processing on the node based on the initial subgraph neural network corresponding to the node and the initial subgraph neural network corresponding to the node's neighboring nodes to determine the subgraph neural network corresponding to the node. The nodes include node u as an example. The redistribution plan for node u during subgraph redistribution can be determined based on the initial subgraph neural network corresponding to node u and the initial subgraph neural network corresponding to the neighboring nodes of node u. According to the redistribution plan corresponding to node u, node u can be subjected to subgraph redistribution processing to obtain the subgraph neural network corresponding to node u.

[0050] In some embodiments, when it is determined that a node meets a preset reallocation condition based on an initial subgraph neural network corresponding to a node and an initial subgraph neural network corresponding to a neighboring node, the subgraph neural network corresponding to the node is determined based on the initial subgraph neural network corresponding to the neighboring node; and the node is reallocated to the subgraph neural network.

[0051] For example, the preset reallocation conditions may include at least one of the following: an initial subgraph neural network corresponding to a node is different from an initial subgraph neural network corresponding to a neighboring node, and the number of neighboring nodes corresponding to the same initial subgraph neural network is greater than or equal to a preset number threshold.

[0052] Take the initial subgraph neural network corresponding to node u as the subgraph For example.

[0053] In an exemplary embodiment, the initial subgraph neural network corresponding to the neighboring nodes of node u is a subgraph In the case of With subgraph Different, determine that node u meets the preset redistribution condition. Accordingly, the graph neural network can be based on the subgraph Determine the redistribution plan corresponding to node u to determine the subgraph neural network corresponding to node u.

[0054] In another exemplary embodiment, among the neighboring nodes of node u, there are multiple neighboring nodes corresponding to the same initial subgraph neural network, such as all corresponding to the subgraph In the case of , if the number of neighboring nodes corresponding to the same initial subgraph neural network is greater than or equal to the preset number threshold, the graph neural network can determine that node u meets the preset reallocation condition. Determine the redistribution plan corresponding to the node u to determine the subgraph neural network corresponding to the node u. The preset quantity threshold may be determined based on the maximum total quantity of neighboring nodes corresponding to the same initial subgraph neural network.

[0055] Of course, it is not limited to this. In the process of determining the subgraph neural network corresponding to node u, the graph neural network can also comprehensively consider that the initial subgraph neural network corresponding to node u is different from the initial subgraph neural network corresponding to the neighboring nodes, and the number of neighboring nodes corresponding to the same initial subgraph neural network is greater than or equal to the preset number threshold, to determine that node u meets the preset reallocation condition. Determine the redistribution plan corresponding to node u to determine the subgraph neural network corresponding to node u.

[0056] Based on the subgraph In the process of determining the reallocation plan corresponding to the node u, the reallocation plan can be used to indicate the subgraph neural network corresponding to the node u, so as to subsequently reallocate the node u to the subgraph neural network.

[0057] The reallocation plan of node u can be expressed as plan T u Plan T u It can be expressed in the form of a triple, such as plan T u It can be expressed as:

[0058]

[0059] Among them, Plan T u The subscript letter of can be used to indicate that the node corresponding to the plan is node u, and u is used to indicate node u. Used to indicate the initial subgraph neural network corresponding to node u, The initial subgraph neural network corresponding to the neighboring nodes of node u and the subgraph neural network corresponding to node u are used to indicate the current total number of nodes of the initial subgraph neural network corresponding to the neighboring nodes of node u, such as subgraph The current total number of nodes.

[0060] When the reallocation plan corresponding to node u is determined, the graph neural network can allocate node u to the corresponding subgraph neural network according to the reallocation plan corresponding to node u. Accordingly, in the process of reallocating node u to the subgraph neural network, the user data features of the users included in node u will also be reallocated to the subgraph neural network.

[0061] In this way, when the node meets the preset reallocation conditions, the subgraph neural network corresponding to the node is determined according to the initial subgraph neural network corresponding to the neighboring nodes, and the node and the neighboring nodes of the node can be reallocated to the same subgraph neural network as much as possible. Since the number of neighboring nodes of different nodes is limited, and the number of neighboring nodes of different nodes is generally the same, by reallocating the nodes and the neighboring nodes of the nodes to the same subgraph neural network, it is beneficial to improve the balance of the number of nodes between different subgraph neural networks. Based on the improvement of the balance of the number of nodes between different subgraph neural networks, the balance of computing resources required for training of subgraph neural networks is also improved accordingly. If it is necessary to adjust the network parameters of one or some subgraph neural networks in the future, it is not easy for a subgraph neural network to include too many nodes, resulting in the network parameters of the subgraph neural network being too complex, or for another subgraph neural network to include too few nodes, resulting in the network parameters of the subgraph neural network being too simple, which is beneficial to improve the balance of network parameter adjustment between different subgraph neural networks. Correspondingly, the improvement of the balance of network parameter adjustment between different sub-graph neural networks will be conducive to the subsequent improvement of the convenience and efficiency of network parameter adjustment of sub-graph neural networks, which will in turn be conducive to improving the convenience and efficiency of data memory deletion of graph neural networks.

[0062] Exemplarily, when the graph neural network determines the sub-graph neural network corresponding to each node, the graph neural network can also determine the reallocation order between the nodes so as to allocate the nodes to the sub-graph neural network according to the reallocation order.

[0063] In some embodiments, the reallocation order between the nodes is determined based on the size relationship between the current total number of nodes of the sub-graph neural network corresponding to each node; and the nodes are reallocated to the sub-graph neural network according to the reallocation order.

[0064] For example, based on the consideration of saving the computing resource overhead required for subgraph redistribution processing of nodes, the graph neural network can determine the order of reallocation between nodes according to the size relationship between the current total number of nodes of the subgraph neural network corresponding to each node. The order of reallocation between nodes can be directly proportional to the size relationship between the current total number of nodes of the subgraph neural network corresponding to each node. The larger the current total number of nodes of the subgraph neural network corresponding to the node, the higher the redistribution order of the node. For example, the graph neural network can sort the order of reallocation between nodes according to the size of the current total number of nodes θ in each node's respective redistribution plan, and obtain the corresponding set of reallocation plans. The set of reallocation plans can be expressed as The order between the reallocation plans in the reallocation plan set may be used to indicate the reallocation order.

[0065] Based on this, when the graph neural network reallocates nodes to subgraph neural networks in the order of redistribution, the graph neural network can give priority to filling the subgraph neural network with the largest current total number of nodes. Accordingly, after filling the subgraph neural network with the largest current total number of nodes, the size relationship between the current total number of nodes of each subgraph neural network will not change, and the graph neural network can continue to fill the subgraph neural network with the second largest current total number of nodes. And so on. Since the computing resource overhead required to fill the subgraph neural network with the largest current total number of nodes is the smallest, and the graph neural network does not need to use other computing resources to update the size relationship between the current total number of nodes of the subgraph neural network, it is beneficial to reduce the computing resource overhead required for subgraph redistribution processing of nodes, and thus help improve the efficiency of subgraph redistribution processing of nodes.

[0066] In an exemplary embodiment, when the graph neural network reallocates nodes to the subgraph neural network in accordance with the reallocation order, it can determine whether the current total number of nodes of the subgraph neural network corresponding to the node is less than the maximum node capacity of the subgraph neural network. In the case where the current total number of nodes is less than the maximum node capacity, the graph neural network can reallocate the nodes to the subgraph neural network in accordance with the reallocation order. Accordingly, in the case of reallocating the node to the subgraph neural network, the node originally included in the initial subgraph neural network will be deleted.

[0067] Take node u as an example. The redistribution plan T of node u u for: When the current total number of nodes θ is less than the maximum capacity of the node δ, the graph neural network can reallocate the node u to the subgraph Accordingly, graph neural networks can be Delete node u. The process of subgraph redistribution of node u can be expressed as:

[0068]

[0069]

[0070] in, Used to indicate that node u is reallocated to a subgraph Sub-images after Used to indicate that node u is reallocated to a subgraph Previous sub-image Accordingly, Used to indicate the subgraph after deleting node u Used to indicate the subgraph before deleting node u

[0071] In another exemplary embodiment, when the total number of current nodes is greater than or equal to the maximum capacity of the node, the graph neural network may determine that the subgraph neural network cannot accommodate more nodes temporarily, and the graph neural network may temporarily not perform subgraph redistribution processing on the node, and perform subgraph redistribution processing on subsequent nodes in the redistribution order until the total number of current nodes of the subgraph neural network is less than the maximum capacity of the node, and then perform subgraph redistribution processing on the node. Of course, this is not limited to this and is not limited here.

[0072] In the process of reallocating nodes to subgraph neural networks according to the reallocation order, the graph neural network can also detect whether the nodes included in each subgraph neural network have changed. If the nodes included in each subgraph neural network have changed, the graph neural network can determine that it is also necessary to perform subgraph reallocation processing on the nodes included in each subgraph neural network. Based on this, the graph neural network can re-determine the reallocation plan corresponding to each node and the reallocation order between the nodes to continue to perform subgraph reallocation processing on the nodes included in the graph neural network.

[0073] Correspondingly, if the nodes included in each subgraph neural network no longer change, the graph neural network can determine that there is no need to continue to perform subgraph reallocation processing on the nodes included in each subgraph neural network. The subgraph reallocation processing of the nodes included in the graph neural network has been completed, and the graph neural network can be determined based on the subgraph neural networks corresponding to each node.

[0074] In this way, the graph neural network can determine the reallocation order between each node based on the current total number of nodes in the sub-graph neural network corresponding to each node, and allocate the nodes to the sub-graph neural network according to the reallocation order, which is beneficial to reduce the computing resource overhead required for sub-graph reallocation processing of the nodes, and is beneficial to improve the processing convenience and efficiency of sub-graph reallocation processing of the nodes.

[0075] Exemplarily, when determining each sub-graph neural network, the graph neural network can determine the weight coefficient of each sub-graph neural network based on the user data features corresponding to the nodes in each sub-graph neural network, and then determine the graph neural network.

[0076] In some embodiments, the weight coefficient of the subgraph neural network is determined based on the user data characteristics corresponding to the nodes in the subgraph neural network; the graph neural network is determined based on each subgraph neural network and the weight coefficient of each subgraph neural network.

[0077] Since different sub-graph neural networks in a graph neural network include different nodes, different nodes correspond to different users, and different users correspond to different user data, the user data features corresponding to different nodes may also be different. Based on this, in the process of determining the graph neural network based on different sub-graph neural networks, different sub-graph neural networks may have their own weight coefficients. The weight coefficients of different sub-graph neural networks may be different or the same, so it is necessary to determine the weight coefficients of different sub-graph neural networks. Figure 2 As shown, the sub-graph neural network 1 to sub-graph neural network k included in the graph neural network can have different nodes, and the user data features corresponding to different nodes can be different, so the weight coefficients of different sub-graph neural networks need to be determined.

[0078] Exemplarily, the graph neural network can determine the weight coefficient of the sub-graph neural network based on the user data features corresponding to the nodes included in the sub-graph neural network.

[0079] For example, based on the user data characteristics corresponding to the node and the neighboring user data characteristics corresponding to the node's neighboring nodes, determine the first correlation relationship between the data prediction value corresponding to the node and the weight coefficient of the subgraph neural network; based on the loss value between the data prediction value and the data prediction label corresponding to the node, determine the second correlation relationship between the loss value and the weight coefficient; based on a preset weight coefficient optimization function, determine the weight coefficient of the subgraph neural network according to the first correlation relationship and the second correlation relationship.

[0080] The data prediction value can be used to indicate the predicted push demand of the user corresponding to the node. The data prediction label can be used to indicate the actual push demand of the user corresponding to the node. In different scenarios, the user's predicted push demand and actual push demand may be different. For financial scenarios, the graph neural network can predict the financial issues that the user is interested in to determine the user's predicted push demand, so as to push the corresponding financial issues and the solution process of the financial issues to the user in the future. For social scenarios, the graph neural network can predict the social content that the user is interested in to determine the user's predicted push demand, so as to push the corresponding social content to the user in the future. Of course, it is not limited to this and is not limited here. The data prediction label can be pre-marked or set by the user, which is not limited here.

[0081] Image Network Included nodes v, graph network Include sub Figure 1 Take subgraph k as an example. Can be integrated Figure 1 The influence of subgraph k on node v is used to predict the data value p corresponding to node v. v Determine. Data prediction value p v It can be by Figure 1 to subgraph k. Graph Network You can get the user data features corresponding to node v, such as feature X v , and the neighboring user data features corresponding to the neighboring nodes of node v, such as feature Y v Graph Network For example, by evaluating Figure 1 To each subgraph k for feature X v And feature Y v The influence of the node v is used to determine the data prediction value p corresponding to the node v. v . Accordingly, the graph network The data prediction value p corresponding to node v can be established v Separately with the child Figure 1 To the weight coefficient α of each subgraph k i The first correlation between the two. Data prediction value p v It can be expressed as:

[0082]

[0083] Among them, α i ={α1,α2,…,α k} is used to indicate the child Figure 1 To the weight coefficient of each subgraph k. i ={F1,F2,…,F k} is used to indicate the child Figure 1To subgraph k. F i (X v ,Y v ) is used to indicate the evaluation sub Figure 1 To each subgraph k, the data prediction value p corresponding to the node v v impact.

[0084] Based on this, the graph network You can get the feature X corresponding to node v v , and the features Y corresponding to the neighboring nodes of node v v , and substitute the data prediction value p corresponding to node v v In the expression of, we get the data prediction value p corresponding to node v v With the child Figure 1 To the weight coefficient α of each subgraph k i The first relationship between them.

[0085] Graph Network You can also get the data prediction label corresponding to node v, such as label z. The value p can be predicted based on the data v And label z, predict the value p of the data v The loss value between the label z is determined. For example, the data prediction value p v The loss value between the label z can be expressed as L(p v ,z).

[0086] In determining the data prediction value p v In the case of the loss value between the label z, the graph network A second correlation relationship between the loss value and the weight coefficient of the subgraph neural network may be determined. For example, the second correlation relationship may be expressed as:

[0087]

[0088] Here, λ may be preset and is not limited here.

[0089] When the first association relationship and the second association relationship are determined, the graph network The first association relationship and the second association relationship may be processed based on a preset weight coefficient optimization function to determine the sub- Figure 1 To the weight coefficient of each subgraph k.

[0090] The preset weight coefficient optimization function can be expressed as:

[0091]

[0092] Based on the preset weight coefficient optimization function, the graph network Can pair Figure 1 The weight coefficients of each subgraph k are minimized to obtain the final subgraph Figure 1 To the weight coefficient of each subgraph k.

[0093] In determining the Figure 1 When the weight coefficients of each subgraph k are obtained, the subgraphs can be combined Figure 1 To subgraph k and subgraph Figure 1 To the weight coefficient of each subgraph k, determine the graph network

[0094] In this way, the weight coefficient of the subgraph neural network can be determined based on the user data characteristics corresponding to the nodes in the subgraph neural network, which is conducive to improving the convenience of determining the weight coefficient of the subgraph neural network. When determining the weight coefficient of the subgraph neural network, the subgraph neural network can be combined to determine the graph neural network, which is conducive to improving the convenience of determining the graph neural network.

[0095] When the graph neural network is obtained, the graph neural network can be used to subsequently determine the target subgraph neural network, and then adjust the network parameters of the target subgraph neural network.

[0096] Step S102: Determine a target subgraph neural network according to a data deletion request of a target user; the target subgraph neural network includes user data features of the target user.

[0097] For example, when the graph neural network receives a data deletion request from any user, the graph neural network can determine the user who initiated the data deletion request as the target user. Accordingly, the graph neural network can determine the target subgraph neural network in response to the data deletion request of the target user. The target subgraph neural network includes the user data features of the target user.

[0098] Graph neural networks include graph networks Graph Network Include sub Figure 1 Take subgraph k as an example. When receiving a data deletion request from a target user, if the subgraph neural network where the target user’s user data features are located is a subgraph neural network, Figure 1 , then the graph network You can Figure 1 Determined as the target subgraph neural network.

[0099] In some embodiments, when the cumulative number of data deletion requests received by a sub-graph neural network is greater than or equal to a preset number threshold, the sub-graph neural network is determined to be a target sub-graph neural network.

[0100] For example, the graph neural network can count the cumulative number of data deletion requests received by each sub-graph neural network. The data deletion requests received by the same sub-graph neural network can be issued by the same target user or by different target users, without limitation.

[0101] Graph neural networks include graph networks Graph Network Include sub Figure 1 Take subgraph k as an example. Recorded sub Figure 1 The cumulative number of data deletion requests corresponding to each subgraph k can be expressed as record C i . Record C i It can be expressed as: C i ={C1,C2,…,C k}.

[0102] When determining the target subgraph neural network, the graph neural network can determine the target subgraph neural network based on the cumulative number of data deletion requests received by each subgraph neural network. For example, when the cumulative number of subgraph neural networks corresponding to the existence is greater than or equal to the preset number threshold, the subgraph neural network is determined to be the target subgraph neural network. When the cumulative number of subgraph neural networks corresponding to the existence is less than the preset number threshold, the subgraph neural network is determined to be a non-target subgraph neural network. And so on. The preset number threshold can be pre-set or set by the user, and is not limited here.

[0103] In this way, when the graph neural network determines the target subgraph neural network, the target subgraph neural network can be used for subsequent adjustment of network parameters, thereby deleting data memory of the graph neural network.

[0104] Step S103: adjusting the network parameters of the target subgraph neural network according to other user data features in the target subgraph neural network except the user data features indicated by the data deletion request.

[0105] For example, in the case of determining the target subgraph neural network, the graph neural network can delete the user data features indicated by the data deletion request in the target subgraph neural network in response to the data deletion request received by the target subgraph neural network, and obtain the processed user data features, that is, other user data features except the user data features indicated by the data deletion request. Accordingly, in response to the change of the user data features in the target subgraph neural network, the graph neural network can adjust the network parameters of the target subgraph neural network according to the user data features other than the user data features indicated by the data deletion request in the target subgraph neural network. Based on this, in the process of adjusting the network parameters of the target subgraph neural network, the graph neural network can avoid the influence of the user data features indicated by the data deletion request on the network parameters of the target subgraph neural network, which is equivalent to deleting the data memory corresponding to the user data features indicated by the data deletion request, which is conducive to improving the convenience and efficiency of adjusting the network parameters of the target subgraph neural network, so as to improve the convenience and efficiency of deleting data memory of the graph neural network.

[0106] For graph neural networks, except for the target subgraph neural network, there is no need to adjust the network parameters because there is no deletion of user data features.

[0107] In this way, when a data deletion request is received, the graph neural network only needs to adjust the network parameters of the target subgraph neural network corresponding to the data deletion request, without adjusting the network parameters of the non-target subgraph neural network. This is beneficial to improving the convenience and efficiency of network parameter adjustment of the target subgraph neural network in the graph neural network, thereby improving the convenience and efficiency of data memory deletion of the graph neural network.

[0108] The processing method of the graph neural network provided by the above embodiment is by obtaining a graph neural network, the graph neural network includes at least one sub-graph neural network, the sub-graph neural network includes user data characteristics of at least one user; according to the data deletion request of the target user, determining the target sub-graph neural network; the target sub-graph neural network includes user data characteristics of the target user; according to other user data characteristics in the target sub-graph neural network except the user data characteristics indicated by the data deletion request, adjusting the network parameters of the target sub-graph neural network to improve the convenience of adjusting the network parameters of the target sub-graph neural network in the graph neural network and the efficiency of adjusting the network parameters, thereby improving the convenience of deleting data memory of the graph neural network and the efficiency of deleting data memory.

[0109] For example, the AI ​​model may include a graph neural network. If a target user initiates a data deletion request, the graph neural network may determine a target subgraph neural network from at least one subgraph neural network in response to the data deletion request, and the target subgraph neural network includes user data features of the target user. Accordingly, the graph neural network may delete the user data features indicated by the data deletion request in the target subgraph neural network, and the graph neural network may adjust the network parameters of the target subgraph neural network based on other user data in the target subgraph neural network except the user data features indicated by the data deletion request. Based on this, when a data deletion request is obtained, the graph neural network may delete the user data features indicated by the data deletion request, and the graph neural network may adjust the network parameters of the target subgraph neural network based on other user data features except the user data features indicated by the data deletion request without retraining the graph neural network, which is conducive to improving the convenience of adjusting the network parameters of the target subgraph neural network in the graph neural network and the efficiency of adjusting the network parameters, thereby improving the convenience of deleting data memory of the graph neural network and the efficiency of deleting data memory. For example, the processing method of the graph neural network can be used in financial scenarios. After the graph neural network answers the finance-related questions raised by the user, if the target user initiates a data deletion request to the graph neural network for the user data features involved in the question-and-answer process of financial questions, the graph neural network can respond to the data deletion request, delete the corresponding user data features from the target sub-graph neural network where the user data features involved in the question-and-answer process of financial questions are located, and adjust the network parameters of the target sub-graph neural network based on other user data features in the target sub-graph neural network, thereby achieving the effect of deleting data memory of the graph neural network.

[0110] See also Figure 3 , Figure 3 It is a schematic block diagram of a processing device of a graph neural network provided in an embodiment of the present application. The processing device can be configured in a server or an electronic device to execute the aforementioned processing method of the graph neural network.

[0111] like Figure 3 As shown, the processing device includes: an acquisition module 110 , a sub-image determination module 120 and a sub-image adjustment module 130 .

[0112] An acquisition module 110, configured to acquire a graph neural network, wherein the graph neural network includes at least one sub-graph neural network, and the sub-graph neural network includes user data features of at least one user;

[0113] The subgraph determination module 120 is used to determine a target subgraph neural network according to a data deletion request of a target user; the target subgraph neural network includes user data features of the target user;

[0114] The subgraph adjustment module 130 is used to adjust the network parameters of the target subgraph neural network according to other user data features in the target subgraph neural network except the user data features indicated by the data deletion request.

[0115] It should be noted that technicians in the relevant field can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described devices and modules and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0116] The method of the present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can 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. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0117] Exemplarily, the above method and apparatus may be implemented in the form of a computer program, which may be run on a computer device.

[0118] See also Figure 4 , Figure 4 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server or an electronic device.

[0119] like Figure 4 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a storage medium and an internal memory.

[0120] The storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any one of the graph neural network processing methods.

[0121] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0122] The internal memory provides an environment for the operation of the computer program in the storage medium. When the computer program is executed by the processor, the processor can execute any graph neural network processing method.

[0123] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will appreciate that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0124] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0125] In one embodiment, the processor is used to execute a computer program and implement the following steps when executing the computer program:

[0126] Acquire a graph neural network, wherein the graph neural network includes at least one sub-graph neural network, and the sub-graph neural network includes user data features of at least one user;

[0127] Determine a target subgraph neural network according to a data deletion request of a target user; the target subgraph neural network includes user data features of the target user;

[0128] Adjust the network parameters of the target subgraph neural network according to other user data features in the target subgraph neural network except the user data features indicated by the data deletion request.

[0129] It should be noted that technical personnel in the relevant field can clearly understand that, for the convenience and conciseness of description, the specific working process of the above-mentioned description of the processing of the graph neural network can refer to the corresponding process in the aforementioned graph neural network processing method embodiment, and will not be repeated here.

[0130] An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. The method implemented when the computer program is executed by a processor can refer to the various embodiments of the processing method of the graph neural network of the present application.

[0131] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., equipped on the computer device.

[0132] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0133] It should also be understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations. It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system 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, article or system. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or system including the element.

[0134] It should also be understood that the non-Company software tools or components appearing in the embodiments of the present application are merely examples and do not represent actual use.

[0135] The serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments. The above description is only a specific implementation mode of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A processing method of a graph neural network, characterized in that: include: Acquire a graph neural network, wherein the graph neural network includes at least one sub-graph neural network, and the sub-graph neural network includes user data features of at least one user; Determine the target subgraph neural network according to the target user's data deletion request; The target subgraph neural network includes user data features of the target user; Adjust the network parameters of the target subgraph neural network according to other user data features in the target subgraph neural network except the user data features indicated by the data deletion request.

2. The processing method according to claim 1, characterized in that: The step of determining a target subgraph neural network according to a data deletion request of a target user includes: When the cumulative number of data deletion requests received by a sub-graph neural network is greater than or equal to a preset number threshold, the sub-graph neural network is determined to be a target sub-graph neural network.

3. The processing method according to claim 1 or 2, characterized in that: The obtaining of the graph neural network comprises: Determine a node of the graph neural network according to at least one user; the node corresponds to the user one-to-one, and the node includes user data features of the user; Randomly assigning the node to an initial subgraph neural network of the graph neural network; According to the initial subgraph neural network corresponding to the node and the initial subgraph neural network corresponding to the neighboring nodes of the node, subgraph redistribution processing is performed on the node to obtain the subgraph neural network corresponding to the node; The graph neural network is determined based on each of the sub-graph neural networks.

4. The processing method according to claim 3, characterized in that: The subgraph redistribution processing is performed on the node according to the initial subgraph neural network corresponding to the node and the initial subgraph neural network corresponding to the neighboring nodes of the node to obtain the subgraph neural network corresponding to the node, including: When it is determined that the node meets the preset reallocation condition according to the initial subgraph neural network corresponding to the node and the initial subgraph neural network corresponding to the neighboring node, the subgraph neural network corresponding to the node is determined according to the initial subgraph neural network corresponding to the neighboring node; Reassigning the nodes to the subgraph neural network.

5. The processing method according to claim 4, characterized in that: The reallocating the nodes to the subgraph neural network comprises: Determine the order of reallocation between the nodes according to the size relationship between the total number of current nodes of the subgraph neural network corresponding to each of the nodes; According to the reallocation order, the nodes are reallocated to the subgraph neural network.

6. The processing method according to claim 3, characterized in that: Determining the graph neural network according to each of the sub-graph neural networks includes: Determining a weight coefficient of the subgraph neural network according to user data features corresponding to nodes in the subgraph neural network; The graph neural network is determined according to each of the sub-graph neural networks and the weight coefficient of each of the sub-graph neural networks.

7. The processing method according to claim 6, characterized in that: The step of determining a weight coefficient of the subgraph neural network according to user data features corresponding to nodes in the subgraph neural network includes: Determine a first correlation relationship between a data prediction value corresponding to the node and a weight coefficient of the subgraph neural network according to a user data feature corresponding to the node and a neighboring user data feature corresponding to a neighboring node of the node; Determining a second association relationship between the loss value and the weight coefficient according to a loss value between the data prediction value and the data prediction label corresponding to the node; Based on a preset weight coefficient optimization function, the weight coefficient of the subgraph neural network is determined according to the first association relationship and the second association relationship.

8. A processing device for a graph neural network, characterized in that: The processing device comprises: An acquisition module, used to acquire a graph neural network, wherein the graph neural network includes at least one sub-graph neural network, and the sub-graph neural network includes user data features of at least one user; A subgraph determination module, used to determine a target subgraph neural network according to a data deletion request of a target user; the target subgraph neural network includes user data features of the target user; A subgraph adjustment module is used to adjust the network parameters of the target subgraph neural network according to other user data features in the target subgraph neural network except the user data features indicated by the data deletion request.

9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the processing method of the graph neural network as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for processing a graph neural network as described in any one of claims 1 to 7 are implemented.

Images