User association analysis method and device based on graph neural network, and computer equipment

Abstract

The invention relates to the technical field of graph neural networks, in particular to a user association analysis method and device based on a graph neural network, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring at least one kind of business data, wherein the types of the business data comprise user attribute data and user interaction data; performing data fusion processing on the user attribute data and the user interaction data, and generating a dynamic relation graph; according to the dynamic relation graph, risk propagation quantitative calculation and weight calculation are carried out by using a graph neural network, and a dynamic weight calculation result is obtained; and adjusting a dynamic relation graph by using the dynamic weight calculation result, and carrying out user association analysis according to the adjusted dynamic relation graph. By adopting the method, the association relationship between the users can be updated in real time in a multi-platform operation mode, the risk conduction effect is effectively captured, and the accuracy of risk identification and credit evaluation is improved.

Description

Technical Field

[0001] The present application relates to the technical field of graph neural networks, and in particular to a method, apparatus, computer device, computer-readable storage medium, and computer program product for user association analysis based on graph neural networks. Background Art

[0002] In the current multi-platform operating model, the supporting role of middle-office technology is becoming increasingly prominent as business continues to grow. The parallel development of multiple specialized platforms requires robust middle-office technology to enable resource sharing, efficiency improvements, and service optimization. A data-based user credit and capability evaluation system is crucial for ensuring the healthy operation of these platforms, and user association analysis plays a crucial role in credit assessment and risk control.

[0003] However, traditional assessment methods primarily rely on data from a single platform, such as transaction flows and user reviews, which presents significant limitations. First, this single data source makes it impossible to integrate multi-dimensional interconnected information, such as supply chain and equity relationships, resulting in an incomplete risk assessment. Second, existing methods are mostly static assessments that focus solely on individual user characteristics, ignoring the risk transmission effects within the interconnected network, making it difficult to identify supply chain risks. Furthermore, user relationships change over time, and traditional models are unable to update network topology in real time, resulting in lagging assessment results.

[0004] Therefore, there is an urgent need for a user association analysis method, device, computer equipment, computer-readable storage medium and computer program product based on graph neural networks, which can update the association relationship between users in real time in a multi-platform operation mode, effectively capture the risk transmission effect, and improve the accuracy of risk identification and credit assessment. Summary of the Invention

[0005] Based on this, it is necessary to provide a user association analysis method, device, computer equipment, computer-readable storage medium and computer program product based on graph neural network, which can update the association relationship between users in real time in a multi-platform operation mode, effectively capture the risk transmission effect, and improve the accuracy of risk identification and credit assessment.

[0006] In a first aspect, the present application provides a user association analysis method based on a graph neural network, comprising:

[0007] Acquire at least one type of business data, where the types of the business data include user attribute data and user interaction data;

[0008] Performing data fusion processing on the user attribute data and the user interaction data, and generating a dynamic relationship graph;

[0009] Based on the dynamic relationship graph, the graph neural network is used to perform risk propagation quantification and weight calculation to obtain dynamic weight calculation results;

[0010] The dynamic weight calculation result is used to adjust the dynamic relationship map, and user association analysis is performed based on the adjusted dynamic relationship map.

[0011] In one embodiment, the graph neural network includes a graph attention network and a temporal convolutional network; the graph neural network is used to perform risk propagation quantification and weight calculation based on the dynamic relationship graph to obtain a dynamic weight calculation result, including:

[0012] Using graph attention network, we generate node embedding vectors based on dynamic relationship graphs;

[0013] Using a temporal graph convolutional network, we obtain the relationship between node features and time according to the node embedding vector, and use the relationship to update the node status.

[0014] Perform risk propagation calculation based on the dynamic relationship graph and the node status, and update the risk value of the risk source node;

[0015] According to the risk value of the risk source node, the dynamic weight calculation result is calculated.

[0016] In one embodiment, adjusting the dynamic relationship graph using the dynamic weight calculation result includes:

[0017] Analyze the user association relationship between each node according to the dynamic weight calculation result;

[0018] When a new association relationship appears in the user association relationship representation, a new association edge is added to the dynamic relationship graph and an initial weight is set.

[0019] In one embodiment, the model of the graph attention network is:

[0020] ;

[0021] Among them, h i (l+1) Represents the feature vector of node i in the l+1 layer; σ represents the activation function, which is a nonlinear function; ∑ represents the summation symbol, which means summing the weighted features of all adjacent nodes; j∈N(i) represents node j in the adjacent node set N(i) of node i; α ij represents the attention coefficient of node i to adjacent node j, which is used to measure the importance of node j to node i; W (l) represents the weight matrix of the lth layer, which is used to map the features of the nodes from the lth layer to the l+1th layer; h j(l)Represents the feature vector of node j in layer l.

[0022] In one embodiment, performing risk propagation calculation based on the dynamic relationship graph and the node status to update the risk value of the risk source node includes:

[0023] Identifying a risk source node in a dynamic relationship graph using a risk propagation model, wherein the risk source node has an initial risk value;

[0024] Calculating the risk propagation intensity in the dynamic relationship graph based on the initial risk value and the node states on the risk propagation path;

[0025] According to the risk propagation intensity in the dynamic relationship graph, the risk values of the risk source nodes within the affected range in the dynamic relationship graph are updated, and the risk source nodes whose risk values exceed the preset threshold are marked as high-risk nodes.

[0026] In one embodiment, the user attribute data includes order data, user evaluation data, after-sales service data, penalty record data, and basic industrial and commercial data, association data, and public opinion monitoring data obtained through a third-party interface;

[0027] The user interaction data includes customer service conversation text, logistics data and user relationship data.

[0028] In a second aspect, the present application also provides a user association analysis device based on a graph neural network, comprising:

[0029] A data acquisition module, configured to acquire at least one type of business data, wherein the types of the business data include user attribute data and user interaction data;

[0030] A data processing module, configured to perform data fusion processing on the user attribute data and the user interaction data, and generate a dynamic relationship graph;

[0031] The dynamic weight calculation module is used to perform risk propagation quantification and weight calculation based on the dynamic relationship graph using the graph neural network to obtain the dynamic weight calculation results;

[0032] The user association analysis module is used to adjust the dynamic relationship map using the dynamic weight calculation result, and perform user association analysis based on the adjusted dynamic relationship map.

[0033] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0034] Acquire at least one type of business data, where the types of the business data include user attribute data and user interaction data;

[0035] Performing data fusion processing on the user attribute data and the user interaction data, and generating a dynamic relationship graph;

[0036] Based on the dynamic relationship graph, the graph neural network is used to perform risk propagation quantification and weight calculation to obtain dynamic weight calculation results;

[0037] The dynamic weight calculation result is used to adjust the dynamic relationship map, and user association analysis is performed based on the adjusted dynamic relationship map.

[0038] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0039] Acquire at least one type of business data, where the types of the business data include user attribute data and user interaction data;

[0040] Performing data fusion processing on the user attribute data and the user interaction data, and generating a dynamic relationship graph;

[0041] Based on the dynamic relationship graph, the graph neural network is used to perform risk propagation quantification and weight calculation to obtain dynamic weight calculation results;

[0042] The dynamic weight calculation result is used to adjust the dynamic relationship map, and user association analysis is performed based on the adjusted dynamic relationship map.

[0043] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the following steps:

[0044] Acquire at least one type of business data, where the types of the business data include user attribute data and user interaction data;

[0045] Performing data fusion processing on the user attribute data and the user interaction data, and generating a dynamic relationship graph;

[0046] Based on the dynamic relationship graph, the graph neural network is used to perform risk propagation quantification and weight calculation to obtain dynamic weight calculation results;

[0047] The dynamic weight calculation result is used to adjust the dynamic relationship map, and user association analysis is performed based on the adjusted dynamic relationship map.

[0048] The above-mentioned user association analysis method, device, computer equipment, computer-readable storage medium and computer program product based on graph neural network can integrate data from different sources and types, including but not limited to transaction flows, user evaluations, supply chain information, equity relations, etc., by obtaining user attribute data and user interaction data from different platforms. This integration of multi-dimensional data makes risk assessment more comprehensive and accurate; the integrated data is processed to generate a dynamic relationship map, which can reflect the association relationships and their changes between users in real time. Compared with traditional static assessment methods, it can more accurately capture the dynamic characteristics of user association networks; using graph neural networks to perform quantitative calculations of risk propagation can effectively capture the transmission effect of risks in user association networks, and can identify and quantify risks such as supply chain risks that are difficult to detect with traditional methods; dynamic weight calculation results are obtained through graph neural networks. These weights reflect the strength and importance of associations between users, making risk assessment and user credit assessment more accurate; the dynamic weight calculation results are used to adjust the dynamic relationship map, and user association analysis is performed based on the adjusted map. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying any creative work.

[0050] Figure 1 This is a diagram of an application environment of a user association analysis method based on a graph neural network in one embodiment;

[0051] Figure 2 1 is a flow chart of a method for analyzing user associations based on a graph neural network in one embodiment;

[0052] Figure 3 Schematic diagram of a flow chart of a user association analysis method based on a graph neural network in another embodiment;

[0053] Figure 4 This is a structural block diagram of a user association analysis device based on a graph neural network in one embodiment;

[0054] Figure 5 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0056] The user association analysis method based on graph neural network provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown, the terminal 102 communicates with the server 104 via a network. The data storage system can store data that the server 104 needs to process. The data storage system can be integrated on the server 104 or placed on the cloud or other network servers.

[0057] The server 104 obtains at least one type of business data through the terminal 102, and the types of business data include user attribute data and user interaction data; the server 104 performs data fusion processing on the user attribute data and the user interaction data, and generates a dynamic relationship graph; based on the dynamic relationship graph, a graph neural network is used to perform risk propagation quantitative calculation and weight calculation to obtain a dynamic weight calculation result; using the dynamic weight calculation result, the dynamic relationship graph is adjusted, and user association analysis is performed based on the adjusted dynamic relationship graph.

[0058] Terminal 102 may include, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices may include smart speakers, smart TVs, smart air conditioners, smart car devices, and projectors. Portable wearable devices may include smart watches, smart bracelets, and head-mounted devices. Head-mounted devices may include virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, and the like. Server 104 may be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing cloud computing services.

[0059] In an exemplary embodiment, Figure 2 As shown in the figure, a user association analysis method based on graph neural network is provided. Figure 1 The server in the example is used to illustrate the method, which includes the following steps S202 to S208.

[0060] Step S202: Acquire at least one type of business data, where the types of business data include user attribute data and user interaction data.

[0061] In one embodiment, user attribute data includes order data, user evaluation data, after-sales service data, penalty record data, as well as basic industrial and commercial data, association data, and public opinion monitoring data obtained through a third-party interface; user interaction data includes customer service conversation text, logistics data, and user relationship data.

[0062] Specifically, order data, user evaluation data, after-sales service data, and penalty record data are usually stored in the enterprise's internal database system and can be directly accessed through the enterprise resource planning (ERP) system, customer relationship management (CRM) system, or a dedicated order management system.

[0063] Basic business data, relationship data, and public opinion monitoring data may need to be obtained through third-party service providers. Third-party services include business information query services, data analysis services, and public opinion monitoring services. Customer service conversation text can be analyzed using NLP technology to extract key indicators such as "service attitude." This text data can come from customer service systems or online chat platforms. For unstructured documents, such as supply chain contracts or credit reports, OCR technology can be used to convert images or scanned documents into editable and searchable data. Logistics data can be obtained through logistics tracking systems, which record the delivery time, location, and other logistics-related information of packages. User relationship data can be obtained from social networks or trading platforms, which record interactions and transaction behaviors between users.

[0064] Step S204: perform data fusion processing on the user attribute data and the user interaction data, and generate a dynamic relationship graph.

[0065] Specifically, the collected user attribute data and user interaction data are cleaned and preprocessed, including noise removal, missing value filling, format standardization, and error correction, to ensure data quality and consistency. The preprocessed user attribute data and user interaction data are then merged, correlated, and transformed to form a unified data view.

[0066] Useful features are extracted from the integrated data view. These features can describe user behavior, preferences, credit status, and other aspects. A dynamic relationship graph is constructed using these extracted features. In a dynamic relationship graph, users and other entities (such as products, services, and other users) are represented as nodes, while interactions and relationships between users are represented as edges. This graph can intuitively display complex relationships and interaction patterns between users.

[0067] Step S206: Based on the dynamic relationship graph, a graph neural network is used to perform risk propagation quantitative calculation and weight calculation to obtain a dynamic weight calculation result.

[0068] Specifically, risk propagation quantification refers to the process of simulating and calculating how risk propagates from one node (e.g., a user or entity) to other nodes in a graph. In fields such as social networks and supply chain management, risk can propagate through the relationships between entities. Graph neural networks are deep learning models specialized for processing graph-structured data. They can capture complex relationships between nodes and perform various tasks such as node classification, link prediction, and graph classification by learning node features and structural information.

[0069] In graph neural networks, weight computation typically refers to learning the weights of edges in a graph. These weights can represent the importance or strength of relationships between nodes. In the context of risk propagation, edge weights can represent the likelihood or impact of a risk spreading from one node to another.

[0070] Dynamic weights mean that these weights are not static, but can be updated in real time based on new data or events. This allows graph neural networks to adapt to changes in the relationships between nodes in the graph, such as new connections, changes in relationship strength, or updates to node attributes.

[0071] During the calculation process, the graph neural network aggregates information from neighboring nodes and updates the feature representation of each node based on the weight of the edge. This process can capture how risks spread in the dynamic relationship graph and the potential risk level of each node.

[0072] Step S208: Using the dynamic weight calculation result, the dynamic relationship graph is adjusted, and user association analysis is performed based on the adjusted dynamic relationship graph.

[0073] Specifically, the dynamic weight calculation results can be used to adjust the edges in the dynamic relationship graph, including adding new edges, deleting old edges, or modifying the weights of edges. This step allows the graph to reflect the latest changes in relationships between users in real time.

[0074] The graph is dynamically adjusted, meaning it updates whenever new data or events occur. Through these adjustments and updates, the dynamic relationship graph can more accurately reflect the actual connections and influences between users, thereby improving the accuracy and reliability of analysis.

[0075] In the above-mentioned user association analysis method based on graph neural network, by obtaining user attribute data and user interaction data from different platforms, it is possible to integrate data from different sources and types, including but not limited to transaction flows, user evaluations, supply chain information, equity relations, etc. This integration of multi-dimensional data makes risk assessment more comprehensive and accurate; the integrated data is processed to generate a dynamic relationship map, which can reflect the association relationships and their changes between users in real time. Compared with traditional static assessment methods, it can more accurately capture the dynamic characteristics of user association networks; using graph neural networks for quantitative calculation of risk propagation can effectively capture the transmission effect of risks in user association networks, and can identify and quantify risks such as supply chain risks that are difficult to detect by traditional methods; dynamic weight calculation results are obtained through graph neural networks. These weights reflect the strength and importance of associations between users, making risk assessment and user credit assessment more accurate; the dynamic weight calculation results are used to adjust the dynamic relationship map, and user association analysis is performed based on the adjusted map.

[0076] In one embodiment, Figure 3 As shown in the figure, the neural network includes a graph attention network and a temporal convolutional network. Based on the dynamic relationship graph, the graph neural network is used to perform risk propagation quantification and weight calculation, and the dynamic weight calculation results are obtained, including:

[0077] Step S302: Generate node embedding vectors based on the dynamic relationship graph using a graph attention network.

[0078] Step S304: Using a temporal graph convolutional network, according to the node embedding vector, obtain the change relationship of the node features over time, and use the change relationship to update the node state;

[0079] Step S306: perform risk propagation calculation based on the dynamic relationship graph and node status, and update the risk value of the risk source node;

[0080] Step S308: Calculate the dynamic weight calculation result according to the risk value of the risk source node.

[0081] Specifically, a graph attention network (GAT) is used to process nodes in the graph and generate an embedding vector for each node. By considering the node's neighbor information and the relative importance of nodes (attention coefficient), GAT generates a comprehensive node embedding vector for each node. This node embedding vector captures the node's local neighborhood information, providing a rich feature representation for subsequent analysis.

[0082] Using a temporal graph convolutional network (TGCN) to analyze changes in node features over time: TGCN captures the temporal dependencies of node states by considering changes in node features and graph structure at different time points. This helps us understand how node features evolve over time and how these changes are affected by the dynamics of the graph structure.

[0083] Calculating risk propagation based on a dynamic relationship graph and node states involves simulating how risk propagates from one node to another in the graph. Using the weights of the associated edges in the graph and the embedding vectors of the nodes, we can calculate the extent to which risk propagates from one node to its neighboring nodes, thereby updating the risk value of each node.

[0084] Based on the risk value of the risk source node, the dynamic weight calculation results are calculated, and the updated risk value is used to adjust the weights of the edges in the graph. These dynamic weights reflect the importance of the relationship between nodes and the potential for risk propagation. The adjustment of weights helps to more accurately identify and evaluate key risk nodes in the network.

[0085] This example combines the strengths of graph attention networks and temporal graph convolutional networks to not only capture local neighborhood information and temporal dynamics of nodes, but also simulate and quantify the propagation of risk within the network. Ultimately, by dynamically adjusting edge weights, risk within the network can be more accurately assessed and managed, providing support for decision-making.

[0086] In one embodiment, the dynamic relationship graph is adjusted using the dynamic weight calculation result, including:

[0087] Analyze the user association relationship between nodes based on the dynamic weight calculation results;

[0088] When a new association relationship appears in the user association relationship representation, a new association edge is added to the dynamic relationship graph and an initial weight is set.

[0089] Specifically, dynamic weight calculations are used to analyze the relationships between nodes (users or entities) in the dynamic relationship graph. Dynamic weights reflect the importance or influence of relationships between nodes. By analyzing these weights, key nodes and strong relationships in the dynamic relationship graph can be identified, and new relationships can be discovered within user relationships and represented in the dynamic relationship graph.

[0090] As time passes and the environment changes, new interactions or relationships may emerge between users. For example, two users with no previous transaction history may begin a transaction, or two entities may form a new partnership. These newly added relationships need to be captured and reflected in the graph in a timely manner.

[0091] When a new user association relationship is identified, a new association edge needs to be added to the dynamic relationship graph to represent the relationship, and new nodes can also be added according to the situation.

[0092] An initial weight is set for the newly added associated edges to reflect their relative importance in the graph. The initial weight can be set based on a variety of factors, such as transaction amount, interaction frequency, cooperation intensity, etc. The initial weight helps the graph neural network correctly evaluate the impact of these new edges in subsequent calculations.

[0093] In this embodiment, by analyzing the user association relationships between each node, the dynamic relationship graph can reflect changes in the network structure in real time, including newly added user association relationships; the calculation and application of dynamic weights enables the graph to not only capture static structural characteristics, but also adapt to and reflect dynamic changes in the network, thereby providing more accurate and timely information for graph-based analysis and decision-making.

[0094] In one embodiment, the model of the graph attention network is:

[0095] ;

[0096] Among them, h i (l+1) Represents the feature vector of node i in the l+1 layer; σ represents the activation function, which is a nonlinear function; ∑ represents the summation symbol, which means summing the weighted features of all adjacent nodes; j∈N(i) represents node j in the adjacent node set N(i) of node i; α ij represents the attention coefficient of node i to adjacent node j, which is used to measure the importance of node j to node i; W (l) represents the weight matrix of the lth layer, which is used to map the features of the nodes from the lth layer to the l+1th layer; h j(l) Represents the feature vector of node j in layer l.

[0097] The model of the temporal graph convolutional network is:

[0098] H(t)=TGCN(H(t−1),A(t)); where A(t) is the adjacency matrix at time t and H(t) is the hidden state of the node.

[0099] Specifically, GAT can adaptively learn the correlation between nodes through the attention mechanism, thereby generating richer and more discriminative feature representations for each node. ij Allowing the model to dynamically adjust its reliance on neighbor node information helps capture complex dependencies in the graph. ij It directly reflects the mutual influence between nodes. Therefore, the GAT model improves the interpretability of graph neural networks and makes the decision-making process of the model more transparent.

[0100] TGCN uses temporal convolution operations to capture changes in node features over time. Because it can handle time-varying graph structures, it is particularly well-suited for processing non-stationary time series data. By combining graph structure and time series information, TGCN improves the accuracy of predicting future node states. TGCN can adapt to dynamic changes in graph edges and nodes, such as added or deleted edges, which is crucial for real-time analysis and prediction.

[0101] In this implementation, combining GAT and TGCN allows for a more comprehensive understanding of the data by simultaneously considering both the spatial information of the graph structure and the temporal information of the time series. This improves the accuracy of risk propagation and node importance assessments. The combined use of graph attention networks and temporal graph convolutional networks significantly improves the ability to analyze dynamic relationship graphs and time series data.

[0102] In one embodiment, performing risk propagation calculation based on the dynamic relationship graph and node status to update the risk value of the risk source node includes:

[0103] The risk propagation model is used to identify the risk source nodes in the dynamic relationship graph, where the risk source nodes have initial risk values;

[0104] Calculate the risk propagation intensity in the dynamic relationship graph based on the initial risk value and the node status on the risk propagation path;

[0105] According to the risk propagation intensity in the dynamic relationship graph, the risk values of the risk source nodes within the affected range in the dynamic relationship graph are updated, and the risk source nodes whose risk values exceed the preset threshold are marked as high-risk nodes.

[0106] Specifically, possible risk source nodes are identified in the dynamic relationship graph. Risk source nodes refer to those nodes with initial risk values. These nodes may be considered as potential risk points due to certain reasons (such as financial problems, bad behavior records, etc.).

[0107] Calculating the intensity of risk propagation in the graph, based on the initial risk value and the states of the nodes along the risk propagation path, involves analyzing the ability of the risk source node to propagate risk to other nodes through its connected edges (i.e., its relationships with other nodes). The intensity of risk propagation depends on multiple factors, including edge weights (indicating relationship strength), node states (such as credit scores and historical behavior), and the parameters of the risk propagation model.

[0108] Based on the calculated risk propagation intensity, the risk values of the risk source nodes within the affected range are updated. The update process involves applying the calculated risk propagation intensity to the affected nodes, thereby adjusting their risk values. This step may involve complex calculations, such as considering the cumulative effect of risk and risk attenuation.

[0109] Identify and mark nodes whose risk values exceed a preset threshold. Once a node's risk value exceeds a certain preset threshold, it will be marked as a high-risk node. This helps to quickly identify risk points that require special attention and management.

[0110] In this embodiment, by dynamically updating risk values and identifying high-risk nodes, potential risk points can be more accurately identified. By identifying key risk nodes, monitoring and management resources can be more effectively allocated, improving the efficiency of risk management. By dynamically analyzing and updating risk values in the graph, an effective risk management and control method is provided, helping to maintain network stability and security.

[0111] The most detailed embodiment of this application is:

[0112] Step 1: Data access phase:

[0113] Source: Data is accessed from multiple sub-e-commerce platforms (such as e-commerce platform 1, e-commerce platform 2, etc.) and external data sources (such as industrial and commercial information, public opinion monitoring, etc.).

[0114] Types: include structured data (such as order data and user review data), unstructured data (such as customer service conversation text), time series data (such as logistics data), and merchant relationship data (such as equity and supply relationships).

[0115] Step 2: Data preprocessing stage:

[0116] Data cleaning: Clean the received data, including removing noise, filling missing values, and format standardization.

[0117] Data fusion: Integrate data from different sources according to a unified standard model to form a unified data view.

[0118] Step 3: Dynamic relationship graph construction phase:

[0119] Layer construction: Define heterogeneous node types (such as merchants, enterprises, and legal entities) and edge relationships (such as shareholdings, transactions, and regions).

[0120] Graph generation: Construct a dynamic graph G(t)=(V,E(t),W(t)), where V represents the node set, E(t) represents the edge set that changes over time, and W(t) represents the edge weight set.

[0121] Step 4: Model calculation phase:

[0122] Graph Attention Network (GAT): Aggregates first-order neighbor features to generate node embedding vectors.

[0123] ;

[0124] Among them, h i (l+1) Represents the feature vector of node i in the l+1 layer; σ represents the activation function, which is a nonlinear function; ∑ represents the summation symbol, which means summing the weighted features of all adjacent nodes; j∈N(i) represents node j in the adjacent node set N(i) of node i; α ij represents the attention coefficient of node i to adjacent node j, which is used to measure the importance of node j to node i; W (l) represents the weight matrix of the lth layer, which is used to map the features of the nodes from the lth layer to the l+1th layer; h j(l) Represents the feature vector of node j in layer l.

[0125] Temporal Graph Convolutional Network (TGCN): Capture temporal dependencies and update node states.

[0126] H(t)=TGCN(H(t−1),A(t));

[0127] Among them, A(t) is the adjacency matrix at time t, and H(t) is the hidden state of the node.

[0128] Step 5: Risk propagation calculation phase:

[0129] Risk source node identification: Use the risk propagation model to identify risk source nodes in the dynamic relationship graph, which have initial risk values.

[0130] Calculation of risk propagation intensity: Based on the initial risk value and the node status on the risk propagation path, the risk propagation intensity in the dynamic relationship graph is calculated.

[0131] Risk value update: Based on the risk propagation intensity in the dynamic relationship graph, the risk value of the risk source nodes within the affected range in the dynamic relationship graph is updated, and the risk source nodes whose risk value exceeds the preset threshold are marked as high-risk nodes.

[0132] Step 6: Apply the results:

[0133] Risk and transmission information visualization: Visualize the model calculation results, including merchant risk and transmission information.

[0134] Data analysis and decision-making application: Apply analysis results to the data analysis and decision-making module to support merchant management, including merchant capability credit result scoring, merchant access and reward and punishment management, etc.

[0135] Step 7: Continue to optimize:

[0136] Model evaluation and optimization: Regularly evaluate risk propagation results and optimize the parameters of the risk propagation model to improve the accuracy and timeliness of risk warnings.

[0137] Dynamic update: Apply the optimized risk propagation model to the next round of risk propagation calculations to form a closed loop and continuously improve the accuracy and timeliness of risk warnings.

[0138] Through the above steps, the entire technical solution realizes the complete process from data access, processing, analysis to application. It is particularly suitable for fields that need to deal with complex network relationships and dynamically changing environments, such as financial risk management, supply chain management, social network analysis, etc.

[0139] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0140] Based on the same inventive concept, the embodiments of the present application also provide a graph neural network-based user association analysis device for implementing the aforementioned graph neural network-based user association analysis method. The implementation solution provided by this device is similar to the implementation solution described in the aforementioned method. Therefore, the specific limitations of one or more graph neural network-based user association analysis device embodiments provided below can be found in the above-mentioned limitations of the graph neural network-based user association analysis method, and will not be repeated here.

[0141] In an exemplary embodiment, Figure 4 As shown, a user association analysis device based on graph neural network is provided, comprising:

[0142] A data acquisition module 402 is configured to acquire at least one type of business data, where the types of business data include user attribute data and user interaction data;

[0143] The data processing module 404 is used to perform data fusion processing on the user attribute data and the user interaction data and generate a dynamic relationship map;

[0144] Dynamic weight calculation module 406 is used to perform risk propagation quantification and weight calculation based on the dynamic relationship graph using a graph neural network to obtain a dynamic weight calculation result;

[0145] The user association analysis module 408 is used to adjust the dynamic relationship graph using the dynamic weight calculation result, and perform user association analysis based on the adjusted dynamic relationship graph.

[0146] In an exemplary embodiment, the dynamic weight calculation module 406 is also used to use a graph attention network to generate a node embedding vector based on a dynamic relationship graph; use a temporal graph convolutional network to obtain the changing relationship of node features over time based on the node embedding vector, and use the changing relationship to update the node status; perform risk propagation calculation based on the dynamic relationship graph and node status, and update the risk value of the risk source node; perform weight calculation based on the risk value of the risk source node to obtain a dynamic weight calculation result.

[0147] In an exemplary embodiment, the dynamic weight calculation module 406 is also used to analyze the user association relationship between each node based on the dynamic weight calculation result; when a new association relationship appears in the user association relationship representation, a new association edge is added to the dynamic relationship graph and an initial weight is set.

[0148] In an exemplary embodiment, the model of the graph attention network is:

[0149] ;

[0150] Among them, h i (l+1) Represents the feature vector of node i in the l+1 layer; σ represents the activation function, which is a nonlinear function; ∑ represents the summation symbol, which means summing the weighted features of all adjacent nodes; j∈N(i) represents node j in the adjacent node set N(i) of node i; α ij represents the attention coefficient of node i to adjacent node j, which is used to measure the importance of node j to node i; W (l) represents the weight matrix of the lth layer, which is used to map the features of the nodes from the lth layer to the l+1th layer; h j(l) Represents the feature vector of node j in layer l.

[0151] In an exemplary embodiment, the dynamic weight calculation module 506 is also used to use the risk propagation model to identify risk source nodes in the dynamic relationship graph, where the risk source nodes have an initial risk value; based on the initial risk value and the node status on the risk propagation path, the risk propagation intensity in the dynamic relationship graph is calculated; according to the risk propagation intensity in the dynamic relationship graph, the risk value of the risk source node within the affected range in the dynamic relationship graph is updated, and the risk source node whose risk value exceeds a preset threshold is marked as a high-risk node.

[0152] In an exemplary embodiment, user attribute data includes order data, user evaluation data, after-sales service data, penalty record data, as well as basic industrial and commercial data, association data and public opinion monitoring data obtained through third-party interfaces; user interaction data includes customer service conversation text, logistics data and user relationship data.

[0153] Each module in the above-mentioned graph neural network-based user association analysis device can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the above modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each of the above modules.

[0154] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 5 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store business data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a user association analysis method based on a graph neural network is implemented.

[0155] Those skilled in the art will understand that Figure 5The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0156] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0157] Acquire at least one type of business data, where the types of business data include user attribute data and user interaction data;

[0158] Perform data fusion processing on user attribute data and user interaction data, and generate dynamic relationship graphs;

[0159] Based on the dynamic relationship graph, the graph neural network is used to perform risk propagation quantification and weight calculation to obtain dynamic weight calculation results;

[0160] The dynamic weight calculation results are used to adjust the dynamic relationship map, and user association analysis is performed based on the adjusted dynamic relationship map.

[0161] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0162] Graph neural networks include graph attention networks and temporal convolutional networks;

[0163] Using graph attention network, we generate node embedding vectors based on dynamic relationship graphs;

[0164] Using the temporal graph convolutional network, we can obtain the changing relationship of node features over time based on the node embedding vector, and use the changing relationship to update the node status.

[0165] Perform risk propagation calculations based on dynamic relationship graphs and node status, and update the risk value of the risk source node;

[0166] According to the risk value of the risk source node, the dynamic weight calculation result is calculated.

[0167] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0168] Analyze the user association relationship between nodes based on the dynamic weight calculation results;

[0169] When a new association relationship appears in the user association relationship representation, a new association edge is added to the dynamic relationship graph and an initial weight is set.

[0170] In one embodiment, the model of the graph attention network is:

[0171] ;

[0172] Among them, h i (l+1) Represents the feature vector of node i in the l+1 layer; σ represents the activation function, which is a nonlinear function; ∑ represents the summation symbol, which means summing the weighted features of all adjacent nodes; j∈N(i) represents node j in the adjacent node set N(i) of node i; α ij represents the attention coefficient of node i to adjacent node j, which is used to measure the importance of node j to node i; W (l) represents the weight matrix of the lth layer, which is used to map the features of the nodes from the lth layer to the l+1th layer; h j(l) Represents the feature vector of node j in layer l.

[0173] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0174] The risk propagation model is used to identify the risk source nodes in the dynamic relationship graph, where the risk source nodes have initial risk values;

[0175] Calculate the risk propagation intensity in the dynamic relationship graph based on the initial risk value and the node status on the risk propagation path;

[0176] According to the risk propagation intensity in the dynamic relationship graph, the risk values of the risk source nodes within the affected range in the dynamic relationship graph are updated, and the risk source nodes whose risk values exceed the preset threshold are marked as high-risk nodes.

[0177] In one embodiment, user attribute data includes order data, user evaluation data, after-sales service data, penalty record data, as well as basic industrial and commercial data, association data, and public opinion monitoring data obtained through a third-party interface;

[0178] User interaction data includes customer service conversation texts, logistics data, and user relationship data.

[0179] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0180] Acquire at least one type of business data, where the types of business data include user attribute data and user interaction data;

[0181] Perform data fusion processing on user attribute data and user interaction data, and generate dynamic relationship graphs;

[0182] Based on the dynamic relationship graph, the graph neural network is used to perform risk propagation quantification and weight calculation to obtain dynamic weight calculation results;

[0183] The dynamic weight calculation results are used to adjust the dynamic relationship map, and user association analysis is performed based on the adjusted dynamic relationship map.

[0184] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0185] Graph neural networks include graph attention networks and temporal convolutional networks;

[0186] Using graph attention network, we generate node embedding vectors based on dynamic relationship graphs;

[0187] Using the temporal graph convolutional network, we can obtain the changing relationship of node features over time based on the node embedding vector, and use the changing relationship to update the node status.

[0188] Perform risk propagation calculations based on dynamic relationship graphs and node status, and update the risk value of the risk source node;

[0189] According to the risk value of the risk source node, the dynamic weight calculation result is calculated.

[0190] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0191] Analyze the user association relationship between nodes based on the dynamic weight calculation results;

[0192] When a new association relationship appears in the user association relationship representation, a new association edge is added to the dynamic relationship graph and an initial weight is set.

[0193] In one embodiment, the model of the graph attention network is:

[0194] ;

[0195] Among them, h i (l+1) Represents the feature vector of node i in the l+1 layer; σ represents the activation function, which is a nonlinear function; ∑ represents the summation symbol, which means summing the weighted features of all adjacent nodes; j∈N(i) represents node j in the adjacent node set N(i) of node i; α ij represents the attention coefficient of node i to adjacent node j, which is used to measure the importance of node j to node i; W (l) represents the weight matrix of the lth layer, which is used to map the features of the nodes from the lth layer to the l+1th layer; h j(l) Represents the feature vector of node j in layer l.

[0196] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0197] The risk propagation model is used to identify the risk source nodes in the dynamic relationship graph, where the risk source nodes have initial risk values;

[0198] Calculate the risk propagation intensity in the dynamic relationship graph based on the initial risk value and the node status on the risk propagation path;

[0199] According to the risk propagation intensity in the dynamic relationship graph, the risk values of the risk source nodes within the affected range in the dynamic relationship graph are updated, and the risk source nodes whose risk values exceed the preset threshold are marked as high-risk nodes.

[0200] In one embodiment, user attribute data includes order data, user evaluation data, after-sales service data, penalty record data, as well as basic industrial and commercial data, association data, and public opinion monitoring data obtained through a third-party interface;

[0201] User interaction data includes customer service conversation texts, logistics data, and user relationship data.

[0202] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0203] Acquire at least one type of business data, where the types of business data include user attribute data and user interaction data;

[0204] Perform data fusion processing on user attribute data and user interaction data, and generate dynamic relationship graphs;

[0205] Based on the dynamic relationship graph, the graph neural network is used to perform risk propagation quantification and weight calculation to obtain dynamic weight calculation results;

[0206] The dynamic weight calculation results are used to adjust the dynamic relationship map, and user association analysis is performed based on the adjusted dynamic relationship map.

[0207] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0208] Graph neural networks include graph attention networks and temporal convolutional networks;

[0209] Using graph attention network, we generate node embedding vectors based on dynamic relationship graphs;

[0210] Using the temporal graph convolutional network, we can obtain the changing relationship of node features over time based on the node embedding vector, and use the changing relationship to update the node status.

[0211] Perform risk propagation calculations based on dynamic relationship graphs and node status, and update the risk value of the risk source node;

[0212] According to the risk value of the risk source node, the dynamic weight calculation result is calculated.

[0213] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0214] Analyze the user association relationship between nodes based on the dynamic weight calculation results;

[0215] When a new association relationship appears in the user association relationship representation, a new association edge is added to the dynamic relationship graph and an initial weight is set.

[0216] In one embodiment, the model of the graph attention network is:

[0217] ;

[0218] Among them, h i (l+1) Represents the feature vector of node i in the l+1 layer; σ represents the activation function, which is a nonlinear function; ∑ represents the summation symbol, which means summing the weighted features of all adjacent nodes; j∈N(i) represents node j in the adjacent node set N(i) of node i; α ij represents the attention coefficient of node i to adjacent node j, which is used to measure the importance of node j to node i; W (l) represents the weight matrix of the lth layer, which is used to map the features of the nodes from the lth layer to the l+1th layer; h j(l) Represents the feature vector of node j in layer l.

[0219] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0220] The risk propagation model is used to identify the risk source nodes in the dynamic relationship graph, where the risk source nodes have initial risk values;

[0221] Calculate the risk propagation intensity in the dynamic relationship graph based on the initial risk value and the node status on the risk propagation path;

[0222] According to the risk propagation intensity in the dynamic relationship graph, the risk values of the risk source nodes within the affected range in the dynamic relationship graph are updated, and the risk source nodes whose risk values exceed the preset threshold are marked as high-risk nodes.

[0223] In one embodiment, user attribute data includes order data, user evaluation data, after-sales service data, penalty record data, as well as basic industrial and commercial data, association data, and public opinion monitoring data obtained through a third-party interface;

[0224] User interaction data includes customer service conversation texts, logistics data, and user relationship data.

[0225] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0226] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), quantum computing-based data processing logic devices, artificial intelligence (AI) processors, and the like.

[0227] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0228] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A user association analysis method based on graph neural network, characterized in that: The method comprises: Acquire at least one type of business data, where the types of the business data include user attribute data and user interaction data; Performing data fusion processing on the user attribute data and the user interaction data, and generating a dynamic relationship graph; Based on the dynamic relationship graph, the graph neural network is used to perform risk propagation quantification and weight calculation to obtain dynamic weight calculation results; The dynamic weight calculation result is used to adjust the dynamic relationship map, and user association analysis is performed based on the adjusted dynamic relationship map.

2. The method according to claim 1, characterized in that The graph neural network includes a graph attention network and a temporal convolutional network. The graph neural network is used to perform risk propagation quantification and weight calculation based on the dynamic relationship graph to obtain dynamic weight calculation results, including: Using graph attention network, we generate node embedding vectors based on dynamic relationship graphs; Using a temporal graph convolutional network, we obtain the relationship between node features and time according to the node embedding vector, and use the relationship to update the node status. Perform risk propagation calculation based on the dynamic relationship graph and the node status, and update the risk value of the risk source node; According to the risk value of the risk source node, weight calculation is performed to obtain the dynamic weight calculation result.

3. The method according to claim 2, characterized in that The adjusting the dynamic relationship graph by using the dynamic weight calculation result includes: Analyze the user association relationship between each node according to the dynamic weight calculation result; When a new association relationship appears in the user association relationship representation, a new association edge is added to the dynamic relationship graph and an initial weight is set.

4. The method according to claim 2, characterized in that The model of the graph attention network is: ； Among them, h i (l+1) Represents the feature vector of node i in the l+1 layer; σ represents the activation function, which is a nonlinear function; ∑ represents the summation symbol, which means summing the weighted features of all adjacent nodes; j∈N(i) represents node j in the adjacent node set N(i) of node i; α ij represents the attention coefficient of node i to adjacent node j, which is used to measure the importance of node j to node i; W (l) represents the weight matrix of the lth layer, which is used to map the features of the nodes from the lth layer to the l+1th layer; h j(l) Represents the feature vector of node j in layer l.

5. The method according to claim 2, characterized in that The risk propagation calculation based on the dynamic relationship graph and the node status, and updating the risk value of the risk source node, includes: Identifying a risk source node in a dynamic relationship graph using a risk propagation model, wherein the risk source node has an initial risk value; Calculating the risk propagation intensity in the dynamic relationship graph based on the initial risk value and the node states on the risk propagation path; According to the risk propagation intensity in the dynamic relationship graph, the risk values of the risk source nodes within the affected range in the dynamic relationship graph are updated, and the risk source nodes whose risk values exceed the preset threshold are marked as high-risk nodes.

6. The method according to claim 1, characterized in that The user attribute data includes order data, user evaluation data, after-sales service data, penalty record data, as well as basic industrial and commercial data, association data and public opinion monitoring data obtained through third-party interfaces; The user interaction data includes customer service conversation text, logistics data and user relationship data.

7. A user association analysis device based on graph neural network, characterized in that: The device comprises: A data acquisition module, configured to acquire at least one type of business data, wherein the types of the business data include user attribute data and user interaction data; A data processing module, configured to perform data fusion processing on the user attribute data and the user interaction data, and generate a dynamic relationship graph; The dynamic weight calculation module is used to perform risk propagation quantification and weight calculation based on the dynamic relationship graph using the graph neural network to obtain the dynamic weight calculation results; The user association analysis module is used to adjust the dynamic relationship map using the dynamic weight calculation result, and perform user association analysis based on the adjusted dynamic relationship map.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images