Data processing method and device, nonvolatile storage medium and electronic equipment

Abstract

The invention discloses a data processing method and device, a nonvolatile storage medium and electronic equipment. The method comprises the following steps: generating a call network diagram according to a calling number, a called number, the number of calls between the calling number and the called number and the call duration; determining an influence label corresponding to the number according to the call network diagram; generating a graph network structure according to the user data corresponding to the numbers, the influence labels corresponding to the numbers and the incidence relation between the numbers; updating a graph network structure by using a propagation mechanism based on a PageRank algorithm to obtain a target feature representation; and inputting the target feature representation into the graph neural network model to obtain an influence index corresponding to a number output by the graph neural network model. According to the method and the device, the technical problem of relatively low prediction precision of the node influence in the social network caused by difficulty in capturing the complexity of the node influence in the social network only depending on a single algorithm or model is solved.

Description

Technical Field

[0001] This application relates to the field of big data mining technology, and more specifically, to a data processing method and apparatus, a non-volatile storage medium, and an electronic device. Background Technology

[0002] In today's era of rapid development in big data and artificial intelligence (AI) technologies, the integration of social network and telecommunications service data has become a key area for studying users' social influence. Social influence refers to the degree to which an individual affects the behavior, attitudes, and decisions of other individuals within a social network, playing a particularly important role in information dissemination, brand promotion, and user behavior analysis. Telecommunications service data, such as call records, internet usage behavior, and user package information, contains rich information on user behavior patterns and social relationships. Therefore, predicting social influence based on telecommunications service data can not only provide more accurate user profiles but also provide strong support for personalized services and social network analysis.

[0003] In utilizing telecommunications service data, traditional statistical analysis methods, such as the PageRank algorithm, can calculate preliminary social influence based on users' social circles. However, these methods often overlook multi-dimensional data such as users' basic attributes, service plans, and app preferences, resulting in limited accuracy and coverage of predictions. The PageRank algorithm itself is based on the importance assessment of hyperlink structures; when directly applied to social network data, its complexity and non-linearity in social relationships may prevent it from fully capturing the true state of social influence.

[0004] In summary, relying solely on the PageRank algorithm or a single neural network model is insufficient to capture the complexity of node influence in social networks.

[0005] There is currently no effective solution to the above problems. Summary of the Invention

[0006] This application provides a data processing method and apparatus, a non-volatile storage medium, and an electronic device to at least solve the technical problem of low prediction accuracy of node influence in social networks, which is caused by the difficulty in capturing the complexity of node influence in social networks due to relying on only a single algorithm or model.

[0007] According to one aspect of the embodiments of this application, a data processing method is provided, comprising: generating a call network graph based on a calling number, a called number, the number of calls between the calling number and the called number, and the call duration, wherein the nodes in the call network graph are the calling number and the called number, and the edges in the call network graph are the number of calls between the calling number and the called number, and the call duration; determining an influence label corresponding to a number based on the call network graph; generating a graph network structure based on user data corresponding to a number, the influence label corresponding to a number, and the association between numbers, wherein the nodes in the graph network structure are user data corresponding to a number and influence labels corresponding to a number, and the edges in the graph network structure are associations between numbers; updating the graph network structure using a propagation mechanism based on the PageRank algorithm to obtain a target feature representation; and inputting the target feature representation into a graph neural network model to obtain the influence index corresponding to the number output by the graph neural network model.

[0008] Optionally, the influence label corresponding to the number is determined based on the call network graph, including: setting the initial PageRank value of all nodes in the call network graph, wherein when learning the information of any node, the node is initialized to 1 and other nodes are initialized to 0, and the PageRank value is used to represent the influence value corresponding to the number; step S1, for the first node, calculate the number of connections a between the first node and the second node, if the PageRank value of the first node is PR(p), then the PageRank value passed from the first node to each second node is PR(p) / a, wherein the first node and the second node are different nodes; step S2, for the second node, determine the set of nodes pointing to the second node, obtain the PageRank value passed from the first node to the second node, and update the PageRank value of the second node according to the PageRank value passed from the first node to the second node; repeat steps S1 and S2 to perform multiple iterative calculations until the preset convergence condition is met to obtain the influence label corresponding to the number, wherein the preset convergence condition includes: the change value of the PageRank value of all nodes between two consecutive iterations is less than a preset threshold.

[0009] Alternatively, the PageRank value of a node in the call network graph can be calculated using the following formula: Among them, pr(p i ) is node p i PageRank value, pr(p j ) is node p j PageRank value, T ij It is node p i Received node p j Number of calls, L j It is node pj The number of outgoing calls, q is the preset damping coefficient, and N is the number of nodes in the call network diagram.

[0010] Optionally, the graph network structure is updated using the propagation mechanism based on the PageRank algorithm to obtain the target feature representation, including: constructing the adjoint matrix of the graph network structure according to the association relationship between numbers, wherein if there are n nodes in the graph network structure and there is an edge connection between node i and node j, then A in the adjoint matrix ij Elements are assigned non-zero values ​​based on the weights of edge connections, where i and j are positive integers and n is a positive integer greater than 1; a node feature matrix representing the features of graph network nodes is received, where the rows of the node feature matrix are the feature vectors corresponding to the nodes; the adjoint matrix is ​​normalized to obtain the target adjoint matrix, and the linear combination coefficient matrix of the identity matrix and the target adjoint matrix is ​​calculated; based on the coefficient matrix and the node feature matrix, iterative calculation is performed. In each iteration, node feature information is propagated and fused along the edge structure of the graph network according to the PageRank propagation mechanism, so that each node obtains a comprehensive feature representation from its neighboring nodes and its own historical feature information; if the iteration process meets the preset convergence judgment condition, the node feature matrix is ​​used as the target feature representation after the graph network structure is updated.

[0011] Optionally, the target feature representation is input into a graph neural network model to obtain the influence index corresponding to the number output by the graph neural network model. This includes: using the target feature representation as the initial feature representation of the nodes in the graph neural network model; determining the set of neighboring nodes corresponding to each node in the graph neural network model; for each node, performing an aggregation operation on the feature representations of its neighboring nodes through a preset information aggregation function to obtain the aggregated features of the neighboring nodes; updating the initial feature representation of each node through a preset update function based on the aggregated features of the neighboring nodes and the current feature representation of each node itself, and constructing a transformation graph based on the updated feature representation of the nodes; inputting the node feature vectors in the transformation graph into a fully connected layer for feature transformation, and inputting the feature transformation result into the output layer to obtain the influence index output by the output layer.

[0012] Optionally, before generating the graph network structure based on the user data corresponding to the number, the influence tag corresponding to the number, and the relationship between the numbers, the method further includes: obtaining the user data corresponding to the number; scanning for missing values ​​for each feature in the user data; if the value of the target feature in one or more samples is a preset missing value, then recording that the target feature has a missing value and determining the target sample where the missing value is located; calculating the mean corresponding to all non-missing samples in the target sample and replacing the missing value with the mean; calculating the mean and standard deviation of each numerical feature in the user data, setting an outlier judgment threshold, and determining that the feature value is an outlier if the feature value of the target data is greater than the outlier judgment threshold; correcting the outlier using a preset Gaussian mixture model; standardizing the corrected data and performing one-hot encoding on the standardized data.

[0013] Optionally, user data includes at least one of the following: business object domain data, deep packet inspection internet access data, and social circle data, wherein social circle data includes: contact list data and social relationship attribute information.

[0014] According to another aspect of the embodiments of this application, a data processing apparatus is also provided, comprising: a first generation module, configured to generate a call network graph based on a calling number, a called number, the number of calls between the calling number and the called number, and the call duration, wherein the nodes in the call network graph are the calling number and the called number, and the edges in the call network graph are the number of calls between the calling number and the called number, and the call duration; a determination module, configured to determine an influence label corresponding to a number based on the call network graph; a second generation module, configured to generate a graph network structure based on user data corresponding to a number, an influence label corresponding to a number, and the association between numbers, wherein the nodes in the graph network structure are user data corresponding to a number and influence labels corresponding to a number, and the edges in the graph network structure are the association between numbers; an update module, configured to update the graph network structure using a propagation mechanism based on the PageRank algorithm to obtain a target feature representation; and a processing module, configured to input the target feature representation into a graph neural network model to obtain the influence index corresponding to the number output by the graph neural network model.

[0015] According to another aspect of the embodiments of this application, a non-volatile storage medium is also provided, the storage medium including a stored program, wherein the program controls the device where the storage medium is located to execute the above data processing method when it runs.

[0016] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory and a processor, wherein the processor is configured to run a program stored in the memory, wherein the program executes the above-described data processing method when it runs.

[0017] According to another aspect of the embodiments of this application, a computer program is also provided, wherein the computer program, when executed by a processor, implements the above-described data processing method.

[0018] According to another aspect of the embodiments of this application, a computer program product is also provided, the computer program product including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and the computer program implements the above-described data processing method when executed by a processor.

[0019] In this embodiment, a call network graph is generated based on the calling number, the called number, the number of calls between the calling and called numbers, and the call duration. The nodes in the call network graph are the calling and called numbers, and the edges are the number of calls between the calling and called numbers and the call duration. Based on the call network graph, influence tags corresponding to the numbers are determined. A graph network structure is generated based on the user data corresponding to the numbers, the influence tags corresponding to the numbers, and the relationships between the numbers. The nodes in the graph network structure are the user data corresponding to the numbers and the influence tags corresponding to the numbers, and the edges are the relationships between the numbers. The propagation mechanism based on the PageRank algorithm is used to update... The graph network structure is used to obtain the target feature representation. The target feature representation is then input into the graph neural network model to obtain the influence index corresponding to the number output by the graph neural network model. By combining the PageRank algorithm with the graph neural network model, the user data corresponding to the number, the influence label corresponding to the number, and the correlation between the numbers are processed to obtain the influence index corresponding to the number. This achieves the goal of effectively capturing the complexity of node influence in social networks, thereby improving the technical effect of predicting the influence of nodes in social networks. It also solves the technical problem of low prediction accuracy of node influence in social networks caused by relying on a single algorithm or model, which is difficult to capture the complexity of node influence in social networks. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0021] Figure 1 This is a flowchart of a data processing method according to an embodiment of this application;

[0022] Figure 2 This is a schematic diagram of a call network according to an embodiment of this application;

[0023] Figure 3This is a schematic diagram of a caller ID communication circle according to an embodiment of this application;

[0024] Figure 4 This is a flowchart of a method for predicting influence according to an embodiment of this application;

[0025] Figure 5 This is a schematic diagram of personalized PageRank propagation according to an embodiment of this application;

[0026] Figure 6 This is a schematic diagram of a graph neural network operation according to an embodiment of this application;

[0027] Figure 7 This is a schematic diagram of a diagram conversion according to an embodiment of this application;

[0028] Figure 8 This is a flowchart of a graph neural network algorithm according to an embodiment of this application;

[0029] Figure 9 This is a flowchart of an algorithm for predicting influence using a graph neural network according to an embodiment of this application;

[0030] Figure 10 This is a structural diagram of a data processing apparatus according to an embodiment of this application;

[0031] Figure 11 This is a hardware structure block diagram of a computer terminal according to an embodiment of the present application for a data processing method. Detailed Implementation

[0032] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0033] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0034] According to an embodiment of this application, a method embodiment for data processing is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0035] Figure 1 This is a flowchart of a data processing method according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0036] Step S101: Generate a call network graph based on the calling number, the called number, the number of calls between the calling number and the called number, and the call duration. In the call network graph, the nodes are the calling number and the called number, and the edges are the number of calls between the calling number and the called number, and the call duration.

[0037] To construct the call network graph, nodes and edges need to be defined. In this application, nodes are the calling and called numbers, representing individual users; edges are the connections between the calling and called numbers, and their weights are determined by the number of calls and the call duration. For example, if user A frequently calls user B and each call is relatively long, then the edge weight from A to B will be higher, and vice versa.

[0038] Step S102: Determine the influence label corresponding to the number based on the call network diagram.

[0039] For each node (user) in the call network graph, a PageRank value is first initialized. The out-degree and in-degree of each node are calculated: the out-degree is the number of edges a node sends, i.e., the number of calls a node makes to other nodes; the in-degree is the number of edges a node receives, i.e., the number of calls other nodes make to that node. The PageRank value of each node is updated iteratively. During the iteration, a node's PageRank value is updated based on the "votes" (i.e., calls) it receives from other nodes and the PageRank values ​​of those nodes themselves.

[0040] The PageRank value of each node is iteratively updated until the change in the PageRank value is less than a given threshold (e.g., 0.00001), or a predetermined number of iterations is reached. Once the PageRank values ​​converge, an influence label is assigned to each node based on its final PageRank value. Nodes with higher PageRank values ​​have more significant influence in the network graph and can therefore be assigned a higher influence label.

[0041] Step S103: Generate a graph network structure based on the user data corresponding to the number, the influence tag corresponding to the number, and the relationship between the numbers. The nodes in the graph network structure are the user data corresponding to the number and the influence tag corresponding to the number, and the edges in the graph network structure are the relationship between the numbers.

[0042] After determining the user influence tags, further integrate multi-dimensional data such as user basic attributes, plan information, and APP preferences to construct a more comprehensive graph network structure. The specific steps are as follows: 1. Convert user basic attributes, plan information, and APP preferences into a feature matrix. This data can be numerical (e.g., call duration, age) or categorical (e.g., plan type, APP category). 2. Combine the node information in the call network graph with the above feature matrix. Each user's node includes not only the weight of its call connections but also its basic attributes, plan information, and APP preferences. 3. Use the graph's adjoint matrix and self-loop matrix to describe the connections between users and the information transmission mechanisms within user nodes themselves.

[0043] Step S104: Update the graph network structure using the propagation mechanism based on the PageRank algorithm to obtain the target feature representation.

[0044] For each node in the graph, its PageRank value is initialized based on the number of "votes" it receives from other nodes in the call network (i.e., the number of calls it receives) and the weight of the call duration. Here, each "vote" carries a certain weight, which is proportional to the call frequency and call duration.

[0045] In personalized PageRank, a propagation probability α is introduced. This means that a node has an α probability of transmitting its information to its neighboring nodes and a 1-α probability of transmitting the information back to itself during propagation. This probability setting helps information spread more widely in the graph while maintaining the node's own influence. The personalized PageRank propagation formula is used to update information propagation for each node in the graph.

[0046] The node features after information propagation serve as input to the graph neural network (Graph Neural Network). In each layer of the Graph Neural Network, nodes are updated based on information from their neighbors and their own PageRank-propagated features. This update is based on a strategy of neighbor information aggregation and node state updates. Neighbor information aggregation involves collecting and fusing information from a node's neighbors, while node state updates involve recalculating features by combining the node's current state with neighbor information. In the Graph Neural Network, each node's features are updated to reflect its influence within the network structure and its interactions with neighbors. The update process consists of two parts: neighbor information aggregation and node state updates. Neighbor information aggregation is achieved through the adjacency matrix and propagation mechanism of the Graph Neural Network, while node state updates are performed using an update function within the Graph Neural Network. These two steps are repeated until the network reaches a predetermined depth or the node features converge.

[0047] After information propagation and node feature updates are completed, the updated graph network structure is obtained, in which each node has a target feature representation that reflects its influence in the network and its interaction with other nodes.

[0048] Step S105: Input the target feature representation into the graph neural network model to obtain the influence index corresponding to the number output by the graph neural network model.

[0049] The graph neural network model described above was trained using the following method: First, the input data was organized into an appropriate graph structure, including node feature matrices, edge indices or adjacency matrices, and node labels. In the social influence prediction scenario, node features include basic user information, package information, social circle data, and app preference data; edge indices describe the relationships between users; and node labels represent the influence magnitude initially calculated using the PageRank algorithm. Second, the parameters of the graph neural network were initialized, including the weight matrix of node features, the parameters of the aggregation operation, and the parameters of the update function. Then, forward propagation was performed: for each node, the aggregation function was used to aggregate the feature representations of its neighboring nodes to form aggregated features of neighbor information. The node's current features were combined with the aggregated features of its neighbors, and the node's feature representation was updated using the update function. In the personalized PageRank mechanism, a node has a probability α of transmitting information back to itself and a probability 1-α of transmitting information to its neighboring nodes. Through multiple iterations, information was propagated in the graph until a preset number of iterations or feature representation convergence was reached. After the final propagation layer and node state update, the node's feature representation was input into the fully connected layer and the output layer to obtain the predicted influence magnitude. Calculate the loss between the predicted values ​​and the true labels, such as mean squared error or other loss functions suitable for regression or classification. Next, perform backpropagation: use the backpropagation algorithm to calculate the gradient of the loss function with respect to the model parameters. Update the model parameters using an optimization algorithm (such as gradient descent or its variants) to minimize the loss function. Evaluate the model performance on the validation set, adjusting model parameters such as learning rate, number of network layers, aggregation and update functions based on performance metrics. Optimize hyperparameters using methods such as grid search, random search, or Bayesian optimization. Finally, terminate training when the model's performance on the validation set no longer significantly improves or reaches the preset number of training epochs. Evaluate the model's final performance using test set data to confirm its generalization ability.

[0050] Based on the above steps, by combining the PageRank algorithm with a graph neural network model, the user data corresponding to the number, the influence tags corresponding to the number, and the correlation between the numbers are processed to obtain the influence index corresponding to the number. This achieves the goal of effectively capturing the complexity of node influence in social networks, thereby improving the technical effect of predicting the influence of nodes in social networks.

[0051] The following are Figure 1 The steps shown are illustrated and explained by way of example.

[0052] According to some optional embodiments of this application, the influence label corresponding to a number can be determined based on the call network graph by the following method: Set the initial PageRank value of all nodes in the call network graph, wherein when learning information about any node, that node is initialized to 1, and other nodes are initialized to 0. The PageRank value is used to characterize the influence value corresponding to the number; Step S1: For the first node, calculate the number of connections a between the first node and the second node. If the PageRank value of the first node is PR(p), then the PageRank value passed from the first node to each second node is PR(p) / a, where the first node and the second node are different nodes; Step S2: For the second node, determine the set of nodes pointing to the second node, obtain the PageRank value passed from the first node to the second node, and update the PageRank value of the second node based on the PageRank value passed from the first node to the second node; Repeat steps S1 and S2 to perform multiple iterative calculations until a preset convergence condition is met to obtain the influence label corresponding to the number, wherein the preset convergence condition includes: the change in PageRank value of all nodes between two consecutive iterations is less than a preset threshold.

[0053] Figure 2 This is a schematic diagram of a call network diagram according to an embodiment of this application. The call network diagram includes the calling number, the called number, and the call relationship between them.

[0054] As the algorithm iterates, a node's influence is reassessed based on its links (call relationships) within the network. For any first node p in the network, the number a of its connections to all second nodes is calculated, i.e., the number of times node p makes calls to other nodes. If node p's current PageRank value is PR(p), then the PageRank value it passes to each second node is PR(p) / a. This means that node p's influence is distributed to other nodes based on who it calls and how many times it makes those calls.

[0055] For the second node q, determine the set of all nodes that point to it. The PageRank value passed from each first node p to q is calculated following the steps described above.

[0056] Repeat the above steps—information propagation and PageRank value update—for each node in the network. After each iteration, check the changes in the PageRank values ​​of all nodes to determine if a preset convergence condition is met. The convergence condition is set so that the changes in the PageRank values ​​of all nodes between two consecutive iterations are less than a preset threshold (e.g., 0.0001). Once this condition is met, the iteration process ends, and the final PageRank value will be used as the node's influence label.

[0057] Once the iterative computation reaches convergence, the PageRank value of each node stabilizes, and these values ​​become the node's (user's) influence label. The higher the PageRank value of each node, the greater its influence in the social network. These labels can be used for subsequent graph neural network training to predict a user's social influence.

[0058] Specifically, the PageRank value of a node in the call network graph is calculated using the following formula: Among them, pr(p i ) is node p i PageRank value, pr(p j ) is node p j PageRank value, T ij It is node p i Received node p j Number of calls, L j It is node p j The number of outgoing calls, q is the preset damping coefficient, and N is the number of nodes in the call network diagram.

[0059] Understandably, if a number is called by many other numbers, it indicates that the number is relatively important, meaning its PageRank value will be relatively high. If a number with a high PageRank value calls other numbers, the PageRank value of the called numbers will increase accordingly.

[0060] Figure 3 This is a schematic diagram of a caller ID social circle according to an embodiment of this application. Assume that the social circle of number A is as follows: Figure 3 As shown, the influence of number A is: PR(A) = PR(B) * 2 / 15 + PR(C) * 1 / 13 + PR(D) * 5 / 21. The number of users who influence A is: SORT(PR(C) * 50, PR(B) * 32, PR(D) * 10).

[0061] According to some alternative embodiments of this application, updating the graph network structure using the propagation mechanism based on the PageRank algorithm to obtain the target feature representation can be achieved as follows: Based on the association relationships between numbers, construct the adjoint matrix of the graph network structure, where if there are n nodes in the graph network structure, and there is an edge connection between node i and node j, then A in the adjoint matrix... ijElements are assigned non-zero values ​​based on the weights of edge connections, where i and j are positive integers and n is a positive integer greater than 1; a node feature matrix representing the features of graph network nodes is received, where the rows of the node feature matrix are the feature vectors corresponding to the nodes; the adjoint matrix is ​​normalized to obtain the target adjoint matrix, and the linear combination coefficient matrix of the identity matrix and the target adjoint matrix is ​​calculated; based on the coefficient matrix and the node feature matrix, iterative calculation is performed. In each iteration, node feature information is propagated and fused along the edge structure of the graph network according to the PageRank propagation mechanism, so that each node obtains a comprehensive feature representation from its neighboring nodes and its own historical feature information; if the iteration process meets the preset convergence judgment condition, the node feature matrix is ​​used as the target feature representation after the graph network structure is updated.

[0062] Construct an adjoint matrix A based on the relationships between nodes in the network graph. The adjoint matrix is ​​an (n*n) matrix, where (n) is the number of nodes in the graph. If there is an edge connecting node i and node j, the elements A in the matrix are... ij The weight of the edge (number of calls or call duration) will be set to a non-zero value. If there is no direct connection between the two, then A... ij The value is 0. The adjoint matrix reflects the strength of direct connections between nodes in the network. The node feature matrix is ​​a matrix containing the features of all nodes, where each row represents a feature vector of a node. This feature vector can contain basic user attributes, package information, APP preferences, etc., obtained through the data preprocessing stage. The adjoint matrix is ​​normalized to ensure the smoothness of information propagation and to avoid rapid decay or explosion of information during propagation. Normalization usually includes row normalization or column normalization; row normalization is used, meaning the sum of the elements in each row is 1. The linear combination coefficient matrix of the identity matrix (I) and the normalized adjoint matrix is ​​calculated. Iterative calculation is performed based on the linear combination coefficient matrix and the node feature matrix. In each iteration, node feature information is propagated and fused along the edge structure of the graph network according to the PageRank propagation mechanism, so that each node obtains a comprehensive feature representation from its neighboring nodes and its own historical feature information. Iterative calculation continues until a preset convergence criterion is met. The convergence criterion is generally set as the change in the node feature matrix between two consecutive iterations being less than a preset threshold, or reaching the maximum number of iterations.

[0063] Through the above steps, the graph neural network structure can be effectively constructed and updated, ensuring the personalization of information dissemination and the full utilization of the graph structure, thereby improving the accuracy and stability of social influence prediction.

[0064] In some optional embodiments of this application, inputting the target feature representation into a graph neural network model to obtain the influence index corresponding to the number output by the graph neural network model can be achieved by the following method: using the target feature representation as the initial feature representation of the nodes in the graph neural network model; determining the set of neighboring nodes corresponding to each node in the graph neural network model; for each node, performing an aggregation operation on the feature representations of its neighboring nodes through a preset information aggregation function to obtain the aggregated features of the neighboring nodes; updating the initial feature representation of each node through a preset update function based on the aggregated features of the neighboring nodes and the current feature representation of each node itself, and constructing a transformation graph based on the updated feature representation of the nodes; inputting the node feature vectors in the transformation graph into a fully connected layer for feature transformation, and inputting the feature transformation result into the output layer to obtain the influence index output by the output layer.

[0065] Assume h v (l) Let U represent the initial feature representation of node v at layer l. For each node to be updated, information is aggregated from its neighboring nodes and calculated using a statistical aggregation function. Let U be the aggregated information representation of node v. v (l) ,but:

[0066]

[0067] Where N(v) is the set of neighboring nodes of v, h μ (l) Let g(·) represent the feature representation of node μ at layer l. Let g(·) represent the neighbor information aggregation function. Indicates aggregation operation

[0068] Combining the current information of the node with the aggregated information, the state is updated using an update function. Let's assume the updated state of the current node is h. v (l+1) Then h v (l+1) =f(h) v (l) U v (l) ).

[0069] A readout function integrates the representations of all nodes into a single overall graph representation. Predictions for each node are obtained by introducing a fully connected layer and an output layer. Let the prediction for the current node be y. pred , then y pred =Linear(Relu(h) v (l+1) )).

[0070] As some optional embodiments of this application, before generating the graph network structure based on the user data corresponding to the number, the influence tag corresponding to the number, and the association between the numbers, the following steps may also be performed: obtaining the user data corresponding to the number; scanning for missing values ​​for each feature in the user data, if the value of the target feature in one or more samples is a preset missing identifier, then recording that the target feature has a missing value and determining the target sample where the missing value is located; calculating the mean corresponding to all non-missing samples in the target sample, and replacing the missing value with the mean; calculating the mean and standard deviation of each numerical feature in the user data, setting an outlier judgment threshold, if the feature value of the target data is greater than the outlier judgment threshold, determining that the feature value is an outlier; correcting the outlier using a preset Gaussian mixture model; standardizing the corrected data, and performing one-hot encoding on the standardized data.

[0071] Retrieve various user data corresponding to each number from the telecommunications service database, including but not limited to basic attributes, package information, internet access behavior, and app preferences. For each feature in the user data, perform a missing value scan. If a feature is found to have a value with a preset missing identifier (such as NaN or None) in one or more samples, record that the feature has missing values ​​in these samples and identify these samples as target samples. Once the target samples with missing values ​​are identified, calculate the mean of all non-missing samples in the target samples. Subsequently, replace the missing values ​​of that feature in the target samples with the calculated mean.

[0072] Next, the mean and standard deviation of each numerical feature in the user data are calculated to determine its distribution. An outlier threshold is set, such as the mean plus or minus three times the standard deviation. If a feature value of a sample exceeds this threshold, it is considered an outlier. To correct for outliers, a pre-defined Gaussian Mixture Model (MM) is used for fitting. GMM can identify multiple Gaussian distributions in the data, thus more accurately estimating the distribution to which outliers belong. Replacing outliers with the most probable values ​​predicted by the GMM reduces the impact of outliers on model training.

[0073] After handling missing and outlier values, the corrected data is standardized to eliminate the influence of dimensions between features, making model training more stable. Standardization methods include min-max standardization and standard deviation standardization. The former scales the data to the interval ([0,1]), while the latter standardizes the data to a distribution with a mean of 0 and a standard deviation of 1.

[0074] For categorical features, such as user package type and app preferences, one-hot encoding is used. One-hot encoding converts categorical features into binary vectors, with each category corresponding to one dimension of the vector. A value of 1 in this dimension indicates that the sample belongs to that category, while the values ​​of the other dimensions are 0. This avoids the model misinterpreting the order of categorical features during training, thus improving the model's generalization ability.

[0075] Preferably, the user data includes at least one of the following: business object domain data, deep packet inspection internet access data, and social circle data, wherein the social circle data includes: contact list data and social relationship attribute information.

[0076] Comprehensive collection and preprocessing of user data is crucial for building a social influence prediction system based on telecommunications business data. This data covers Business Object Domain (BO domain) data, Deep Packet Inspection (DPI) data, and Social Networking Data, each providing a different perspective on user behavior and social networks. The following is a detailed description of these data types and how to process contact list data and social relationship attribute information within Social Networking Data.

[0077] Object domain data (BO domain data): This includes user account information, transaction records, package details, service usage, etc. BO domain data provides direct evidence of user interactions with telecom operators and can be used to analyze user consumption habits and service preferences.

[0078] Deep Packet Inspection (DPI) data: DPI technology can deeply analyze network packets to extract information about a user's online behavior, including the websites visited, applications used, and data traffic patterns. This data helps to understand a user's online social activities, interests, and browsing habits, and is one of the important indicators for measuring a user's social influence.

[0079] Social Network Data: Social network data provides a detailed picture of a user's social connection network, including contact list data, social relationship attributes, and more. This data helps identify a user's key social partners, the tightness of their social network, and the nature of their social relationships.

[0080] As described above, this application calculates a user's social influence based on user-level social circle call data using the PageRank algorithm. Based on this result, highly active users and their phone numbers are selected. In addition to the original features, the application expands upon these features with dimensions such as basic user attributes, service plans, and app preferences. Using the selected user data, a graph neural network model is trained based on the PageRank propagation mechanism. The trained model is then used to predict and correct the social influence of all users.

[0081] This application cleverly combines the PageRank algorithm with graph neural networks to deepen the network layers of the model, enabling the network model to learn node information more comprehensively and accurately, thereby achieving accurate prediction of users' social influence.

[0082] Figure 4 This is a flowchart of a method for predicting influence according to an embodiment of this application, such as... Figure 4 The method includes the following steps:

[0083] Step S401: Input the user's basic information, package information, social circle data and APP preference data, the connection relationship between users and the social influence size label output by PageRank.

[0084] Step S402: Analyze the correlation between data, and perform missing value and outlier handling, as well as data standardization and one-hot encoding.

[0085] Step S403: Generate a graph network structure based on the three elements: feature data, edge index, and label.

[0086] Step S404: Based on the generated graph network structure, information is propagated using a personalized PageRank propagation mechanism. This propagation mechanism allows node information to be transmitted back to itself with a probability of α, and to re-enter the hypergraph neural network with a probability of 1-α.

[0087] Figure 5 This is a schematic diagram of personalized PageRank propagation according to an embodiment of this application. The personalized PageRank propagation mechanism will be explained in detail below.

[0088] For the root node x1, information propagation occurs between connected nodes. During the journey from one vertex to the next, there is an α probability of returning to x1 itself, and a 1-α probability of propagating to x2, x3, and x4. Simultaneously, in the next propagation, these three vertices also have an α probability of propagating back to x1. This propagation process is centered on x1, meaning it serves as the root node for this propagation, and the ultimately learned information is the information representation of that node.

[0089] Personalized PageRank is a PageRank algorithm tailored to an individual. Each time a new page traversal is performed, it starts from the root node, the user node μ. When learning information about node μ, it is initialized to 1, while other nodes are initialized to 0. The formula for personalized PageRank is as follows:

[0090] πi x =(1-α)Aπi x +αi x

[0091] i x πi is the eigenvector of the root node; A is the adjacency matrix after normalization and self-looping; x The characteristics after propagation; α is the propagation probability.

[0092] Figure 6 This is a schematic diagram of a graph neural network operation according to an embodiment of this application, such as... Figure 6 As shown, graph neural networks are neural network models suitable for graph data. The initial feature representation of user nodes is input into the graph neural network model, and the output is the feature learned by the user nodes through the graph neural network model. By combining PageRank and graph neural networks, the number of network layers of the model is deepened by a specific propagation mechanism, so that the algorithm can learn the information of the nodes more comprehensively and accurately.

[0093] Figure 7 This is a schematic diagram of a diagram conversion according to an embodiment of this application, such as... Figure 7 As shown, the initial graph input includes user node features, influence labels, and connections between users. A personalized PageRank propagation mechanism is needed to update the information in the initial graph input, resulting in the transformed graph data structure.

[0094] Suppose the initial graph has n nodes and m features, and G = (V, E) represents a graph containing a set of nodes V and a set of edges E. X ∈ R n×f Let f represent the node feature matrix, where f represents the number of features for each node. Y∈R n×1 Represents the matrix label. A∈R n ×n The adjoint matrix represents the graph and is used to describe the connectivity relationships within the graph. The adjoint matrix representing a self-loop requires considering its own information when updating the information of each root node, where I... n It is an identity matrix. For example... Figure 7 The dashed line portion of the self-connected node A.

[0095] for Figure 7 For the root node A to be updated, it will update its node feature X with a probability of (1-α). A The information is passed to nodes B, C, and D, and simultaneously returned to itself with probability α. Nodes B, C, and D will also pass the information to root node A with probability α. Therefore, for root node A, the updated information feature expression is as follows: Where, π ppr (X A ) represents the updated information characteristics of root node A.

[0096] In step S405, after information propagation through each layer, the data enters the graph neural network for the next feature learning iteration until a new node representation is obtained. Based on the new node representation, it is input into the fully connected layer to learn the network weight parameters.

[0097] Figure 8 This is a flowchart of a graph neural network algorithm according to an embodiment of this application, such as... Figure 8 As shown, the method includes:

[0098] Step S801, Node Representation Initialization: Initialize the feature representation for each node in the graph, assuming h v (l) This represents the feature representation of node v at layer l.

[0099] Step S802, Neighbor Information Aggregation: For each node to be updated, information is aggregated from neighboring nodes and calculated using a statistical aggregation function. Let the aggregated information for node v be represented as U. v (l) ,but: Where N(v) is the set of neighboring nodes of v, h μ (l) Let g(·) represent the feature representation of node μ at layer l. g(·) represents the neighbor information aggregation function. This indicates an aggregation operation.

[0100] Step S803, Node State Update: Combining the current information of the node and the aggregated information, the state is updated using an update function. Assume the updated state of the current node is h. v (l+1) Then h v (l+1) =f(h) v (l) U v (l) ).

[0101] Step S804, Transform Graph Generation: Integrate the representations of all nodes into a total graph representation using a readout function.

[0102] Step S804: Obtain the prediction result of the node by introducing a fully connected layer and an output layer. Let the prediction result of the current node be y. pred , then y pred =Linear(Relu(h) v (l+1) )).

[0103] Step S406: Based on the learned weights, input them into the output layer to output the magnitude of social influence.

[0104] Figure 9 This is a flowchart of an algorithm for predicting influence using a graph neural network according to an embodiment of this application, such as... Figure 9 As shown, the model construction process is as follows: 1. Data Input: The model inputs user basic information, package information, APP usage, call information, and other dimensional data. 2. Feature Engineering: The input data needs to undergo the following feature engineering processes: removing weakly correlated numerical features; handling missing values ​​and outliers; feature transformation and data binning; feature derivation; data standardization and one-hot encoding. 3. Model Construction: First, the training and test sets are divided in an 80:20 ratio. The training set is used to train the model's performance, and the test set is used to test the model's performance. Then, the PageRank graph neural network algorithm is applied to train the model.

[0105] Figure 10 This is a structural diagram of a data processing apparatus according to an embodiment of this application, such as... Figure 10 As shown, the device includes:

[0106] The first generation module 1001 is used to generate a call network graph based on the calling number, the called number, the number of calls between the calling number and the called number, and the call duration. The nodes in the call network graph are the calling number and the called number, and the edges in the call network graph are the number of calls between the calling number and the called number, and the call duration.

[0107] The determination module 1002 is used to determine the influence label corresponding to the number based on the call network diagram.

[0108] The second generation module 1003 is used to generate a graph network structure based on the user data corresponding to the number, the influence tag corresponding to the number, and the association between the numbers. The nodes in the graph network structure are the user data corresponding to the number and the influence tag corresponding to the number, and the edges in the graph network structure are the association between the numbers.

[0109] The update module 1004 is used to update the graph network structure using the propagation mechanism based on the PageRank algorithm to obtain the target feature representation.

[0110] The processing module 1005 is used to input the target feature representation into the graph neural network model to obtain the influence index corresponding to the number output by the graph neural network model.

[0111] Optionally, the determining module 1002 is further configured to perform the following steps: setting the initial PageRank value of all nodes in the call network graph, wherein when learning information of any node, the node is initialized to 1 and other nodes are initialized to 0, and the PageRank value is used to characterize the influence value corresponding to the number; step S1, for the first node, calculating the number of connections a between the first node and the second node, if the PageRank value of the first node is PR(p), then the PageRank value passed from the first node to each second node is PR(p) / a, wherein the first node and the second node are different nodes; step S2, for the second node, determining the set of nodes pointing to the second node, obtaining the PageRank value passed from the first node to the second node, and updating the PageRank value of the second node according to the PageRank value passed from the first node to the second node; repeating steps S1 and S2 to perform multiple iterative calculations until the preset convergence condition is met to obtain the influence label corresponding to the number, wherein the preset convergence condition includes: the change value of the PageRank value of all nodes between two consecutive iterations is less than a preset threshold.

[0112] Alternatively, the PageRank value of a node in the call network graph can be calculated using the following formula: Among them, pr(p i ) is node p i PageRank value, pr(p j ) is node p j PageRank value, T ij It is node p i Received node p j Number of calls, L j It is node p j The number of outgoing calls, q is the preset damping coefficient, and N is the number of nodes in the call network diagram.

[0113] Optionally, the update module 1004 is further configured to perform the following steps: constructing an adjoint matrix of a graph network structure based on the association between numbers, wherein if there are n nodes in the graph network structure and there is an edge connection between node i and node j, then A in the adjoint matrix... ijElements are assigned non-zero values ​​based on the weights of edge connections, where i and j are positive integers and n is a positive integer greater than 1; a node feature matrix representing the features of graph network nodes is received, where the rows of the node feature matrix are the feature vectors corresponding to the nodes; the adjoint matrix is ​​normalized to obtain the target adjoint matrix, and the linear combination coefficient matrix of the identity matrix and the target adjoint matrix is ​​calculated; based on the coefficient matrix and the node feature matrix, iterative calculation is performed. In each iteration, node feature information is propagated and fused along the edge structure of the graph network according to the PageRank propagation mechanism, so that each node obtains a comprehensive feature representation from its neighboring nodes and its own historical feature information; if the iteration process meets the preset convergence judgment condition, the node feature matrix is ​​used as the target feature representation after the graph network structure is updated.

[0114] Optionally, the processing module 1005 is further configured to perform the following steps: using the target feature representation as the initial feature representation of the nodes in the graph neural network model; determining the set of neighboring nodes corresponding to each node in the graph neural network model, and for each node, performing an aggregation operation on the feature representations of its neighboring nodes through a preset information aggregation function to obtain the aggregated features of the neighboring nodes; updating the initial feature representation of each node through a preset update function based on the aggregated features of the neighboring nodes and the current feature representation of each node itself, and constructing a transformation graph based on the updated feature representation of the nodes; inputting the node feature vectors in the transformation graph into a fully connected layer for feature transformation, and inputting the feature transformation result into an output layer to obtain the influence index output by the output layer.

[0115] Optionally, the data processing device is further configured to perform the following steps before generating the graph network structure based on the user data corresponding to the number, the influence tag corresponding to the number, and the association between the numbers: obtaining the user data corresponding to the number; scanning for missing values ​​for each feature in the user data, and if the value of the target feature in one or more samples is a preset missing identifier, then recording that the target feature has a missing value and determining the target sample where the missing value is located; calculating the mean corresponding to all non-missing samples in the target sample, and replacing the missing value with the mean; calculating the mean and standard deviation of each numerical feature in the user data, setting an outlier judgment threshold, and determining that the feature value is an outlier if the feature value of the target data is greater than the outlier judgment threshold; correcting the outlier using a preset Gaussian mixture model; standardizing the corrected data, and performing one-hot encoding on the standardized data.

[0116] Optionally, user data includes at least one of the following: business object domain data, deep packet inspection internet access data, and social circle data, wherein social circle data includes: contact list data and social relationship attribute information.

[0117] It should be noted that the above Figure 10 The modules in can be program modules (e.g., a set of program instructions that implement a specific function) or hardware modules. For the latter, they can be represented in the following forms, but are not limited to these: each of the above modules is represented by a processor, or the functions of each of the above modules are implemented by a processor.

[0118] It should be noted that, Figure 10 Preferred embodiments of the shown examples can be found in [reference needed]. Figure 1 The relevant descriptions of the embodiments shown will not be repeated here.

[0119] Figure 11 A hardware structure block diagram of a computer terminal for implementing a data processing method is shown. Figure 11 As shown, the computer terminal 110 may include one or more processors 1102 (shown as 1102a, 1102b, ..., 1102n in the figure) (processor 1102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 1104 for storing data, and a transmission module 1106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 11 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, the computer terminal 110 may also include... Figure 11 The more or fewer components shown, or having the same Figure 11 The different configurations shown.

[0120] It should be noted that the aforementioned one or more processors 1102 and / or other data processing circuitry are generally referred to herein as "data processing circuitry". This data processing circuitry may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuitry may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 110. As involved in the embodiments of this application, this data processing circuitry serves as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0121] The memory 1104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the data processing method in this embodiment. The processor 1102 executes various functional applications and data processing by running the software programs and modules stored in the memory 1104, thereby implementing the aforementioned data processing method. The memory 1104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 1104 may further include memory remotely located relative to the processor 1102, and these remote memories can be connected to the computer terminal 110 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0122] The transmission module 1106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 110. In one example, the transmission module 1106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission module 1106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0123] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 110.

[0124] It should be noted here that, in some optional embodiments, the above... Figure 11 The computer terminal shown may include hardware elements (including circuitry), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware and software elements. It should be noted that... Figure 11 This is only one instance of a specific particular instance, and is intended to illustrate the types of components that may exist in the aforementioned computer terminal.

[0125] It should be noted that, Figure 11 The computer terminal shown is used to execute Figure 1 The data processing method shown above is also applicable to this electronic device, and will not be repeated here.

[0126] This application also provides a non-volatile storage medium, which includes a stored program, wherein the program controls the device where the storage medium is located to execute the above data processing method when it runs.

[0127] A program on a non-volatile storage medium performs the following functions: It generates a call network graph based on the calling number, the called number, the number of calls between the calling and called numbers, and the call duration. The nodes in the call network graph are the calling and called numbers, and the edges are the number of calls between the calling and called numbers and the call duration. Based on the call network graph, it determines the influence label corresponding to each number. Based on the user data corresponding to each number, the influence label corresponding to each number, and the relationships between numbers, it generates a graph network structure. The nodes in the graph network structure are the user data corresponding to each number and the influence label corresponding to each number, and the edges are the relationships between numbers. It updates the graph network structure using a propagation mechanism based on the PageRank algorithm to obtain the target feature representation. Finally, it inputs the target feature representation into a graph neural network model to obtain the influence index corresponding to the number output by the graph neural network model.

[0128] This application also provides an electronic device, including a memory and a processor, wherein the processor is used to run a program stored in the memory, wherein the program executes the above-described data processing method during runtime.

[0129] The processor runs a program that performs the following functions: It generates a call network graph based on the calling number, the called number, the number of calls between the calling and called numbers, and the call duration. The nodes in the call network graph are the calling and called numbers, and the edges are the number of calls between the calling and called numbers and the call duration. Based on the call network graph, it determines the influence label corresponding to each number. Based on the user data corresponding to each number, the influence label corresponding to each number, and the relationships between numbers, it generates a graph network structure. The nodes in the graph network structure are the user data corresponding to each number and the influence label corresponding to each number, and the edges are the relationships between numbers. It updates the graph network structure using a propagation mechanism based on the PageRank algorithm to obtain the target feature representation. Finally, it inputs the target feature representation into a graph neural network model to obtain the influence index corresponding to the number output by the graph neural network model.

[0130] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0131] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0132] In the above embodiments of this application, the information collected is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with relevant laws, regulations and standards, take necessary protective measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0133] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0134] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0135] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0136] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to related technologies, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0137] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A data processing method, characterized in that: include: Generate a call network graph according to the calling number, the called number, the number of calls and the call duration between the calling number and the called number, wherein the nodes in the call network graph are the calling number and the called number, and the edges in the call network graph are the number of calls and the call duration between the calling number and the called number; Determine the influence label corresponding to the number according to the call network diagram; Generate a graph network structure according to the user data corresponding to the number, the influence label corresponding to the number, and the association relationship between the numbers, wherein the nodes in the graph network structure are the user data corresponding to the number and the influence label corresponding to the number, and the edges in the graph network structure are the association relationship between the numbers; Using a propagation mechanism based on the PageRank algorithm, the graph network structure is updated to obtain a target feature representation; The target feature representation is input into a graph neural network model to obtain an influence index corresponding to the number output by the graph neural network model.

2. The method according to claim 1, characterized in that Determining the influence label corresponding to the number according to the call network diagram includes: Setting the initial PageRank values ​​of all nodes in the call network graph, wherein when learning the information of any node, the node is initialized to 1, and other nodes are initialized to 0, and the PageRank value is used to represent the influence value corresponding to the number; Step S1, for a first node, calculate the number of connections a between the first node and the second node, if the PageRank value of the first node is PR(p), then the PageRank value transmitted by the first node to each of the second nodes is PR(p) / a, wherein the first node and the second node are different nodes; Step S2, for the second node, determine a node set pointing to the second node, obtain the PageRank value passed by the first node to the second node, and update the PageRank value of the second node according to the PageRank value passed by the first node to the second node; Repeat steps S1 and S2 to perform multiple iterative calculations until a preset convergence condition is met to obtain an influence label corresponding to the number, wherein the preset convergence condition includes: the PageRank value change values ​​of all nodes between two consecutive iterations are less than a preset threshold.

3. The method according to claim 2, characterized in that The PageRank value of a node in the call network graph is calculated using the following formula: Among them, pr(p i ) is the node p i The PageRank value of pr(p j ) is the node p j PageRank value, T ij is node p i Receive node p j The number of calls, L j is node p j The number of calls, q is the preset damping coefficient, and N is the number of nodes in the call network diagram.

4. The method according to claim 1, characterized in that The propagation mechanism based on the PageRank algorithm is used to update the graph network structure and obtain the target feature representation, including: According to the association relationship between the numbers, the adjoint matrix of the graph network structure is constructed, wherein, if there are n nodes in the graph network structure, and there is an edge connection between node i and node j, then A in the adjoint matrix ij The element is set to a corresponding non-zero value according to the weight of the edge connection, where i, j are positive integers and n is a positive integer greater than 1; Receiving a node feature matrix representing features of a graph network node, wherein the node feature matrix is ​​a feature vector corresponding to a row node; Normalizing the adjoint matrix to obtain a target adjoint matrix, and calculating a linear combination coefficient matrix of a unit matrix and the target adjoint matrix; Based on the coefficient matrix and the node feature matrix, iterative calculation is performed. In each iteration, the node feature information is propagated and integrated along the edge structure of the graph network according to the PageRank propagation mechanism, so that each node obtains a comprehensive feature representation from neighbor nodes and its own historical feature information; When the iterative process meets the preset convergence judgment condition, the node feature matrix is ​​used as the target feature representation after the graph network structure is updated.

5. The method according to claim 1, characterized in that Inputting the target feature representation into the graph neural network model to obtain the influence index corresponding to the number output by the graph neural network model includes: Using the target feature representation as an initial feature representation of a node in the graph neural network model; Determine a set of neighbor nodes corresponding to each node in the graph neural network model, and for each node, perform an aggregation operation on feature representations of its neighbor nodes through a preset information aggregation function to obtain neighbor node aggregation features; Based on the aggregated features of the neighboring nodes and the current feature representation of each node itself, the initial feature representation of each node is updated by a preset update function, and a transformation graph is constructed according to the updated feature representation of the node; The node feature vectors in the conversion graph are input into the fully connected layer for feature transformation, and the feature transformation results are input into the output layer to obtain the influence index output by the output layer.

6. The method according to claim 1, characterized in that Before generating the graph network structure according to the user data corresponding to the number, the influence label corresponding to the number and the association relationship between the numbers, the method further includes: Get the user data corresponding to the number; Scan each feature in the user data for missing values. If the value of the target feature in one or more samples is a preset missing flag, record that the target feature has a missing value, and determine the target sample where the missing value is located; calculate the mean corresponding to all non-missing samples in the target sample, and replace the missing value with the mean; Calculate the mean and standard deviation of each numerical feature in the user data, set an outlier judgment threshold, and if the feature value of the target data is greater than the outlier judgment threshold, determine that the feature value is an outlier; and use a preset Gaussian mixture model to correct the outlier; The corrected data is standardized and the standardized data is one-hot encoded.

7. The method according to claim 6, characterized in that The user data includes at least one of the following: business object domain data, deep packet inspection Internet access data, and social circle data, wherein the social circle data includes: contact list data and social relationship attribute information.

8. A data processing device, characterized in that: include: A first generating module, configured to generate a call network graph according to a calling number, a called number, and the number of calls and the call duration between the calling number and the called number, wherein the nodes in the call network graph are the calling number and the called number, and the edges in the call network graph are the number of calls and the call duration between the calling number and the called number; A determination module, used to determine the influence label corresponding to the number according to the call network diagram; A second generating module is used to generate a graph network structure according to user data corresponding to the number, influence labels corresponding to the number, and associations between numbers, wherein nodes in the graph network structure are user data corresponding to the number and influence labels corresponding to the number, and edges in the graph network structure are associations between numbers; An updating module, used to update the graph network structure by using a propagation mechanism based on the PageRank algorithm to obtain a target feature representation; A processing module is used to input the target feature representation into a graph neural network model to obtain an influence index corresponding to the number output by the graph neural network model.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is executed, the device where the non-volatile storage medium is located is controlled to execute the data processing method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is used to run a program stored in the memory, wherein the program executes the data processing method according to any one of claims 1 to 7 when running.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the data processing method according to any one of claims 1 to 7 is implemented.

Images