WO2022237175A1

WO2022237175A1 - Graph data processing method and apparatus, device, storage medium, and program product

Info

Publication number: WO2022237175A1
Application number: PCT/CN2021/140229
Authority: WO
Inventors: 吴子凡; 张潮宇; 陈天健; 杨强
Original assignee: 深圳前海微众银行股份有限公司
Priority date: 2021-05-10
Filing date: 2021-12-21
Publication date: 2022-11-17
Also published as: CN113240505B; CN113240505A

Abstract

The present application discloses a graph data processing method and apparatus, a device, a storage medium, and a program product. The method comprises: traversing nodes in a first sub-graph during any round of iterations; for each traversed node, searching for a neighbor node of the node in the first sub-graph, performing an aggregation operation according to an eigenvector of the neighbor node of the previous round of iterations to obtain a first aggregation result; if the node has a connection relationship with a node in a second sub-graph, determining a final aggregation result of the node according to the first aggregation result and a second aggregation result, wherein the second aggregation result is determined by a second participant according to the eigenvector of the node during the previous round of iterations of the neighbor node in the second sub-graph; and determining the eigenvector of the current round of iterations of the node according to the final aggregation result. Once the number of iterations meets requirements, the eigenvector of the last round of iterations is used to calculate a prediction result corresponding to the node. The prediction accuracy of graph data can be effectively increased.

Description

Graph data processing method, device, device, storage medium and program product

This application claims the priority of the Chinese patent application with the application number 202110507515.0 and the application name "Graph data processing method, device, equipment, storage medium and program product" submitted to the China Patent Office on May 10, 2021, all of which The contents are incorporated by reference in this application.

technical field

The present application relates to the technical field of data processing, and in particular to a method, device, equipment, storage medium, and program product for processing graph data.

Background technique

With the continuous development of computer technology and big data processing technology, the application of deep learning is becoming more and more extensive. The graph neural network gradually replaces the traditional artificially designed graph features to extract the hidden value behind the graph data and make related recognition predictions. deal with. For example, based on the graph data constructed by financial institutions, it is possible to identify whether users have overdue risks, etc.

In practical applications, massive graph data are often distributed in different organizations, but due to data privacy requirements, a single organization cannot use data from other organizations for analysis and processing, resulting in poor prediction accuracy for graph data.

technical problem

The main purpose of the present application is to provide a processing method, device, equipment, storage medium and program product for graph data, aiming at improving the accuracy of prediction for graph data.

To achieve the above purpose, the present application provides a method for processing graph data, the graph data includes a first subgraph and a second subgraph, the first subgraph includes nodes belonging to a first participant, and the second The subgraph includes nodes belonging to a second party; the method is applied to the first party, the method comprising:

During any round of iteration, traverse the nodes in the first subgraph, and perform the following operations for each traversed node:

Finding the neighbor nodes of the node in the first subgraph, and performing an aggregation operation according to the feature vector of the last round of iteration process of the neighbor nodes, to obtain a first aggregation result;

If the node has a connection relationship with the nodes in the second subgraph, then determine the final aggregation result of the node according to the first aggregation result and the second aggregation result; wherein the second aggregation result is the The second participant is determined according to the eigenvectors of the last round of iterative process of the neighbor nodes of the node in the second subgraph;

According to the final aggregation result, the eigenvector of the current round of iterative process of the node is determined; after the number of iterations meets the requirement, the eigenvector of the last round of iterative process is used to calculate the prediction result corresponding to the node.

Optionally, if the node has a connection relationship with the nodes in the second subgraph, determining the final aggregation result of the node according to the first aggregation result and the second aggregation result includes:

If the node has a connection relationship with a node in the second subgraph, send request information to the second participant, where the request information is used to request the second participant to calculate the first node corresponding to the node two aggregation results and encrypting said second aggregation result;

receiving the encrypted second aggregation result sent by the second participant;

Determine the final aggregation result of the node according to the encrypted second aggregation result.

Optionally, the encrypted second aggregation result is a second aggregation result encrypted with a public key; according to the encrypted second aggregation result, determining the final aggregation result of the node includes:

encrypting the first aggregation result using the public key;

Based on the random mask, calculating the encrypted first aggregation result and the encrypted second aggregation result to obtain the encrypted final aggregation result;

sending the encrypted final aggregation result to a third party, so that the third party uses a private key to decrypt the encrypted final aggregation result;

Receive the decryption result sent by the third participant, and perform a random masking operation on the decryption result to obtain the final aggregation result.

Optionally, searching for neighbor nodes of the node in the first subgraph, and performing an aggregation operation according to the feature vectors of the last iteration process of the neighbor nodes, including:

Find all neighbor nodes of the node in the first subgraph;

Based on the sampling operation with replacement, select a preset number of neighbor nodes from the found neighbor nodes;

The first aggregation result is calculated according to the eigenvectors of the last round of iterative process of the selected neighbor nodes.

Optionally, the method also includes:

constructing nodes in the first subgraph according to user accounts belonging to the first participant;

According to the transfer record of the user account of the first participant, construct the connection relationship of the nodes in the first subgraph, and the connection relationship is used to determine the neighbor nodes;

Correspondingly, after the number of iterations meets the preset requirements, the method further includes:

According to the feature vector corresponding to any node in the first subgraph in the last iteration process, determine the service risk information of the user account corresponding to the node.

Optionally, the method also includes:

determining an initial feature vector corresponding to a node in the first subgraph according to the attribute information of the user account of the first participant;

Wherein, the eigenvector of the last iterative process used in the first iterative process is the initial eigenvector.

Optionally, determining the business risk information of the user account corresponding to the node includes:

Determine whether the user account is an abnormal account, and if it is determined according to the feature vector that the user account belongs to an abnormal account, perform reporting processing; or,

Determine whether the user account has an overdue risk, and if it is determined according to the feature vector that the user account has an overdue risk, monitor the user account, or adjust the credit level of the user account.

Optionally, determining the final aggregation result of the node according to the first aggregation result and the second aggregation result includes:

A final aggregation result is determined through a nonlinear algorithm according to the first aggregation result and the second aggregation result.

Optionally, the graph data is used to implement social behavior analysis, the nodes in the graph data are used to represent users, and the preset association relationships include family relationships and employment relationships, and the preset association relationships are used to determine neighbor nodes.

The present application also provides a graph data processing device, the graph data includes a first subgraph and a second subgraph, the first subgraph includes nodes belonging to the first participant, and the second subgraph includes nodes belonging to A node of a second participant; the device is applied to the first participant, the device comprising:

An execution module, configured to traverse the nodes in the first subgraph during any iteration, and perform the following operations for each traversed node;

A search module, configured to search for neighbor nodes of the node in the first subgraph, perform an aggregation operation according to the feature vectors of the last round of iteration process of the neighbor nodes, and obtain a first aggregation result;

An aggregation module, configured to determine the final aggregation result of the node according to the first aggregation result and the second aggregation result when the node has a connection relationship with the nodes in the second sub-graph; wherein, the first The second aggregation result is determined by the second participant according to the eigenvectors of the last iteration process of the neighbor nodes of the node in the second subgraph;

The determination module is configured to determine the eigenvector of the current round of iterative process of the node according to the final aggregation result; after the number of iterations meets the requirement, the eigenvector of the last round of iterative process is used to calculate the prediction result corresponding to the node.

Optionally, the aggregation module is specifically used for:

Optionally, the encrypted second aggregation result is a second aggregation result encrypted with a public key; the aggregation module determines the final aggregation result of the node according to the encrypted second aggregation result , specifically for:

encrypting the first aggregation result using the public key;

Optionally, the search module is specifically used for:

Find all neighbor nodes of the node in the first subgraph;

Optionally, the execution module is also used for:

Correspondingly, after the number of iterations meets the preset requirements, the executing module is also used for:

Optionally, the execution module is also used for:

Optionally, when the execution module determines the business risk information of the user account corresponding to the node, it is specifically used to:

Optionally, when the aggregation module determines the final aggregation result of the node according to the first aggregation result and the second aggregation result, it is specifically used to:

The present application also provides a graph data processing device, the graph data processing device includes: a memory, a processor, and a graph data processing program stored in the memory and operable on the processor, the When the graph data processing program is executed by the processor, the steps of the graph data processing method described in any one of the preceding items are realized.

The present application also provides a computer-readable storage medium, the computer-readable storage medium stores a processing program for graph data, and when the processing program for graph data is executed by a processor, the graph as described in any one of the preceding items is realized. The steps of the data processing method.

The present application also provides a computer program product, including a computer program. When the computer program is executed by a processor, the method described in any one of the preceding items is implemented.

In this application, graph data including a first subgraph including nodes belonging to a first participant and a second subgraph including nodes belonging to a second participant may be processed. node, in any round of iteration process, the first participant can traverse the nodes in the first subgraph, and for each node traversed, find the neighbor nodes of the node in the first subgraph, Perform an aggregation operation according to the eigenvectors of the last iteration of the neighbor node to obtain a first aggregation result, if the node has a connection relationship with a node in the second sub-graph, then according to the first aggregation result and the second aggregation result to determine the final aggregation result of the node, wherein the second aggregation result is the last round of iteration process of the second participant according to the neighbor nodes of the node in the second subgraph Determined by the eigenvector of the node, according to the final aggregation result, determine the eigenvector of the current round of iterative process of the node, after the number of iterations meets the requirements, the eigenvector of the last round of iterative process is used to calculate the prediction result corresponding to the node, In the process of node analysis, the aggregation results of the first participant on the neighbor nodes of the first subgraph and the aggregation results of the second participant on the neighbor nodes of the second subgraph are integrated, so that the results can be more comprehensive and accurate Extracting the characteristics of nodes can comprehensively utilize the data of all parties in the case of barriers to data interoperability, jointly realize the processing of graph data, and effectively improve the accuracy of prediction based on graph data.

Description of drawings

FIG. 1 is a schematic diagram of an application scenario provided by an embodiment of the present application;

Fig. 2 is a schematic diagram of a kind of graph data provided by the embodiment of the present application;

FIG. 3 is a schematic flowchart of a method for processing graph data provided in an embodiment of the present application;

FIG. 4 is a system architecture diagram of graph data processing provided by an embodiment of the present application;

FIG. 5 is a schematic flowchart of another graph data processing method provided by the embodiment of the present application;

FIG. 6 is a schematic structural diagram of a graph data processing device provided in an embodiment of the present application;

FIG. 7 is a schematic structural diagram of a graph data processing device provided by an embodiment of the present application.

The realization, functional features and advantages of the present application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Embodiments of the present invention

Exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present application can be more thoroughly understood, and the scope of the present application can be fully conveyed to those skilled in the art.

Association analysis, or graph analysis, is an important class of analysis methods. Through such methods, operators can easily model and analyze the relationship between entities, and determine whether there is a specific connection form or important node in the network. Traditional analysis methods mainly use artificially designed graph features as the analysis object, such as PageRank (web page ranking), centrality and so on. In recent years, with the rise of deep learning, graph neural networks have gradually replaced traditional artificially designed graph features to extract the value hidden behind graph data.

The application of graph neural network needs to rely on graph data, but in actual operation, graph data often involves multiple institutions. Subject to the relevant requirements of data privacy, these data cannot be collected to form an effective network to obtain more accurate prediction results.

FIG. 1 is a schematic diagram of an application scenario provided by an embodiment of the present application. As shown in Figure 1, Institution 1 and Institution 2 are both banks, and both Institution 1 and Institution 2 have multiple user accounts. For example, Institution 1 includes User Account A, User Account B, User Account C, etc. Institution 2 includes User Account Account D, user account E, user account F, etc. The user account number may specifically be a bank card number or the like. Connections between different user accounts represent transfer records between them.

Based on the transfer relationship between user accounts and the basic attribute information of user accounts, graph data can be constructed, because the graph data contains the user's basic attribute information and related financial characteristics, so the overdue risk prediction of user accounts can be performed based on the graph data Or identify abnormal accounts, etc., to meet monitoring requirements.

In practical applications, the user account in organization 1 will not only have a transfer relationship with other user accounts in the organization, but may also have a transfer relationship with the user account in organization 2. As shown in Figure 1, user account B in organization 1 has a transfer relationship with user account A and user account C of the organization, and also has a transfer relationship with user account E in organization 2. However, due to the incompatibility of data between institutions, institution 1 cannot obtain the detailed information of user account E in institution 2, and can only analyze and process user account B based on the detailed information of user accounts A and C of its own institution, thus losing The valid information of user account B in other institutions leads to poor accuracy of prediction and identification.

In view of this, the embodiment of the present application provides a method for processing graph data, which can process graph data in cooperation with different participants. The graph data may comprise a first subgraph comprising nodes belonging to a first party and a second subgraph comprising nodes belonging to a second party. Taking the application of the graph data in financial institutions as an example, the first participant and the second participant can both be financial institutions such as banks, the nodes in the graph data can be used to represent user accounts, and the connection relationship between nodes can be represented by Indicates the transfer relationship between user accounts.

When processing the nodes in the graph data, the first participant can perform an aggregation operation according to the neighbor nodes of the node in the first sub-graph to obtain the first aggregation result; the second participant can obtain the first aggregation result according to the node Neighboring nodes in the second subgraph perform an aggregation operation to obtain a second aggregation result; according to the first aggregation result and the second aggregation result, a final aggregation result corresponding to the node can be obtained; performing subsequent processing according to the final aggregation result, The prediction results corresponding to the nodes can be obtained.

Fig. 2 is a schematic diagram of a kind of graph data provided by the embodiment of the present application. As shown in Figure 2, the graph data G may include a first subgraph G1 and a second subgraph G2, the small circles in the figure represent nodes, and the first subgraph G1 includes nodes v11, v12, v13, v14 belonging to the first participant , v15, the second subgraph G2 includes nodes v21, v22, v23, and v24 belonging to the second participant, each node represents a user account, and the connection lines in the figure show the connection relationship between the nodes.

When processing nodes in graph data, there can be multiple rounds of iterative process. In each round of iteration, all nodes can be traversed, and for any node, data from multiple parties can be combined for analysis and processing. For example, for node v12, the first participant can perform an aggregation operation based on its neighbor nodes in the first subgraph G1, that is, nodes v11 and v14, to obtain the first aggregation result, and the second participant can obtain the first aggregation result based on its neighbor nodes in the first subgraph G1. Neighbor nodes in the second subgraph G2, ie, nodes v21, v22, and v23, perform an aggregation operation to obtain a second aggregation result. According to the first aggregation result and the second aggregation result, the final aggregation result corresponding to node v12 can be obtained. According to the final aggregation result, the feature vector of node v12 in this round can be determined, and the feature vector of this round is used for the next round of aggregation operate. After the number of iterations meets the requirements, the final feature vector can be used to calculate the prediction result, such as whether node v12 has an overdue risk, etc.

In the method provided by the embodiment of this application, the data of each participant does not leave the local area, and it is possible to jointly predict graph data with multiple parties while ensuring data security. Moreover, since the aggregation result of the first participant on the neighbor nodes of the first subgraph and the aggregation result of the second participant on the neighbor nodes of the second subgraph are integrated during the node analysis process, it can be more comprehensive , Accurately reflect the transfer relationship of the nodes, so as to extract the fund flow characteristics of the nodes more accurately, and effectively improve the accuracy of predictions on the basis of ensuring data security.

Some implementations of the present application will be described in detail below in conjunction with the accompanying drawings. Under the condition that there is no conflict between the various embodiments, the following embodiments and the features in the embodiments can be combined with each other.

FIG. 3 is a schematic flowchart of a method for processing graph data provided by an embodiment of the present application. The graph data includes a first subgraph comprising nodes belonging to a first party and a second subgraph comprising nodes belonging to a second party. The method provided in this embodiment may be applied to the first participant. The first participant may process the graph data through multiple rounds of iterative process. As shown in Figure 3, in any round of iteration, the nodes in the first subgraph can be traversed, and for each node traversed, the following operations are performed:

Step 301 : Find the neighbor nodes of the currently traversed node in the first subgraph, perform an aggregation operation according to the eigenvectors of the last iteration process of the neighbor nodes, and obtain a first aggregation result.

Taking the current iteration process as the kth iteration process as an example, for the currently traversed node, you can find the neighbor nodes of the current node in the first subgraph, where the neighbor nodes can be nodes that have a direct connection relationship with the current node , according to the eigenvectors of neighbor nodes in the k-1th round, the first aggregation result of the k-th round of the current node can be determined.

When k=1, the eigenvector of the previous round of iterative process can be the initial eigenvector. Optionally, before performing the iterative process, each node in the graph data may be initialized to determine the initial feature vector of each node.

In the embodiment of the present application, the feature vector of a node may be any information that can characterize the feature of the node. Optionally, after the graph data is obtained, the node may be assigned a value according to the attribute information of each node in the graph data to obtain a corresponding initial feature vector.

In the first iteration process, the first participant can traverse each node in the first subgraph, and perform an aggregation operation according to the initial feature vectors corresponding to the neighbor nodes of the current node to obtain the first aggregation result of the current node.

Step 302: If the node has a connection relationship with the nodes in the second sub-graph, determine the final aggregation result of the node according to the first aggregation result and the second aggregation result.

Wherein, the second aggregation result is determined by the second participant according to the eigenvectors of the node's neighbor nodes in the second subgraph in the last round of iterative process.

Optionally, judging whether any two nodes have a connection relationship can be achieved in the following manner: judging whether the two nodes have a preset association relationship, wherein the preset association relationship can be set according to actual needs, there are Two nodes with a preset association relationship can be regarded as having a connection relationship in graph data. The connection relationship can be used to determine neighbor nodes.

For example, in a risk control scenario, the preset association relationship may be a transfer relationship. Specifically, the nodes in the graph data can be used to represent user accounts, and the connection relationship between nodes can be used to represent the transfer relationship between user accounts; if a certain node in the first sub-graph and a certain node in the second sub-graph If there is a transfer record between them, it can be considered that the two have a connection relationship. For another example, in a social behavior analysis scenario, nodes in graph data may be used to represent users, and the preset association relationship may be family relationship, employment relationship, and the like.

During the first iteration, if the current node traversed by the first participant has neighbor nodes in the second subgraph, the second participant can correspond to the current node according to the neighbor nodes in the second subgraph. The initial eigenvector of , perform an aggregation operation, and obtain the second aggregation result of the current node.

The first participant may determine the final aggregation result of the first round of iteration process of the current node according to the first aggregation result and the second aggregation result of the first round of iteration process of the current node.

Optionally, the final aggregation result may be the sum of the first aggregation result and the second aggregation result.

Or, when calculating the final aggregation result according to the first aggregation result and the second aggregation result, the final aggregation result can be determined through a nonlinear algorithm, such as a log function, an exponential function, taking a maximum value, taking a minimum value, etc., so that the result With higher nonlinearity, it can fit more complex situations.

Step 303 , according to the final aggregation result, determine the feature vector of the current round of iteration process of the node.

For example, after the final aggregation result of the first round of iteration process of the current node is determined, the feature vector of the first round of iteration process of the current node may be determined according to the final aggregation result. After traversing all nodes of the first subgraph, the feature vectors of the first round of iterative process of all nodes are obtained.

In the second round of iteration, according to the feature vector of the first round of iteration, repeat the above steps to obtain the feature vector of the second round of iteration, and so on, until the feature vector of the last round of iteration is obtained. After the number of iterations meets the requirements, the eigenvectors of the last round of iterative process can be used to calculate the prediction results corresponding to the nodes.

Optionally, for any node, the feature vector of the last round of iterative process of the node can be input to the predictor, or input to the Sigmoid function to obtain the corresponding prediction result.

In practical applications, the first participant and the second participant can respectively construct the first subgraph and the second subgraph in the graph data, and the nodes in the first subgraph and the nodes in the second subgraph can have a connection relationship . The first participant can process the node in the first subgraph, calculate the aggregation result of the neighbor nodes of the node in the first subgraph, and obtain the neighbor nodes of the node in the second subgraph from the second participant , and determine the final aggregation result according to the aggregation results corresponding to each subgraph.

Similarly, the second participant can also use a similar method to process the second subgraph, traverse the nodes in the second subgraph, calculate the aggregation result of the node's neighbor nodes in the second subgraph, and at the same time from the first The participant obtains the aggregation result of the neighbor nodes of the node in the first subgraph, and determines the final aggregation result according to the aggregation results corresponding to each subgraph. Through the above-mentioned operations of the first participant and the second participant, the results corresponding to all the nodes in the graph data can be obtained, and the processing of the graph data can be completed without the original data of the nodes leaving the local area.

Optionally, the number of second participants in the embodiment of the present application may be multiple, and correspondingly, the number of second subgraphs may also be multiple, and each second participant may The neighbor node of the arrived node in the corresponding second subgraph calculates the corresponding second aggregation result and sends it to the first participant, and the first participant processes it according to the first aggregation result and multiple second aggregation results to obtain The final aggregation result.

The graph data processing method provided in this embodiment can process graph data including a first subgraph and a second subgraph, the first subgraph includes nodes belonging to the first participant, and the second subgraph Including nodes belonging to the second participant, in any iteration process, the first participant can traverse the nodes in the first subgraph, and for each node traversed, search for the nodes in the first subgraph The neighbor nodes in the sub-graph are aggregated according to the eigenvectors of the last iteration process of the neighbor nodes to obtain the first aggregation result. If the node has a connection relationship with the nodes in the second sub-graph, then Determine the final aggregated result of the node according to the first aggregated result and the second aggregated result, wherein the second aggregated result is based on the neighbors of the node in the second subgraph by the second participant The eigenvector of the previous round of iteration process of the node is determined, and according to the final aggregation result, the eigenvector of the current round of iteration process of the node is determined, and after the number of iterations meets the requirements, the eigenvector of the last round of iteration process is used for Calculate the prediction results corresponding to the nodes, because in the process of analyzing the nodes, the aggregation results of the first participant on the neighbor nodes of the first subgraph and the aggregation results of the second participant on the neighbor nodes of the second subgraph are integrated, In this way, the characteristics of nodes can be extracted more comprehensively and accurately, and in the case of barriers to data interoperability, data from all parties can be comprehensively utilized to jointly realize the processing of graph data and effectively improve the prediction accuracy of graph data.

On the basis of the technical solutions provided in the foregoing embodiments, optionally, a SecureAggregate (secure aggregation) function may be used to perform a secure aggregation operation on the first aggregation result and the second aggregation result to obtain a final aggregation result. The implementation principle of the SecureAggregate function is described below.

Specifically, if the node has a connection relationship with the nodes in the second subgraph, determining the final aggregation result of the node according to the first aggregation result and the second aggregation result may include: if the node has a connection relationship with a node in the second subgraph, then send request information to the second participant, where the request information is used to request the second participant to calculate the second aggregation result corresponding to the node and Encrypting the second aggregation result; receiving the encrypted second aggregation result sent by the second participant; and determining the final aggregation result of the node according to the encrypted second aggregation result.

Wherein, when a node having a connection relationship with a node in the second subgraph is traversed in the first subgraph, the request information is sent to the second subgraph to request the second participant to process it, which can save unnecessary The amount of calculation can effectively improve the efficiency of graph data processing. In addition, the encrypted aggregation result can be transmitted between the first participant and the second participant, which effectively improves the security of data transmission.

In an optional implementation manner, after obtaining the encrypted second aggregation result sent by the second participant, the first participant may perform processing according to the encrypted second aggregation result to obtain a final aggregation result.

In another optional implementation manner, a third party may also be introduced to process the aggregation result.

Optionally, the encrypted second aggregation result is a second aggregation result encrypted with a public key. According to the encrypted second aggregation result, determining the final aggregation result of the node may include: using the public key to encrypt the first aggregation result; based on the random mask, encrypting the encrypted first aggregation result The result is calculated with the encrypted second aggregated result to obtain the encrypted final aggregated result; the encrypted final aggregated result is sent to the third participant, so that the third participant uses the private key to pair The encrypted final aggregated result is decrypted; the decrypted result sent by the third participant is received, and a random masking operation is performed on the decrypted result to obtain the final aggregated result.

FIG. 4 is a system architecture diagram of graph data processing provided by an embodiment of the present application. As shown in Fig. 4, the first participant processes the first subgraph, the second participant processes the second subgraph, and introduces the third participant as a collaborator. The third party holds the keys, which may include public keys and private keys. Optionally, the calculated results can be encrypted using homomorphic encryption. The public key used for encryption is sent by the third party to the first party and the second party, and the private key is held by the third party alone. Have.

When processing the nodes in the first subgraph, the first participant uses the public key to encrypt the first aggregation result, the second participant uses the public key to encrypt the second aggregation result, and the second participant encrypts the encrypted The second aggregation result of is sent to the first participant, and the first participant calculates the sum of the two and sends it to the third participant after adding a random mask. After the third participant receives the data from the first participant, it decrypts it with its own private key, and sends it to the first participant for de-random masking to obtain the final aggregation result of the node.

Similarly, when processing the nodes in the second subgraph, the first participant uses the public key to encrypt the first aggregation result, the second participant uses the public key to encrypt the second aggregation result, and the first participant The encrypted first aggregation result is sent to the second participant, and the second participant calculates the sum of the two and sends it to the third participant after adding a random mask. After the third participant receives the data from the second participant, it decrypts it with its own private key, and sends the decrypted result to the second participant, and the second participant removes the random mask from the result and obtains The final aggregation result of the node.

By introducing an independent third party to assist in the encryption and decryption of aggregation results, it can effectively improve the security of graph data processing and reduce the risk of data leakage.

FIG. 5 is a schematic flowchart of another graph data processing method provided by the embodiment of the present application. This embodiment provides a specific implementation scheme for joint processing of graph data by the first participant and the second participant. As shown in Figure 5, the method may include:

Step 501. Construct graph data according to the user accounts and transfer records of the first participant and the second participant.

Wherein, the graph data includes a first subgraph and a second subgraph, which are respectively constructed by the first participant and the second participant and initialize the nodes.

Optionally, the first participant may construct and initialize the first subgraph through the following methods: construct nodes in the first subgraph according to user accounts belonging to the first participant; The transfer record of the user account of the participant constructs the connection relationship of the nodes in the first sub-graph, and the connection relationship is used to determine the neighbor nodes.

For example, the first participant can be a bank, the user account can be a bank card number, each node in the first sub-graph represents a bank card number, and the transfer relationship between bank card numbers can be used to form a connection relationship between nodes, for example , there is a transfer record between bank card number A and bank card number B, then there may be a connection line between the node corresponding to bank card number A and the node corresponding to bank card number B. In subsequent processing, two directly connected nodes can be neighbor nodes.

The second participant can also use a similar method to construct the second subgraph.

It should be noted that since the bank card number of the first participant may have a transfer relationship with the bank card number of the second participant, there may be at least some nodes in the first subgraph that are connected to nodes in the second subgraph , the first participant and the second participant store corresponding transfer records, so they can know the connection relationship between their own nodes and nodes of other participants, and these connection relationships can be used for aggregation operations in subsequent steps.

Through the above method, graph data can be constructed based on user accounts and transfer relationships, so that user accounts can be analyzed based on the graph data, and the efficiency of monitoring user accounts can be improved.

Step 502: Initialize all nodes in the graph data to obtain initial feature vectors of the nodes.

Wherein, for any node, an initialization operation may be performed according to its corresponding attribute information. Specifically, the first participant may determine the initial feature vector corresponding to the node in the first sub-graph according to the attribute information of the user account of the first participant; similarly, the second participant may determine the initial feature vector according to the second The attribute information of the user account of the participant determines the initial feature vector corresponding to the node in the second subgraph. Wherein, the eigenvector of the last iterative process used in the first iterative process is the initial eigenvector.

The attribute information may include any information used to characterize the attributes of the user account, such as but not limited to: the user's region, age, gender, education, occupation, income, card opening time, card balance, etc.

Optionally, the initialization operation may refer to performing an assignment operation according to attribute information. A simple example would be 000 for ages between 21 and 30 and 001 for ages 31-40. After the assignment of each item of attribute information is completed, the corresponding initial feature vector is obtained.

Constructing the initial feature vector through the attribute information of the user account can quickly and effectively sort out the user's characteristics and apply it to the subsequent iterative process, so as to comprehensively analyze and process it based on the attribute information of the user account and the transfer relationship, and improve the business. The predictive effect of risk information.

Step 503, set the number of iterations k=1.

Step 504, in the kth iteration process, traverse the nodes in the graph data, for each node, the first participant calculates the first aggregation result of the node in the first sub-graph, and the second participant calculates the For the second aggregation result of the node in the second subgraph, the first participant or the second participant determines the feature vector of the k-th round of the node according to the first aggregation result and the second aggregation result.

Specifically, each node in the graph data may be traversed, and for each traversed node, the following steps a to d may be performed.

Step a. Aggregate the neighbor nodes of the currently traversed node in the first subgraph through an Aggregate (aggregation) function to obtain a first aggregation result.

Specifically, step a may be performed by the first participant. The first participant may search for the neighbor nodes of the node in the first subgraph, perform an aggregation operation according to the feature vectors of the last iteration of the neighbor nodes, and obtain a first aggregation result.

When k=1, the eigenvector of the last iteration process is the initial eigenvector, that is, the first aggregation result of the first round of the current node can be calculated based on the initial eigenvectors of the neighbor nodes of the current node.

Optionally, searching for the neighbor nodes of the node in the first subgraph, and performing an aggregation operation according to the feature vectors of the last iteration process of the neighbor nodes may include: searching for the nodes in the first subgraph All neighbor nodes in the subgraph; based on the sampling operation with replacement, select a preset number of neighbor nodes from the found neighbor nodes; calculate the The first aggregation result.

Among them, the sampling operation with replacement means that after selecting any neighbor node from the set of all neighbor nodes, put the selected neighbor node back into the set, and continue sampling until the sampling times meet the requirements, that is, any Neighbor nodes may be drawn one or more times.

For example, the neighbor nodes of node v include node a, node b and node c, the preset number is 3, then after the sampling operation with replacement, the final selected neighbor nodes may be node a, node b, Node a, that is, node a is selected twice, based on the selected three neighbor nodes (two of which are the same), the first aggregation result corresponding to node v can be calculated.

Wherein, the first aggregation result may be calculated through an Aggregate function. The Aggregate function can be designed according to actual needs, for example, it can be a mean function, that is, the feature vectors of the last iteration process of the selected neighbor nodes are averaged, and used as the first aggregation result of the node.

By performing a sampling operation with replacement in all neighbor nodes, a preset number of neighbor nodes can be quickly found and aggregated, which improves the efficiency of the aggregated operation and realizes a standardized aggregated operation.

Further, when an aggregation operation needs to be performed on nodes in the second subgraph, the second participant can send request information to the first participant, and the first participant can determine the nodes in the second subgraph according to the request information The neighbor node in the first subgraph, and the aggregation result of the node determined by the above method is sent to the second participant. Due to the sampling operation with replacement, the second participant cannot understand each node in the first subgraph. The feature vectors corresponding to the neighbor nodes can reduce the risk of data leakage of the first participant and effectively improve the security of graph data processing.

Step b. Aggregate the neighbor nodes of the node in the second subgraph through an Aggregate function to obtain a second aggregation result.

Specifically, step b may be performed by the second participant. For the specific implementation principle and process, please refer to step a.

Optionally, if a node has no neighbor nodes in the first subgraph or the second subgraph, the corresponding participant may not calculate its corresponding aggregation result, or consider it to be in the first subgraph or the second subgraph The aggregation result in is 0.

Step c. Perform secure aggregation on the first aggregation result and the second aggregation result of the nodes to obtain the final aggregation result of the current round.

Specifically, the SecureAggregate function may be used to perform a secure aggregation operation, and the specific implementation scheme may refer to the foregoing embodiments, which will not be repeated here.

In this embodiment, by distributing the Aggregate function to different participants and aggregating the results of all parties through the SecureAggregate function, the function of comprehensively utilizing data from all parties to realize graph data processing is achieved.

Step d. Determine the feature vector of the current round of the node according to the final aggregation result of the current round of the node.

Optionally, for each node, the feature vector of the current round of the node may be determined according to the final aggregation result of the current round of the node and the feature vector of the previous round of the node.

It can be understood that, in the iteration process of the kth round, the current round may refer to the kth round, the previous round may refer to the k-1th round, and the previous round of the first round may refer to the initialization phase.

Specifically, in the iterative process of the k-th round, for each node, the final aggregation result of the k-th round of the node and the feature vector of the k-1th round of the node can be CONCATed, and the obtained result is the same as The model parameters are multiplied, and the multiplied result is added with a nonlinear component through the σ function to obtain the final result, and the feature vector of the current round is determined according to the final result.

Optionally, the model parameters of each round may be different, and during the iteration process of the kth round, the feature vector may be calculated using the model parameters corresponding to the kth round.

Wherein, the model parameters may be model parameters obtained after training. Specifically, they may be obtained after a single participant uses a training sample set to train the model, or may be obtained after joint training of the model by multiple participants. The model can refer to GraphSAGE or other graph neural network models.

Optionally, after obtaining the final results of all nodes, the final results of each node can be normalized or regularized to obtain the feature vector of the current round of each node, thereby completing the iteration of the kth round operate.

Step 505, judging whether k is equal to K. If not, execute step 506; if yes, execute step 507.

Wherein, K is the number of iterations required, which can be set according to actual needs.

Step 506, after increasing the value of k by 1, re-execute step 504.

Step 507 , according to the eigenvector of the last round of iterative process of each node in the graph data, determine the corresponding prediction result.

Wherein, the feature vector used in step 507 may be the final feature vector obtained after the last iterative process is completed, that is, the feature vector of the Kth round of the node. For any node, the feature vector of the K-th round of the node can be input to the predictor function, or input to the sigmoid function to obtain the corresponding prediction result.

In practical applications, the graph data can be constructed first. After the construction is completed, for any node v, according to its corresponding attribute information

, to initialize it,

Indicates the value of node v after initialization. In the k-th iteration process, for each node v in the graph data, find the neighbor node u of node v, according to the feature vector of the k-1th round of neighbor node u

, calculate the final aggregation result of node v in the k-th round iteration process, and then according to the final aggregation result, the feature vector of the k-1th round

And model parameters, etc., to get the eigenvector of the kth iteration process of node v

. In this way, the feature vector of round k can be determined according to the feature vector of round k-1, and the feature vector of round k is used to calculate the feature vector of round k+1. After the number of iterations reaches the preset number K, according to the final feature vector

, to determine the prediction result.

Optionally, for the first participant, after the number of iterations meets the preset requirements, according to the eigenvector corresponding to any node in the first subgraph in the last iteration process, determine the corresponding Business risk information of the user account. Similarly, the second participant may determine the business risk information of the corresponding user account according to the nodes in the second subgraph.

In an optional implementation solution, determining the business risk information of the user account may be specifically determining whether the user account is an abnormal account. If the user account is determined to be an abnormal account according to the feature vector, it indicates that the user account may have illegal Behavior needs to be reported for processing.

In another optional implementation scheme, determining the business risk information of the user account can specifically determine whether the user account has an overdue risk. If it is determined that the user account has an overdue risk based on the feature vector, it needs to be closely monitored, or the user account must be adjusted. credit rating.

The type of business risk information that is finally predicted can be determined through a training process. For example, if the training process uses whether the account is abnormal as a label, the final model can be used to predict whether the account is abnormal; if the training process uses whether the account is overdue as the label, the final model can be used to predict whether the account will be overdue.

The graph data processing method provided in this embodiment can construct the nodes of the graph data according to the user account, and construct the connection relationship of the nodes in the graph data according to the transfer record of the user account, and the connection relationship is used to determine the neighbor nodes, according to The attribute information of the user account constructs the initial feature vector corresponding to the node in the graph data, and finally after multiple iterations, the obtained feature vector can be used to predict the business risk information of the user account, which can integrate the information of different participants. The information of user accounts together completes the prediction of business risk information, improves the accuracy of prediction of business risks, screens out abnormal user accounts in time, and effectively realizes the monitoring of user accounts.

FIG. 6 is a schematic structural diagram of an apparatus for processing graph data provided by an embodiment of the present application. The graph data includes a first subgraph comprising nodes belonging to a first party and a second subgraph comprising nodes belonging to a second party. The apparatus is applicable to a first party. As shown in Figure 6, the processing means for the graph data may include:

An execution module 601, configured to traverse the nodes in the first subgraph during any iteration, and perform the following operations for each traversed node;

A search module 602, configured to search for neighbor nodes of the node in the first subgraph, perform an aggregation operation according to the feature vectors of the last round of iteration process of the neighbor nodes, and obtain a first aggregation result;

An aggregation module 603, configured to determine the final aggregation result of the node according to the first aggregation result and the second aggregation result when the node has a connection relationship with the nodes in the second sub-graph; wherein, the The second aggregation result is determined by the second participant according to the eigenvectors of the node's neighbor nodes in the second subgraph in the last round of iterative process;

The determination module 604 is configured to determine the eigenvector of the current round of iterative process of the node according to the final aggregation result; after the number of iterations meets the requirements, the eigenvector of the last round of iterative process is used to calculate the prediction result corresponding to the node.

Wherein, the execution module 601 can traverse the nodes in the first sub-graph in any iteration process, and for each node traversed, the search module 602, the aggregation module 603, and the determination module 604 can be used to calculate its The corresponding eigenvectors.

Optionally, the aggregation module 603 is specifically configured to:

Optionally, the encrypted second aggregation result is a second aggregation result encrypted with a public key; the aggregation module 603 determines the final aggregation result of the node according to the encrypted second aggregation result As a result, specifically for:

encrypting the first aggregation result using the public key;

Optionally, the search module 602 is specifically configured to:

Find all neighbor nodes of the node in the first subgraph;

Optionally, the executing module 601 is further configured to:

Correspondingly, after the number of iterations meets the preset requirement, the executing module 601 is further configured to:

Optionally, the executing module 601 is further configured to:

Optionally, when the execution module 601 determines the business risk information of the user account corresponding to the node, it is specifically configured to:

Optionally, when the aggregation module 603 determines the final aggregation result of the node according to the first aggregation result and the second aggregation result, it is specifically used to:

The image data processing device provided by any of the foregoing embodiments is used to implement the technical solution of any of the foregoing method embodiments, and its implementation principles and technical effects are similar, so details are not repeated here.

FIG. 7 is a schematic structural diagram of a graph data processing device provided by an embodiment of the present application. As shown in FIG. 7 , the device may include: a memory 701, a processor 702, and a processing program of graph data stored on the memory 701 and operable on the processor 702, the processing program of the graph data When executed by the processor 702, the steps of the method for processing graph data as described in any of the foregoing embodiments are implemented.

Optionally, the memory 701 can be independent or integrated with the processor 702 .

For the implementation principles and technical effects of the device provided in this embodiment, reference may be made to the foregoing embodiments, and details are not repeated here.

The embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium stores a processing program for image data. The steps of the processing method of the graph data described above.

An embodiment of the present application further provides a computer program product, including a computer program, and when the computer program is executed by a processor, the method described in any of the preceding embodiments is implemented.

In the several embodiments provided in this application, it should be understood that the disclosed devices and methods may be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods, for example, multiple modules can be combined or integrated. to another system, or some features may be ignored, or not implemented.

The above-mentioned integrated modules implemented in the form of software function modules may be stored in a computer-readable storage medium. The above-mentioned software function modules are stored in a storage medium, and include several instructions to make a computer device (which may be a personal computer, server, or network device, etc.) or a processor execute some steps of the methods described in various embodiments of the present application.

It should be understood that the above-mentioned processor may be a central processing unit (Central Processing Unit, referred to as CPU), and may also be other general-purpose processors, digital signal processors (Digital Signal Processor (DSP for short), Application Specific Integrated Circuit (ASIC for short), etc. A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like. The steps of the method disclosed in conjunction with the invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor.

The storage may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk storage, and may also be a U disk, a mobile hard disk, a read-only memory, a magnetic disk, or an optical disk.

The above-mentioned storage medium can be realized by any type of volatile or non-volatile storage device or their combination, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable In addition to programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. A storage media may be any available media that can be accessed by a general purpose or special purpose computer.

An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. Of course, the storage medium may also be a component of the processor. Processors and storage media can be located in application-specific integrated circuits (Application Specific Integrated Circuits, referred to as ASIC). Of course, the processor and the storage medium can also exist in the electronic device or the main control device as discrete components.

It should be noted that, in this document, the term "comprising", "comprising" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or apparatus comprising a set of elements includes not only those elements, It also includes other elements not expressly listed, or elements inherent in the process, method, article, or device. Without further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus comprising that element.

The serial numbers of the above embodiments of the present application are for description only, and do not represent the advantages and disadvantages of the embodiments.

Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware, but in many cases the former is better implementation. Based on this understanding, the essence of the technical solution of this application or the part that contributes to the prior art can be embodied in the form of software products, and the computer software products are stored in a storage medium (such as ROM/RAM, disk, CD-ROM), including several instructions to enable a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present application.

The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structures or equivalent process transformations made by using the description of the application and the accompanying drawings are directly or indirectly used in other related technical fields. , are all included in the patent protection scope of the present application in the same way.

Claims

A method for processing graph data, wherein the graph data includes a first subgraph and a second subgraph, the first subgraph includes nodes belonging to a first participant, and the second subgraph includes nodes belonging to A node of a second party; the method applied to the first party, the method comprising:

During any round of iteration, traverse the nodes in the first subgraph, and perform the following operations for each traversed node:

Finding the neighbor nodes of the node in the first subgraph, and performing an aggregation operation according to the feature vector of the last round of iteration process of the neighbor nodes, to obtain a first aggregation result;

If the node has a connection relationship with the nodes in the second subgraph, then determine the final aggregation result of the node according to the first aggregation result and the second aggregation result; wherein the second aggregation result is the The second participant is determined according to the eigenvectors of the last round of iterative process of the neighbor nodes of the node in the second subgraph;

According to the final aggregation result, the eigenvector of the current round of iterative process of the node is determined; after the number of iterations meets the requirement, the eigenvector of the last round of iterative process is used to calculate the prediction result corresponding to the node.
The method according to claim 1, wherein if the node has a connection relationship with the nodes in the second sub-graph, then determine the final result of the node according to the first aggregation result and the second aggregation result Aggregated results, including:

If the node has a connection relationship with a node in the second subgraph, send request information to the second participant, where the request information is used to request the second participant to calculate the first node corresponding to the node two aggregation results and encrypting said second aggregation result;

receiving the encrypted second aggregation result sent by the second participant;

Determine the final aggregation result of the node according to the encrypted second aggregation result.
The method according to claim 2, wherein the encrypted second aggregation result is a second aggregation result encrypted with a public key; according to the encrypted second aggregation result, the node is determined The final aggregation results of , including:

encrypting the first aggregation result using the public key;

Based on the random mask, calculating the encrypted first aggregation result and the encrypted second aggregation result to obtain the encrypted final aggregation result;

sending the encrypted final aggregation result to a third party, so that the third party uses a private key to decrypt the encrypted final aggregation result;

Receive the decryption result sent by the third participant, and perform a random masking operation on the decryption result to obtain the final aggregation result.
The method according to any one of claims 1-3, characterized in that, searching for the neighbor nodes of the node in the first subgraph, and performing aggregation according to the feature vector of the last round of iterative process of the neighbor nodes operations, including:

Find all neighbor nodes of the node in the first subgraph;

Based on the sampling operation with replacement, select a preset number of neighbor nodes from the found neighbor nodes;

The first aggregation result is calculated according to the eigenvectors of the last round of iterative process of the selected neighbor nodes.
The method according to any one of claims 1-4, further comprising:

constructing nodes in the first subgraph according to user accounts belonging to the first participant;

According to the transfer record of the user account of the first participant, construct the connection relationship of the nodes in the first subgraph, and the connection relationship is used to determine the neighbor nodes;

Correspondingly, after the number of iterations meets the preset requirements, the method further includes:

According to the feature vector corresponding to any node in the first subgraph in the last iteration process, determine the service risk information of the user account corresponding to the node.
The method according to claim 5, further comprising:

determining an initial feature vector corresponding to a node in the first subgraph according to the attribute information of the user account of the first participant;

Wherein, the eigenvector of the last iterative process used in the first iterative process is the initial eigenvector.
The method according to claim 5 or 6, wherein determining the business risk information of the user account corresponding to the node comprises:

Determine whether the user account is an abnormal account, and if it is determined according to the feature vector that the user account belongs to an abnormal account, perform reporting processing; or,

Determine whether the user account has an overdue risk, and if it is determined according to the feature vector that the user account has an overdue risk, monitor the user account, or adjust the credit level of the user account.
The method according to any one of claims 1-7, wherein determining the final aggregation result of the node according to the first aggregation result and the second aggregation result includes:

A final aggregation result is determined through a nonlinear algorithm according to the first aggregation result and the second aggregation result.
The method according to any one of claims 1-8, wherein the graph data is used to implement social behavior analysis, the nodes in the graph data are used to represent users, and the preset association relationships include family relationships, employment relationships, The preset association relationship is used to determine neighbor nodes.
A device for processing graph data, characterized in that the graph data includes a first subgraph and a second subgraph, the first subgraph includes nodes belonging to a first participant, and the second subgraph includes nodes belonging to A node of a second participant; the device is applied to the first participant, the device comprising:

An execution module, configured to traverse the nodes in the first subgraph during any iteration, and perform the following operations for each traversed node;

A search module, configured to search for neighbor nodes of the node in the first subgraph, perform an aggregation operation according to the feature vectors of the last round of iteration process of the neighbor nodes, and obtain a first aggregation result;

An aggregation module, configured to determine the final aggregation result of the node according to the first aggregation result and the second aggregation result when the node has a connection relationship with the nodes in the second sub-graph; wherein, the first The two-aggregation result is determined by the second participant according to the eigenvectors of the last iteration process of the neighbor nodes of the node in the second subgraph;

The determination module is configured to determine the eigenvector of the current round of iterative process of the node according to the final aggregation result; after the number of iterations meets the requirement, the eigenvector of the last round of iterative process is used to calculate the prediction result corresponding to the node.
The device according to claim 10, wherein the aggregation module is specifically used for:

If the node has a connection relationship with a node in the second subgraph, send request information to the second participant, where the request information is used to request the second participant to calculate the first node corresponding to the node two aggregation results and encrypting said second aggregation result;

receiving the encrypted second aggregation result sent by the second participant;

Determine the final aggregation result of the node according to the encrypted second aggregation result.
The device according to claim 11, wherein the encrypted second aggregation result is a second aggregation result encrypted with a public key; and the aggregation module is based on the encrypted second aggregation result , when determining the final aggregation result of the node, it is specifically used for:

encrypting the first aggregation result using the public key;

Based on the random mask, calculating the encrypted first aggregation result and the encrypted second aggregation result to obtain the encrypted final aggregation result;

sending the encrypted final aggregation result to a third party, so that the third party uses a private key to decrypt the encrypted final aggregation result;

Receive the decryption result sent by the third participant, and perform a random masking operation on the decryption result to obtain the final aggregation result.
The device according to any one of claims 10-12, wherein the search module is specifically used for:

Find all neighbor nodes of the node in the first subgraph;

Based on the sampling operation with replacement, select a preset number of neighbor nodes from the found neighbor nodes;

The first aggregation result is calculated according to the eigenvectors of the last round of iterative process of the selected neighbor nodes.
The device according to any one of claims 10-13, wherein the execution module is further configured to:

constructing nodes in the first subgraph according to user accounts belonging to the first participant;

According to the transfer record of the user account of the first participant, construct the connection relationship of the nodes in the first subgraph, and the connection relationship is used to determine the neighbor nodes;

Correspondingly, after the number of iterations meets the preset requirements, the execution module is also used to:

According to the eigenvector corresponding to any node in the first subgraph in the last iteration process, determine the business risk information of the user account corresponding to the node.
The device according to claim 14, wherein the executing module is also used for:

determining an initial feature vector corresponding to a node in the first subgraph according to the attribute information of the user account of the first participant;

Wherein, the eigenvector of the last iterative process used in the first iterative process is the initial eigenvector.
The device according to claim 14 or 15, wherein the execution module is specifically configured to: when determining the business risk information of the user account corresponding to the node:

Determine whether the user account is an abnormal account, and if it is determined according to the feature vector that the user account belongs to an abnormal account, perform reporting processing; or,

Determine whether the user account has an overdue risk, and if it is determined according to the feature vector that the user account has an overdue risk, monitor the user account, or adjust the credit level of the user account.
The device according to any one of claims 10-16, wherein, when the aggregation module determines the final aggregation result of the node according to the first aggregation result and the second aggregation result, it is specifically used for:

A final aggregation result is determined through a nonlinear algorithm according to the first aggregation result and the second aggregation result.
A graph data processing device, characterized in that the graph data processing device includes: a memory, a processor, and a graph data processing program stored in the memory and operable on the processor, the When the image data processing program is executed by the processor, the steps of the image data processing method according to any one of claims 1-9 are realized.
A computer-readable storage medium, characterized in that a processing program for image data is stored on the computer-readable storage medium, and when the processing program for image data is executed by a processor, any one of claims 1-9 is implemented. The steps of the graph data processing method described in item.
A computer program product, comprising a computer program, characterized in that, when the computer program is executed by a processor, the method according to any one of claims 1-9 is implemented.