Data processing

Abstract

Disclosed in the present disclosure are a data processing method and apparatus, a storage medium, and an electronic device. The method comprises: acquiring first graph structure data corresponding to a target transaction; determining at least one piece of transaction subgraph structure data corresponding to the first graph structure data; on the basis of the transaction subgraph structure data, determining a first transaction graph processing model; performing edge structure knowledge distillation processing on the first transaction graph processing model to obtain a second transaction graph processing model; acquiring second graph structure data of the target transaction; on the basis of the second transaction graph processing model, performing graph sparsification processing on the second graph structure data to obtain target transaction graph structure data; and, on the basis of the target transaction graph structure data, performing target transaction processing.

Description

Data processing Technical Field

[0001] This disclosure relates to the field of computer technology, and more particularly to data processing. Background Technology

[0002] In practical applications, graph-structured data is widely present in various business domains. Graph-structured data depicts entities and their relationships as a complex network graph (also known as graph-structured data). For example, graph-structured data in financial trading firms records and analyzes tens of thousands of transactions in the financial market, while graph-structured data in social networks connects the intricate personal relationships between billions of users. These graph-structured data in financial trading networks and social networks often contain hundreds of millions of nodes and tens of billions of edges, forming a highly complex network structure. Graph-structured data may contain important information such as user behavior preferences, social influence, and the propagation of financial risks. Summary of the Invention

[0003] This disclosure provides a data processing method, apparatus, storage medium, and electronic device, the technical solutions of which are as follows.

[0004] In a first aspect, this disclosure provides a data processing method, the method comprising: acquiring first graph structure data corresponding to a target transaction, determining at least one transaction subgraph structure data corresponding to the first graph structure data; determining a first transaction graph processing model based on the transaction subgraph structure data, performing edge structure knowledge distillation on the first transaction graph processing model to obtain a second transaction graph processing model, wherein the second transaction graph processing model is a lightweight model relative to the first transaction graph processing model; acquiring second graph structure data of the target transaction, performing graph sparsification on the second graph structure data based on the second transaction graph processing model to obtain target transaction graph structure data; and performing target transaction processing based on the target transaction graph structure data.

[0005] Secondly, this disclosure provides a data processing apparatus, the apparatus comprising: a data acquisition module, configured to acquire first graph structure data corresponding to a target transaction, and determine at least one transaction subgraph structure data corresponding to the first graph structure data; a knowledge distillation module, configured to determine a first transaction graph processing model based on the transaction subgraph structure data, and perform edge structure knowledge distillation processing on the first transaction graph processing model to obtain a second transaction graph processing model, wherein the second transaction graph processing model is a lightweight model relative to the first transaction graph processing model; a graph sparsification module, configured to acquire second graph structure data of the target transaction, and perform graph sparsification processing on the second graph structure data based on the second transaction graph processing model to obtain target transaction graph structure data; and a transaction processing module, configured to perform target transaction processing based on the target transaction graph structure data.

[0006] Thirdly, this disclosure provides a computer storage medium storing at least one instruction adapted for loading by a processor and executing method steps of one or more embodiments of this disclosure.

[0007] Fourthly, this disclosure provides a computer program product storing at least one instruction adapted to be loaded by a processor and to execute method steps of one or more embodiments of this disclosure.

[0008] Fifthly, this disclosure provides an electronic device that may include: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and to execute method steps of one or more embodiments of this disclosure.

[0009] The beneficial effects of the technical solutions provided by some embodiments of this disclosure include at least the following: In one or more embodiments of this disclosure, a first graph structure data corresponding to a target transaction is obtained, transaction subgraph structure data corresponding to the first graph structure data is determined, a first transaction graph processing model is determined based on the transaction subgraph structure data, an edge structure knowledge distillation process is performed on the first transaction graph processing model to obtain a second transaction graph processing model, and graph sparsification is performed on the second graph structure data of the target transaction based on the second transaction graph processing model to obtain the target transaction graph structure data. By combining the expressive power of the first graph structure data of the target transaction from deep learning with the computational efficiency of the lightweight second transaction graph processing model, the contradiction between computational performance and model accuracy in large-scale graph sparsification is resolved. Subsequently, target transaction processing can be performed based on the target transaction graph structure data, showing significant effects in improving model performance, increasing inference speed, and reducing computational resource consumption, which helps to promote technological innovation and commercial application in related transaction scenarios. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in this disclosure or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 is a schematic diagram of a data processing system provided in this disclosure.

[0012] Figure 2 is a flowchart illustrating a data processing method provided in this disclosure;

[0013] Figure 3 is a schematic diagram of a graph structure data processing scenario provided in this disclosure;

[0014] Figure 4 is a schematic diagram of a method for distilling edge structure knowledge provided in this disclosure;

[0015] Figure 5 is a schematic diagram of the processing flow of a first transaction graph processing model provided in this disclosure;

[0016] Figure 6 is a schematic diagram of the processing flow of a second transaction graph processing model provided in this disclosure;

[0017] Figure 7 is a schematic diagram of a graph sparsification process provided in this disclosure;

[0018] Figure 8 is a schematic diagram of the structure of a data processing device provided in this disclosure;

[0019] Figure 9 is a schematic diagram of the structure of an electronic device provided in this disclosure. Detailed Implementation

[0020] The technical solutions of this disclosure will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0021] In the description of this disclosure, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. In the description of this disclosure, it should be noted that, unless otherwise expressly specified and limited, "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices. Those skilled in the art can understand the specific meaning of the above terms in this disclosure based on the specific circumstances. Furthermore, in the description of this disclosure, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist; for example, A and / or B can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship.

[0022] In related technologies, learning and reasoning on graph-structured data is extremely challenging, partly due to its massive scale and complex structure. Furthermore, the requirements for relationship modeling vary significantly across different scenarios, and different applications often need to optimize the way their graph-structured data is constructed for specific tasks. For example, in recommender systems, user interactions might be valuable connections, while in tasks involving identifying fraudulent account behavior, the same connections might seem irrelevant or even disruptive. Moreover, the network graph data of graph-structured data undergoes real-time dynamic updates over time; outdated edges may disappear, and new relationships are constantly being generated. This daily-level dynamic change requires rapid response and the ability to maintain and update graph-structured data in a rapidly changing environment.

[0023] When processing graph-structured data (often referred to as large-scale graphs), sparsification becomes particularly crucial. Graph sparsification can be understood as the process of reducing non-core edges while preserving the core characteristics of the graph data structure. This not only significantly improves the computational efficiency of subsequent tasks but also enhances the grasp of key information by transaction models (such as content recommendation models and credit processing models), thereby alleviating the computational and storage burden brought about by processing large-scale graphs to some extent. However, traditional graph sparsification methods often rely on heuristic rules or predefined thresholds. Such processing methods are not only difficult to adapt to the personalized needs of different transaction scenarios but also difficult to flexibly adapt to dynamic changes in data. How to intelligently identify and remove noisy edges that have a negative impact on specific tasks, and thus obtain a base graph that is both concise and rich in key information, is one of the challenges facing this field.

[0024] The present disclosure will now be described in detail with reference to specific embodiments.

[0025] Please refer to Figure 1, which is a schematic diagram of a data processing system provided in this disclosure. As shown in Figure 1, the data processing system may include at least a client cluster and a service platform 100.

[0026] The client cluster may include at least one client, as shown in Figure 1, specifically including client 1 corresponding to user 1, client 2 corresponding to user 2, ..., client n corresponding to user n, where n is an integer greater than 0.

[0027] Each client in a client cluster can be an electronic device with communication capabilities, including but not limited to: wearable devices, handheld devices, personal computers, tablets, in-vehicle devices, smartphones, computing devices, or other processing devices connected to a wireless modem. Electronic devices may have different names in different networks, such as: user equipment, access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, electronic device, wireless communication device, user agent or user device, cellular phone, cordless phone, personal digital assistant (PDA), and electronic devices in 5G networks or future evolved networks.

[0028] The service platform 100 can be a standalone server device, such as a rack-mount, blade, tower, or cabinet-type server device, or a workstation, mainframe, or other hardware device with strong computing power; or it can be a server cluster composed of multiple servers. The servers in the service cluster can be composed in a symmetrical manner, wherein each server is functionally and hierarchically equivalent in the transaction chain, and each server can provide services independently. The independent provision of services can be understood as not requiring the assistance of other servers.

[0029] It should be noted that the service platform 100 establishes a communication connection with at least one client in the client cluster via a network for interactive communication. This network can be a wireless network or a wired network. Wireless networks include, but are not limited to, cellular networks, wireless LANs, infrared networks, or Bluetooth networks. Wired networks include, but are not limited to, Ethernet, universal serial bus (USB), or controller area networks. In one or more embodiments of the specification, technologies and / or formats including Hyper Text Markup Language (HTML), Extensible Markup Language (XML), etc., are used to represent data exchanged over the network (such as target compressed packets). Furthermore, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), and Internet Protocol Security (IPsec) can be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can be used to replace or supplement the aforementioned data communication technologies.

[0030] The data processing system embodiments provided in this disclosure and the data processing methods described in one or more embodiments belong to the same concept. The execution entity corresponding to the data processing method involved in one or more embodiments of the specification can be the aforementioned service platform 100; the execution entity corresponding to the data processing method involved in one or more embodiments of the specification can also be the electronic device corresponding to the client, specifically determined based on the actual application environment. The implementation process of the data processing system embodiments can be detailed in the following method embodiments, and will not be repeated here.

[0031] Based on the scenario diagram shown in Figure 1, the data processing method provided by one or more embodiments of this disclosure will be described in detail below.

[0032] Please refer to Figure 2, which is a flowchart illustrating a data processing method provided by one or more embodiments of this disclosure. This method can be implemented using a computer program and can run on a data processing device based on the von Neumann architecture. The computer program can be integrated into an application or run as a standalone utility application. The data processing device can be a service platform.

[0033] Specifically, the data processing method includes the following steps.

[0034] S102: Obtain the first graph structure data corresponding to the target transaction, and determine at least one transaction subgraph structure data corresponding to the first graph structure data.

[0035] Understandably, in numerous different types of transaction scenarios, a large amount of structured data and information is collected and accumulated from various data sources. In practical applications, it is necessary to describe this structured data and information. Therefore, in these practical transaction scenarios, it is often necessary to store this large amount of information and even data in the form of graph-structured data in a structured manner.

[0036] Understandably, in practical applications, graph structure construction technology can be used to record graph structure data in the target transaction (such as financial transaction scenarios, shopping transaction scenarios, instant messaging transaction scenarios, content recommendation transaction scenarios, etc.).

[0037] Graph-structured data refers to graphs with a large number of nodes that change frequently. It typically contains complex network data with a large number of nodes and edges, such as a graph with more than 5000 nodes whose number changes every minute. It can represent entities and the relationships between them. Graph-structured data is widely found in many fields, such as social networks, financial transactions, and logistics networks. In some embodiments, graph-structured data can also be called large-scale graph-structured data. Typically, node thresholds or edge thresholds are set. When the number of nodes in the graph-structured data exceeds the node threshold and / or the number of edges exceeds the edge threshold, the graph-structured data can be considered large-scale graph-structured data.

[0038] Components of graph-structured data: Nodes: The basic units in the graph, representing entities. For example, in a social network, a node can represent a user; in finance, a node can represent a company, account, or transaction. Edges: Lines connecting nodes, representing relationships or interactions between them. For example, in a social network, an edge can represent a friend relationship between users; in financial data, an edge can represent a transaction or the flow of funds.

[0039] The first graph structure data can be historical graph data or original graph data of the service platform on the target transaction. The second transaction graph processing model is obtained by using the historical data or original graph data and performing edge structure knowledge distillation with the help of the first transaction graph processing model.

[0040] Considering that the first graph structure data is usually huge, when constructing the first transaction graph processing model of the target transaction, one or more transaction subgraph structure data are extracted, that is, at least one transaction subgraph structure data corresponding to the first graph structure data is determined; optionally, graph data sampling can be used to sample the first graph structure data into subgraphs, and the original first graph structure data will be sampled into small transaction subgraph structure data.

[0041] S104: Determine a first transaction graph processing model based on the transaction subgraph structure data, and perform edge structure knowledge distillation on the first transaction graph processing model to obtain a second transaction graph processing model. The second transaction graph processing model is a lightweight model relative to the first transaction graph processing model.

[0042] Schematic, transaction subgraph structure data is input into a pre-constructed interpretable first transaction graph processing model. This first transaction graph processing model is typically a graph neural network model. Under different transactions, this disclosure selects corresponding first graph structure data for different transactions and inputs it into the first transaction graph processing model for transaction graph processing training. After completing the transaction graph processing training task, the first transaction graph processing model possesses edge structure knowledge for processing graph structure data under the target transaction. Then, edge structure knowledge distillation is performed on the first transaction graph processing model to obtain a second transaction graph processing model; wherein, the second transaction graph processing model is a lightweight model relative to the first transaction graph processing model.

[0043] In this disclosure, the lightweight model can be understood as comparing the second transaction graph processing model with the first transaction graph processing model, wherein the second transaction graph processing model is a lightweight model relative to the first transaction graph processing model; in terms of processing resource consumption, the processing resource consumption of the second transaction graph processing model is less than that of the first transaction graph processing model; in terms of computational complexity, the computational complexity of the second transaction graph processing model is less than that of the first transaction graph processing model; in terms of model architecture, the model architecture complexity of the second transaction graph processing model is less than that of the first transaction graph processing model; and in terms of storage space, the storage space of the second transaction graph processing model is less than that of the first transaction graph processing model.

[0044] In this disclosure, (edge ​​structure) knowledge distillation is the transfer of knowledge from a large, complex model (referred to as the first transaction graph processing model) to a smaller, simpler model (the second transaction graph processing model). The purpose of knowledge distillation is typically to obtain a model that is still powerful in performance but more efficient.

[0045] In one feasible implementation, the method may involve parsing the edge structure information using a first transaction graph processing model based on the transaction subgraph structure data to obtain the edge structure importance information for the target transaction, and then performing edge structure knowledge distillation processing on the first transaction graph processing model based on the edge structure importance information to obtain a second transaction graph processing model.

[0046] For example, based on multiple transaction subgraph structure data, a first transaction graph processing model is used to train the edge structure information parsing task. In each round of task training, the edge structure information parsing task is to identify and learn the importance contribution of edges in each transaction subgraph structure data to the target transaction task (such as classification tasks, clustering tasks, recommendation tasks, etc. under the target transaction). Each round of task training will output edge structure importance information for the target transaction, which is the importance contribution of edges in the transaction subgraph structure data to the target transaction task. After the edge structure information parsing task training is completed, the first transaction graph processing model is subjected to edge structure knowledge distillation processing based on the edge structure importance information to obtain the second transaction graph processing model.

[0047] Optionally, the first transaction graph processing model can be a model such as a graph neural network (GNN), a graph convolutional network (GCN), or a graph attention network (GAT).

[0048] Optionally, the second transaction graph processing model, as a lightweight model relative to the first transaction graph processing model, refers to a model that, after knowledge distillation or other optimization techniques, has lower computational complexity and storage requirements, but can still effectively process graph-structured data. It possesses one or more characteristics such as fewer parameters, efficient inference, scalability, and ease of maintenance. The second transaction graph processing model can select a lightweight processing model architecture, such as a lightweight graph neural network or a multilayer perceptron (MLP) model.

[0049] In one feasible implementation, the first transaction graph processing model includes a transaction graph processing model based on a graph neural network, and the second transaction graph processing model includes a transaction graph processing model based on a multi-point perceptron network. Utilizing an interpretable graph neural network, the importance of each subgraph structure data or edge in the graph structure data is learned through training. Knowledge distillation techniques are then employed to effectively transfer knowledge containing edge importance probability information to the computationally less expensive second transaction graph processing model. Filtering is then performed using the second transaction graph processing model.

[0050] S106: Obtain the second graph structure data of the target transaction, and perform graph sparsification processing on the second graph structure data based on the second transaction graph processing model to obtain the target transaction graph structure data.

[0051] For example, the second graph structure data can be the online graph structure data that is updated in real time for the target transaction. Using the second transaction graph processing obtained above, graph sparsification can be performed on the rapidly updated second graph structure data for the target transaction to generate a simplified base graph reflecting its core structure, i.e., the target transaction graph structure data. Based on the previously trained second transaction graph processing model, graph sparsification is performed on this online graph structure data. Graph sparsification aims to reduce non-core edges in the graph structure data while retaining key information, thereby generating a more concise target transaction graph structure data. This step significantly improves the efficiency of subsequent analysis and processing, thus achieving efficient and robust graph processing and inference performance.

[0052] Figure 3 illustrates a scenario for graph structure data processing. In Figure 3, several transaction sub-graph structure data are obtained by sampling the first graph structure data under a target transaction. The first graph structure data can be historical graph data or original graph data of the service platform on the target transaction. The transaction sub-graph structure data is input into a pre-constructed interpretable first transaction graph processing model (a graph neural network GNN as shown in Figure 3). The first transaction graph processing model is typically a graph neural network model. Under different transactions, this disclosure selects the corresponding first graph structure data under different transactions and inputs it into the first transaction graph processing model for transaction graph processing training. After completing the transaction graph processing training task, the first transaction graph processing model possesses edge structure knowledge for processing graph structure data under the target transaction. Then, based on the selected second transaction graph processing model (using a multilayer perceptron (MLP) model as shown in Figure 3), the first transaction graph processing model is used to perform edge structure knowledge distillation to obtain a trained second transaction graph processing model. Based on the previously trained second transaction graph processing model, graph sparsification processing is performed on this online graph structure data. Graph sparsification aims to reduce non-core edges in graph structure data while retaining key information, thereby generating a more concise target transaction graph structure data.

[0053] In this disclosure, by introducing edge structure knowledge distillation into the graph structure data processing scenario, the resulting small model (the second transaction graph processing model) not only carries the core recognition capability of the interpretable first transaction graph processing model, but also significantly accelerates the processing speed of the rapidly updated second graph structure data for target transactions due to its smaller scale. This enables the service platform to respond quickly to data changes even in resource-constrained environments, ensuring the real-time updating and deployment of target transaction models (such as content recommendation, credit scoring, and financial risk control models). This graph sparsification method reduces computational complexity while preserving the capture of key information from the original base graph, which has significant application value in research scenarios such as financial risk control, social network analysis, and even bioinformatics.

[0054] S108: Perform target transaction processing based on the target transaction graph structure data.

[0055] For example, a sparsed target transaction graph structure data is used, in which unnecessary edges have been removed and core relationships have been preserved, making it suitable for subsequent analysis and processing of target transactions.

[0056] Optionally, a target transaction knowledge graph can be constructed using knowledge graph technology based on the target transaction graph structure data.

[0057] Optionally, the target transaction graph structure data can be processed based on specific target transaction tasks. For example, target transaction tasks include, but are not limited to, risk assessment tasks, user behavior analysis tasks, recommendation generation tasks, network propagation analysis tasks, etc.

[0058] For example, in a financial scenario: when processing financial transaction data, potential fraudulent transactions can be identified through the target transaction graph structure data, and risk control measures can be triggered based on the analysis results.

[0059] For example, in a social media scenario: by using the interaction graph between users in the target transaction graph structure data, we can analyze which users have a strong influence and then formulate corresponding promotion strategies.

[0060] In one or more embodiments of this disclosure, a first graph structure data corresponding to a target transaction is obtained, transaction subgraph structure data corresponding to the first graph structure data is determined, a first transaction graph processing model is determined based on the transaction subgraph structure data, edge structure knowledge distillation is performed on the first transaction graph processing model to obtain a second transaction graph processing model, and graph sparsification is performed on the second graph structure data of the target transaction based on the second graph processing model to obtain the target transaction graph structure data. By combining the expressive power of the first graph structure data of the target transaction from deep learning with the computational efficiency of the lightweight second transaction graph processing model, the contradiction between computational performance and model accuracy in large-scale graph sparsification is resolved. Subsequently, target transaction processing can be performed based on the target transaction graph structure data, showing significant effects in improving model performance, increasing inference speed, and reducing computational resource consumption, which helps to promote technological innovation and commercial application in related transaction scenarios.

[0061] Please refer to Figure 4, which is a schematic flowchart of an edge structure knowledge distillation method proposed in one or more embodiments of this disclosure. The method involves parsing edge structure information using a first transaction graph processing model based on the transaction subgraph structure data to obtain edge structure importance information for the target transaction. Then, the first transaction graph processing model is subjected to edge structure knowledge distillation processing based on the edge structure importance information to obtain a second transaction graph processing model. This can be achieved by referring to the following approach.

[0062] S202: Train the first transaction graph processing model at least once using all the transaction subgraph structure data, and use the first transaction graph processing model to perform graph structure parsing processing on the transaction subgraph structure data to obtain edge structure importance information.

[0063] For example, to create a first transaction graph processing model for a target transaction, multiple transaction subgraph structure data can be used to train the first transaction graph processing model on the task of edge structure information parsing. In each round of task training, the edge structure information parsing task is to identify and learn the importance contribution of each edge in the transaction subgraph structure data to the target transaction task (such as classification tasks, clustering tasks, recommendation tasks, etc. under the target transaction). Each round of task training will output edge structure importance information for the target transaction, which is the importance contribution of the edges in the transaction subgraph structure data to the target transaction task. After the edge structure information parsing task training is completed, the first transaction graph processing model adapted to the target transaction is obtained.

[0064] For example, refer to Figure 5, which is a schematic diagram of the processing flow of a first transaction graph processing model, specifically including the following steps.

[0065] S3002: Create the initial first transaction graph processing model.

[0066] Optionally, one or more graph processing models, such as graph neural networks (GNN), graph convolutional networks (GCN), and graph attention networks (GAT), can be selected to create an initial first-transaction graph processing model.

[0067] S3004: Input all the transaction subgraph structure data into the initial first transaction graph processing model for at least one round of model training until the model training is completed and the first transaction graph processing model is obtained.

[0068] For example, all transaction subgraph structure data can be used as sample data, and the edge structure importance information labels corresponding to the transaction subgraph structure data can be pre-labeled using expert-side services.

[0069] The input to the (initial) first transaction graph processing model is the transaction subgraph structure data; the (initial) first transaction graph processing model is used to identify the importance contribution of edges in the transaction subgraph structure data to the target transaction task (such as classification task, clustering task, recommendation task, etc. under the target transaction), and output the predicted edge structure importance information.

[0070] In each round of model training, the predicted edge structure importance information output by the (initial) first transaction graph processing model is determined. Based on the predicted edge structure importance information and the edge structure importance information label, the first loss function is used to calculate the first model loss. The model parameters of the (initial) first transaction graph processing model are adjusted using the first model loss until the (initial) first transaction graph processing model meets the model training termination condition, thus obtaining the first transaction graph processing model.

[0071] Optionally, the conditions for ending model training may include, for example, the first model loss of the loss function being less than or equal to a preset loss function threshold, or the number of iterations reaching a preset threshold. Specific conditions for ending model training can be determined based on actual circumstances and are not specifically limited here.

[0072] Optionally, the first loss function can be one or more loss functions such as cross-entropy loss, hinge loss, and Euclidean distance loss from related techniques.

[0073] S3006: Determine the edge structure importance information for the transaction subgraph structure data based on the first transaction graph processing model.

[0074] In a schematic way, after the model is completed, the edge structure importance information of the transaction subgraph structure data can be extracted from the trained first transaction graph processing model. For example, all transaction subgraph structure data can be re-input into the first transaction graph processing model, and the edge structure importance information of each transaction subgraph structure data can be output.

[0075] S204: Extract at least one edge structure data corresponding to the edge structure importance information from the transaction subgraph structure data.

[0076] Edge structure transaction information describes the importance and contribution of "a certain edge" in the transaction subgraph structure data to the target transaction task (such as classification task, clustering task, recommendation task, etc. under the target transaction), while edge structure data describes the edge structure description information of "a certain edge" corresponding to the edge structure transaction information. The edge structure description information can be understood as the meta-information of the edge.

[0077] In this disclosure, at least one edge structure data corresponding to the edge structure importance information is extracted from the transaction subgraph structure data, thereby constructing the model input data of the second transaction graph processing model in the subsequent knowledge transfer stage, and the edge structure importance information corresponding to the edge architecture data is used as output label data.

[0078] S206: Using the edge structure data and the edge structure importance information, perform edge structure knowledge distillation on the first transaction graph processing model to obtain the second transaction graph processing model.

[0079] Edge structure data serves as the model input data for the (initial) second transaction graph processing model in the knowledge distillation stage, while edge structure importance information serves as the output label data. By using the edge structure data and the edge structure importance information, the first transaction graph processing model can be subjected to edge structure knowledge distillation to obtain the second transaction graph processing model.

[0080] In one or more embodiments of this disclosure, by introducing edge structure knowledge distillation into the graph structure data processing scenario, the resulting small model (the second transaction graph processing model) not only carries the core recognition capability of the interpretable first transaction graph processing model, but also, due to its smaller size, significantly accelerates the processing speed of the second graph structure data that is rapidly updated for the target transaction. This enables the service platform to respond quickly to data changes even in resource-constrained environments, ensuring the real-time updating and deployment of the target transaction model (content recommendation, credit scoring, financial risk control, etc.). This graph sparsification method reduces computational complexity while maintaining the capture of key information from the original base graph, which has significant application value in research scenarios such as financial risk control, social network analysis, and even bioinformatics.

[0081] For example, refer to Figure 6, which is a schematic diagram of the processing flow of a second transaction graph processing model, specifically including the following steps.

[0082] S4002: Create an initial second transaction graph processing model, obtain model edge structure knowledge from the first transaction graph processing model, and transfer the model edge structure knowledge to the initial second transaction graph processing model.

[0083] Optionally, the (initial) second transaction graph processing model, as a lightweight model relative to the first transaction graph processing model, refers to a model that, after knowledge distillation or other optimization techniques, has lower computational complexity and storage requirements, but can still effectively process graph-structured data. It possesses one or more characteristics such as fewer parameters, efficient inference, scalability, and ease of maintenance.

[0084] Optionally, the (initial) second transaction graph processing model can be a lightweight processing model architecture, such as a lightweight graph neural network, a multilayer perceptron (MLP) model, etc.

[0085] For example, model edge structure knowledge can be obtained from the first transaction graph processing model. This knowledge is stored in the model network layer parameters of the first transaction graph processing model. In other words, the model network layer parameters can be extracted and applied to the initial second transaction graph processing model to initialize the model parameters, thereby completing the transfer of model edge structure knowledge. The weights or parameters of the (initial) second transaction graph processing model are initialized using edge structure importance information, enabling it to quickly learn the features of important edges.

[0086] S4004: Label the edge structure data with edge structure importance tags based on the edge structure importance information.

[0087] For example, the edge structure importance information determined by the first transaction graph processing model will be used as the label for the edge structure data, which is also the edge structure importance label.

[0088] S4006: Input the edge structure data into the initial second transaction graph processing model for at least one round of model training. During model training, the initial second transaction graph processing model is used to identify the predicted edge structure importance information of the edge structure data for the target transaction.

[0089] The initial second transaction graph processing model is used to train the edge structure data. During the forward propagation training process, the initial second transaction graph processing model learns to predict the importance of the target transaction based on the input edge structure data and outputs the predicted edge structure importance information.

[0090] S4008: Determine the target model loss based on the predicted edge structure importance information and the edge structure importance label, and use the target model loss to adjust the model parameters of the initial second transaction graph processing model until the model training is completed, thereby obtaining the second transaction graph processing model.

[0091] The target model loss is calculated using the corresponding loss function, which combines the predicted edge structure importance information with the actual edge structure importance labels. For example, a suitable loss function, such as mean squared error or cross-entropy loss, can be pre-selected to reflect the gap between predictions and actual values.

[0092] The target model loss is used to adjust the model parameters of the initial second transaction graph processing model. The target model loss is used to adjust the parameters of the initial second transaction graph processing model through a backpropagation optimization algorithm (such as Adam or SGD) to reduce the loss and optimize the model performance. The training is iterated and this process is repeated until the loss function converges and the model reaches the expected performance level, thus obtaining the final second transaction graph processing model.

[0093] In one or more embodiments of this disclosure, the training process of the initial second transaction graph processing model is optimized through the transfer and importance labeling of edge structure knowledge. By continuously adjusting the model parameters, a lightweight model that performs better on the target transaction is finally obtained. This process not only improves the model's accuracy but also enhances its sensitivity to edge structure information, thereby better handling the complexity of practical applications.

[0094] Optionally, you can refer to Figure 7, which is a schematic diagram of a graph sparsification process. Specifically, to perform graph sparsification processing on the second graph structure data based on the second transaction graph processing model to obtain the target transaction graph structure data, you can refer to the following method.

[0095] S5002: Based on the second graph structure data, determine the importance information of the target edge structure corresponding to at least one set of edge structures through the second transaction graph processing model.

[0096] After obtaining the second graph structure data of the target transaction, the second graph structure data is input into the second transaction graph processing model. For example, the daily second graph structure data of the target transaction can be input into the second transaction graph processing model. The second graph structure data includes a massive amount of edge structure. The second transaction graph processing model processes the edge structure information parsing task. The structure information parsing task identifies the degree of importance contribution of the edges in the second graph structure data to the target transaction task (such as classification tasks, clustering tasks, recommendation tasks, etc. under the target transaction), thereby generating target edge structure importance information.

[0097] S5004: Based on the importance information of the target edge structure, perform graph structure filtering processing on the second graph structure data to obtain the target transaction graph structure data.

[0098] For example, the following approach can be used.

[0099] B2: Input the second graph structure data into the second transaction graph processing model. The second transaction graph processing model determines the target edge structure in the second graph structure data that does not match the edge structure importance threshold information. The edge structure importance threshold information includes a threshold set for the importance of the edge structure. If the value of the edge structure importance information is greater than the threshold in the edge structure importance threshold information, the edge structure is considered important to the target transaction task. If the value of the edge structure importance information is less than or equal to the threshold in the edge structure importance threshold information, the edge structure is considered unimportant to the target transaction task. Input the second graph structure data into the second transaction graph processing model. The second transaction graph processing model can determine the target edge structure importance information corresponding to each edge structure. Then, it compares the target edge structure importance information with the edge structure importance threshold information. It determines the target edge structure whose value of the edge structure importance information is less than or equal to the threshold in the edge structure importance threshold information. This target edge structure belongs to the edge structure importance threshold information mismatch.

[0100] B4: Determine the target graph structure corresponding to the target edge structure in the second graph structure data, wherein the target graph structure includes target edges and / or target nodes; determine the target graph structure corresponding to the target edge structure in the second graph structure data, wherein the target graph result includes at least one of target edges and target nodes.

[0101] B6: Filter out the target graph structure from the second graph structure data to obtain the target transaction graph structure data.

[0102] In the second graph structure data, all target graph structures are filtered out to retain the core relationships, and graph sparsification is completed to obtain the target transaction graph structure data.

[0103] In this disclosure, graph sparsification is performed using second graph structure data, which yields better results than related techniques such as edge sampling, node sampling, and community detection. Graph complexity is reduced by extracting representative subgraphs, decreasing the number of nodes in the graph, or clustering the graph into tightly connected communities. Related techniques, primarily relying on graph compression, may oversimplify the graph structure, losing important information. They often require manually setting metrics or depend on specific heuristics and lack task relevance, meaning they may not retain the most valuable information for specific application tasks. In contrast, the method in one or more embodiments of this disclosure, instead of relying on simple heuristics, learns the importance of each edge in the graph using interpretable first graph structure data, ensuring that edge selection is closely related to the specific task. Through knowledge distillation, edge importance decisions are passed to the computationally less expensive second graph structure data, resulting in faster processing speeds while retaining edges valuable for specific tasks—efficient and accurate.

[0104] B8: Monitor graph structure update data for the second graph structure data, the graph structure update data including graph structure update edge information and / or graph structure update node information.

[0105] B10: Input the graph structure update data into the second transaction graph processing model, and execute the step of determining the target edge structure that does not match the edge structure importance threshold information in the second graph structure data through the second transaction graph processing model.

[0106] In this disclosure, the second transaction graph processing model supports dynamic graph data updates relative to the second graph structure data. For dynamically changing graphs, the graph structure update data of the target transaction relative to the second graph structure data can be dynamically monitored, and then the graph structure update data can be dynamically input into the second transaction graph processing model. The second transaction graph processing model determines the target edge structure in the second graph structure data that does not match the edge structure importance threshold information, and filters out the target graph structure in the second graph structure data to obtain the target transaction graph structure data.

[0107] The second transaction graph processing model can continuously adapt to newly generated graph structure update data, including edge change update data and node change update data, to ensure that the sparse graph (target transaction graph structure data) always reflects the latest data structure.

[0108] In one or more embodiments disclosed herein, the following technical effects can be achieved.

[0109] 1. Edge importance learning based on the first transaction graph processing model (as shown in the figure neural network GNN): This scheme learns and understands the importance of each subgraph or edge in a large-scale graph through an interpretable model. This method can intelligently identify and maintain relationships that are beneficial to the final task, rather than simply relying on predetermined rules or thresholds.

[0110] 2. Application of knowledge distillation technology in graph sparsity: The knowledge distillation process is innovatively used to transfer the probability information of edge relationship importance judgment in the complex first-transaction graph processing model to the second-transaction graph processing model (such as the multilayer perceptron (MLP) model) with lower computational cost. This reduces the resource consumption and processing time of graph sparsity tasks on large-scale graphs.

[0111] 3. Balance between efficiency and accuracy: Through knowledge distillation, this scheme not only preserves the fine-grained graph structure information learned by the complex first-transaction graph processing model, but also enables smaller models (such as the second-transaction graph processing model) to perform graph sparsification quickly and accurately while significantly reducing the computational burden.

[0112] 4. Adapting to dynamic graph data changes: This solution is particularly suitable for dynamic graph data and can effectively adapt to graph structures that change over time, such as the addition or deletion of nodes and edges. It supports real-time system updates and self-adaptation without the need for frequent retraining.

[0113] 5. Improved inference speed and reduced computational resource consumption: By simplifying the size and complexity of the graph, this approach significantly improves the inference speed of the model while reducing the demand for storage and computational resources.

[0114] 6. Universality and Flexibility: This sparsification method is not only practical in fields such as finance and social networks, but also shows its broad application potential in various other business scenarios such as recommender systems and bioinformatics, enhancing the practicality and universality of the solution.

[0115] The data processing apparatus provided in this disclosure will now be described in detail with reference to FIG8. It should be noted that the data processing apparatus shown in FIG8 is used to execute the methods of the embodiments shown in FIG1 to FIG7 of this disclosure. For ease of explanation, only the parts related to this disclosure are shown. For specific technical details not disclosed, please refer to the embodiments shown in FIG1 to FIG7 of this disclosure.

[0116] Please refer to Figure 8, which shows a schematic diagram of the structure of the data processing device of this disclosure. This data processing device 1 can be implemented as all or part of a device through software, hardware, or a combination of both. According to some embodiments, the data processing device 1 includes a data acquisition module 11, a knowledge distillation module 12, a graph sparsification module 13, and a transaction processing module 14, specifically used for: the data acquisition module 11, for acquiring first graph structure data corresponding to a target transaction, and determining at least one transaction subgraph structure data corresponding to the first graph structure data; the knowledge distillation module 12, for determining a first transaction graph processing model based on the transaction subgraph structure data, and performing edge structure knowledge distillation processing on the first transaction graph processing model to obtain a second transaction graph processing model, wherein the second transaction graph processing model is a lightweight model relative to the first transaction graph processing model; the graph sparsification module 13, for acquiring second graph structure data of the target transaction, and performing graph sparsification processing on the second graph structure data based on the second transaction graph processing model to obtain target transaction graph structure data; and the transaction processing module 14, for performing target transaction processing based on the target transaction graph structure data.

[0117] Optionally, the knowledge distillation module 12 is configured to: parse the edge structure information based on the transaction subgraph structure data using a first transaction graph processing model to obtain edge structure importance information for the target transaction; and perform edge structure knowledge distillation processing on the first transaction graph processing model based on the edge structure importance information to obtain a second transaction graph processing model.

[0118] Optionally, the knowledge distillation module 12 is configured to: train a first transaction graph processing model for at least one round using all the transaction subgraph structure data; obtain edge structure importance information by performing graph structure parsing processing on the transaction subgraph structure data using the first transaction graph processing model; extract at least one edge structure data corresponding to the edge structure importance information from the transaction subgraph structure data; and perform edge structure knowledge distillation processing on the first transaction graph processing model using the edge structure data and the edge structure importance information to obtain a second transaction graph processing model.

[0119] Optionally, the knowledge distillation module 12 is used to: create an initial first transaction graph processing model; input all the transaction subgraph structure data into the initial first transaction graph processing model for at least one round of model training until the model training is completed and the first transaction graph processing model is obtained; and determine the edge structure importance information for the transaction subgraph structure data based on the first transaction graph processing model.

[0120] Optionally, the knowledge distillation module 12 is configured to: create an initial second transaction graph processing model; obtain model edge structure knowledge from the first transaction graph processing model; transfer the model edge structure knowledge to the initial second transaction graph processing model; label the edge structure data with edge structure importance tags based on the edge structure importance information; input the edge structure data into the initial second transaction graph processing model for at least one round of model training; identify the predicted edge structure importance information of the edge structure data for the target transaction through the initial second transaction graph processing model during model training; determine the target model loss based on the predicted edge structure importance information and the edge structure importance tags; adjust the model parameters of the initial second transaction graph processing model using the target model loss until model training is completed, thereby obtaining the second transaction graph processing model.

[0121] Optionally, the first transaction graph processing model includes a transaction graph processing model created based on a graph neural network, and the second transaction graph processing model includes a transaction graph processing model created based on a multi-point perceptron network.

[0122] Optionally, the graph sparsification module 13 is used to: determine the importance information of at least one set of edge structures corresponding to the target edge structures through the second transaction graph processing model based on the second graph structure data, and perform graph structure filtering processing on the second graph structure data based on the importance information of the target edge structures to obtain the target transaction graph structure data.

[0123] Optionally, the graph sparsification module 13 is configured to: input the second graph structure data into the second transaction graph processing model; determine, through the second transaction graph processing model, a target edge structure in the second graph structure data that does not match the edge structure importance threshold information; determine the target graph structure corresponding to the target edge structure in the second graph structure data, the target graph structure including target edges and / or target nodes; and filter out the target graph structure in the second graph structure data to obtain target transaction graph structure data.

[0124] Optionally, the graph sparsification module 13 is used to: monitor graph structure update data for the second graph structure data, the graph structure update data including graph structure update edge information and / or graph structure update node information; input the graph structure update data into the second transaction graph processing model, and execute the step of determining the target edge structure that does not match the edge structure importance threshold information in the second graph structure data through the second transaction graph processing model.

[0125] It should be noted that the data processing apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when executing the data processing method. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the data processing apparatus and the data processing method embodiments provided in the above embodiments belong to the same concept, and the implementation process can be found in the method embodiments, which will not be repeated here.

[0126] The serial numbers in this disclosure are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0127] In one or more embodiments of this disclosure, a first graph structure data corresponding to a target transaction is obtained, transaction subgraph structure data corresponding to the first graph structure data is determined, a first transaction graph processing model is determined based on the transaction subgraph structure data, edge structure knowledge distillation is performed on the first transaction graph processing model to obtain a second transaction graph processing model, and graph sparsification is performed on the second graph structure data of the target transaction based on the second graph processing model to obtain the target transaction graph structure data. By combining the expressive power of the first graph structure data of the target transaction from deep learning with the computational efficiency of the lightweight second transaction graph processing model, the contradiction between computational performance and model accuracy in large-scale graph sparsification is resolved. Subsequently, target transaction processing can be performed based on the target transaction graph structure data, showing significant effects in improving model performance, increasing inference speed, and reducing computational resource consumption, which helps to promote technological innovation and commercial application in related transaction scenarios.

[0128] This disclosure also provides a computer storage medium that can store multiple instructions adapted for loading and execution by a processor of the data processing method described in the embodiments shown in Figures 1 to 7 above. For details of the execution process, please refer to the specific description of the embodiments shown in Figures 1 to 7, which will not be repeated here.

[0129] This disclosure also provides a computer program product that stores at least one instruction, which is loaded by the processor and executed as described in the embodiments shown in Figures 1 to 7 above. For details of the execution process, please refer to the specific description of the embodiments shown in Figures 1 to 7, which will not be repeated here.

[0130] Please refer to Figure 9, which is a structural block diagram of an electronic device provided in an embodiment of this disclosure. The electronic device in this disclosure may include one or more of the following components: a processor 1010, a memory 1020, an input device 1030, an output device 1040, and a bus 1050. The processor 1010, memory 1020, input device 1030, and output device 1040 can be connected to each other via the bus 1050.

[0131] Processor 1010 may include one or more processing cores. Processor 1010 connects to various parts of the electronic device using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 1020, and by calling data stored in memory 1020. Optionally, processor 1010 may be implemented using at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). Processor 1010 may integrate one or more of a central processing unit (CPU), graphics processing unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the displayed content; and the modem handles wireless communication. It is understood that the modem may also not be integrated into processor 1010 and may be implemented separately through a communication chip.

[0132] The memory 1020 may include random access memory (RAM) or read-only memory (ROM). Optionally, the memory 1020 may include non-transitory computer-readable storage medium. The memory 1020 may be used to store instructions, programs, code, code sets, or instruction sets.

[0133] The input device 1030 is used to receive input instructions or data, and includes, but is not limited to, a keyboard, mouse, camera, microphone, or touch device. The output device 1040 is used to output instructions or data, and includes, but is not limited to, a display device and a speaker. In this embodiment, the input device 1030 can be a temperature sensor for acquiring the operating temperature of the electronic device. The output device 1040 can be a speaker for outputting audio signals.

[0134] In addition, those skilled in the art will understand that the structure of the electronic device shown in the above figures does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements. For example, the electronic device may also include radio frequency circuits, input units, sensors, audio circuits, wireless fidelity (WIFI) modules, power supplies, Bluetooth modules, etc., which will not be described in detail here.

[0135] In this embodiment of the disclosure, the entity executing each step can be the electronic device described above. Optionally, the entity executing each step can be the operating system of the electronic device. The operating system can be Android, iOS, or other operating systems, and this embodiment of the disclosure does not limit this.

[0136] In the electronic device of FIG9, the processor 1010 can be used to call a program stored in the memory 1020 and execute it to implement the data processing method as described in the various method embodiments of the present disclosure.

[0137] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory, or random access memory, etc.

[0138] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in the embodiments of this disclosure are all authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the graph structure data involved in this disclosure were obtained under full authorization.

[0139] The above-disclosed embodiments are merely preferred embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Therefore, any equivalent variations made in accordance with the claims of this disclosure shall still fall within the scope of this disclosure.

Claims

1. A data processing method, the method comprising: Obtain the first graph structure data corresponding to the target transaction, and determine at least one transaction subgraph structure data corresponding to the first graph structure data; Based on the transaction subgraph structure data, a first transaction graph processing model is determined, and the first transaction graph processing model is subjected to edge structure knowledge distillation to obtain a second transaction graph processing model. The second transaction graph processing model is a lightweight model relative to the first transaction graph processing model. Obtain the second graph structure data of the target transaction, and perform graph sparsification processing on the second graph structure data based on the second transaction graph processing model to obtain the target transaction graph structure data; Target transaction processing is performed based on the target transaction graph structure data.

2. The method according to claim 1, wherein determining a first transaction graph processing model based on the transaction subgraph structure data, and performing edge structure knowledge distillation on the first transaction graph processing model to obtain a second transaction graph processing model, comprises: Based on the transaction subgraph structure data, the first transaction graph processing model is used to parse the edge structure information to obtain the edge structure importance information for the target transaction; The first transaction graph processing model is subjected to edge structure knowledge distillation processing based on the edge structure importance information to obtain the second transaction graph processing model.

3. The method according to claim 2, wherein the step of parsing edge structure information using a first transaction graph processing model based on the transaction subgraph structure data to obtain edge structure importance information for the target transaction, and performing edge structure knowledge distillation processing on the first transaction graph processing model based on the edge structure importance information to obtain a second transaction graph processing model, includes: The first transaction graph processing model is trained at least once using all the transaction subgraph structure data, and edge structure importance information is obtained by performing graph structure parsing processing on the transaction subgraph structure data through the first transaction graph processing model. Extract at least one edge structure data corresponding to the edge structure importance information from the transaction subgraph structure data; Using the edge structure data and the edge structure importance information, the first transaction graph processing model is subjected to edge structure knowledge distillation to obtain the second transaction graph processing model.

4. The method according to claim 3, wherein the step of training the first transaction graph processing model with at least one round of all the transaction subgraph structure data, and obtaining edge structure importance information by performing graph structure parsing processing on the transaction subgraph structure data through the first transaction graph processing model, includes: Create the initial first transaction graph processing model; All the transaction subgraph structure data are input into the initial first transaction graph processing model for at least one round of model training until the model training is completed and the first transaction graph processing model is obtained. The importance information of the edge structure for the transaction subgraph structure data is determined based on the first transaction graph processing model.

5. The method according to claim 3, wherein the step of using the edge structure data and the edge structure importance information to perform edge structure knowledge distillation on the first transaction graph processing model to obtain the second transaction graph processing model includes: An initial second transaction graph processing model is created, model edge structure knowledge is obtained from the first transaction graph processing model, and the model edge structure knowledge is transferred to the initial second transaction graph processing model. Label the edge structure importance data with edge structure importance tags based on the edge structure importance information; The edge structure data is input into the initial second transaction graph processing model for at least one round of model training. During the model training, the initial second transaction graph processing model is used to identify the predicted edge structure importance information of the edge structure data for the target transaction. Based on the predicted edge structure importance information and the edge structure importance label, the target model loss is determined. The target model loss is then used to adjust the model parameters of the initial second transaction graph processing model until model training is completed, thus obtaining the second transaction graph processing model.

6. The method according to claim 1, wherein the first transaction graph processing model includes a transaction graph processing model created based on a graph neural network, and the second transaction graph processing model includes a transaction graph processing model created based on a multi-point perceptron network.

7. The method according to claim 1, wherein the step of performing graph sparsification processing on the second graph structure data based on the second transaction graph processing model to obtain the target transaction graph structure data includes: Based on the second graph structure data, the importance information of the target edge structure corresponding to at least one set of edge structures is determined by the second transaction graph processing model. Based on the importance information of the target edge structure, graph structure filtering processing is performed on the second graph structure data to obtain the target transaction graph structure data.

8. The method according to claim 7, wherein determining the importance information of at least one set of edge structures corresponding to the target edge structures based on the second graph structure data and through the second transaction graph processing model includes: The second graph structure data is input into the second transaction graph processing model, and the second transaction graph processing model determines the target edge structure in the second graph structure data that does not match the edge structure importance threshold information. Determine the target graph structure corresponding to the target edge structure in the second graph structure data, wherein the target graph structure includes target edges and / or target nodes; The target graph structure is obtained by filtering out the target graph structure from the second graph structure data.

9. The method according to claim 8, further comprising: Monitor graph structure update data for the second graph structure data, the graph structure update data including graph structure update edge information and / or graph structure update node information; The graph structure update data is input into the second transaction graph processing model, and the step of determining the target edge structure that does not match the edge structure importance threshold information in the second graph structure data through the second transaction graph processing model is executed.

10. A data processing apparatus, the apparatus comprising: The data acquisition module is used to acquire the first graph structure data corresponding to the target transaction and determine at least one transaction subgraph structure data corresponding to the first graph structure data. The knowledge distillation module is used to determine a first transaction graph processing model based on the transaction subgraph structure data, and to perform edge structure knowledge distillation on the first transaction graph processing model to obtain a second transaction graph processing model. The second transaction graph processing model is a lightweight model relative to the first transaction graph processing model. The graph sparsification module is used to obtain the second graph structure data of the target transaction, and perform graph sparsification processing on the second graph structure data based on the second transaction graph processing model to obtain the target transaction graph structure data. The transaction processing module is used to perform target transaction processing based on the target transaction graph structure data.

11. A computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the method steps of any one of claims 1 to 9.

12. A computer program product storing at least one instruction, said at least one instruction being loaded by a processor and executing the method steps of any one of claims 1 to 9.

13. An electronic device comprising: A processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and executed the method steps as claimed in any one of claims 1 to 9.

Images