A method and system for classifying special personnel characteristic information based on graph neural network

Abstract

The present invention relates to a method and system for classifying characteristic information of special personnel based on graph neural networks, comprising: obtaining a special personnel behavior data set and special personnel characteristic information; performing data preprocessing based on the obtained special personnel behavior data set and special personnel characteristic information, including supplementing missing values ​​in the data set by using a median supplementation method; performing normalization operations on the numerical data of the data set; and performing numerical processing on the categorical data of the data set. The present invention provides a method for classifying characteristic information of special personnel based on graph neural networks, proposes a method for constructing a graph network based on the similarity between special personnel individuals, extracting graph network features using complex network analysis, and on this basis, aggregating neighbor features through graph neural networks to evaluate the behavior of special personnel.

Description

Technical Field

[0001] The present invention relates to the field of regulatory technology, and in particular to a method and system for classifying characteristic information of special personnel based on a graph neural network. Background Art

[0002] The special population is a complex, dynamic system with distinct characteristics. Traditional theories struggle to accurately describe and predict the evolution of its behavioral and psychological traits. The integration of big data information science and traditional social sciences can provide a new theoretical framework and tools for the classification of characteristic information. With the advent of the big data era, a growing amount of information generated in management is being accumulated and stored, including basic indicators, activity records, performance, diagnostic data, and various behavioral data related to special individuals. This vast amount of data is closely related to the daily performance and behavior of special individuals. Data of varying types and time periods reflect the individual characteristics of special individuals, making this data of great research value and analytical significance.

[0003] At present, most of the feature information classification work in China is based on traditional paper-based tables, which require manual data entry. There are problems such as low efficiency, poor objectivity, poor data compatibility, and insufficient data utilization. Specifically, it manifests itself in the following two aspects: (1) Evaluating the behavior of special personnel through traditional paper-based tables will lead to insufficient data utilization and low accuracy. In the process of effectively quantifying the behavior of special personnel, the existing method takes all the information of special personnel individuals into consideration, which not only brings a lot of data redundancy but also ignores a lot of important information; (2) Extracting information independently from special personnel individuals has limitations in exploring information, ignores the interactions and connections between subjects in the crowd, and cannot guarantee the effectiveness of the evaluation work.

[0004] Therefore, how to efficiently and reasonably classify the characteristic information of special personnel's behavior is an important issue to be solved now. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a method and system for classifying characteristic information of special personnel based on graph neural networks. By extracting features from data and introducing the similarities between individuals to construct a graph network, the graph network features are extracted using complex network analysis. On this basis, the neighbor features are aggregated through the graph neural network to obtain the behavior of special personnel, which solves the problems of insufficient utilization of model data and ignoring the association between subjects in the crowd, and improves the accuracy of the classification model.

[0006] Explanation of terms:

[0007] Graph Neural Network (GNN): A concept proposed by Gori et al. in 2005. It is a general term for algorithms that use neural networks to learn from graph-structured data, extracting and discovering features and patterns within it, and meeting the requirements of graph learning tasks such as clustering, classification, prediction, segmentation, and generation. GNNs optimize all graph attributes while preserving their symmetry. In other words, GNNs only modify the graph's attributes, not its structure.

[0008] Graph Convolution Network (GCN): GCN can be regarded as the "pioneering work" of graph neural networks. For the first time, it simply applied the convolution operation in image processing to graph structure data processing and gave a specific derivation. Its methods are divided into two categories, spectral domain-based methods and spatial domain-based methods. The spectral domain-based method defines graph convolution by introducing filters from the perspective of graph signal processing, where the graph convolution operation is interpreted as removing noise from the graph signal. The spatial domain-based method represents graph convolution as aggregating feature information from neighbors.

[0009] The technical solution of the present invention is:

[0010] A special personnel feature information classification method based on graph neural network, including:

[0011] Obtain a dataset containing information on special personnel characteristics;

[0012] Based on the acquired dataset containing the characteristic information of special personnel, data preprocessing is performed, including: using the median method to supplement the missing values ​​in the dataset; normalizing the numerical data in the dataset; digitizing the categorical data in the dataset; and dividing the characteristic information of special personnel into basic information of special personnel and risk information of special personnel;

[0013] Construct a special personnel graph network based on the pre-processed special personnel feature information;

[0014] Based on the special personnel graph network, the complex network is used to extract network features, including: using the complex network to extract the network features of the special personnel graph network, the basic features of special personnel, and the risk features of special personnel;

[0015] Based on the extracted network features, the graph neural network is used to aggregate the neighbor features in the network features. The neighbor features are the basic features of special personnel on the nodes adjacent to the target node, the risk features of special personnel, and the network features of the special personnel graph network;

[0016] According to the neighbor features aggregated by the graph neural network, the Softmax function is used to obtain the probability that each node in the graph neural network belongs to each category, and the node classification is completed. According to the classification results, the classification level of the special personnel feature information is obtained.

[0017] Furthermore, in the data preprocessing, a normalization operation is performed on the numerical data of the data set, including using a linear function to normalize the data set and perform a linear transformation so that each numerical data is mapped to the interval [0,1]. The normalization operation formula is:

[0018]

[0019] Among them, X represents the original data; X min 、X max Respectively represent the minimum and maximum values ​​of the data; X norm is the normalized result.

[0020] Furthermore, in the data preprocessing, the categorical data of the data set is digitized, including digitizing the categorical text information in the data set, and uniformly label encoding the discrete fields, the empty value in each discrete field is represented by 0, and all attribute values ​​under each discrete field are encoded in sequence.

[0021] Furthermore, the special personnel graph network is constructed based on the preprocessed special personnel behavior dataset, including taking special personnel as nodes, classification levels as node labels, capturing similarities between special personnel individuals as edges, and constructing the special personnel graph network.

[0022] Furthermore, the method extracts network features based on the special personnel graph network using a complex network, including obtaining the degree centrality, core number, closeness centrality, betweenness centrality and eigenvector centrality of nodes in the special personnel graph network using a node centrality identification algorithm.

[0023] Furthermore, the obtained network features are used to aggregate neighbor features using the graph convolutional neural network model in the graph neural network, specifically including calculating the convolution results of the adjacent nodes of the central node in the special personnel graph network and the convolution kernel by training the convolution kernel coefficient of the graph convolutional neural network model, and then multiplying the result with the feature matrix and the activation function to perform a nonlinear transformation to obtain a matrix of aggregated adjacent node features, thereby realizing the aggregation of neighbor features. The convolution kernel g used in the graph convolutional neural network model is θ for:

[0024]

[0025] in, is a first-order adjacency matrix, is the degree matrix, and θ is the parameter matrix.

[0026] Furthermore, the Softmax function is used to obtain the probability that each node in the graph neural network belongs to each category based on the neighbor features aggregated by the graph neural network, including obtaining the node feature vector using the convolutional layer of the graph neural network, and then passing it through the Softmax function to obtain the probability that each node belongs to each category, thereby completing the node classification.

[0027] A special personnel feature information classification system based on graph neural network, including:

[0028] The data acquisition module is configured to acquire a special person behavior data set and special person feature information;

[0029] The preprocessing module is configured to perform data preprocessing based on the acquired special personnel behavior data set and special personnel characteristic information, including supplementing missing values ​​in the data set using the median method; normalizing the numerical data in the data set; digitizing the categorical data in the data set; and dividing the special personnel characteristic information into special personnel basic information and special personnel risk information;

[0030] The graph network module is configured to construct a special person graph network based on the preprocessed special person behavior dataset and special person feature information;

[0031] The network feature module is configured to extract network features based on the special personnel graph network using a complex network, including extracting network features of the special personnel graph network, basic features of special personnel, and risk features of special personnel using the complex network;

[0032] A neighbor feature module is configured to aggregate neighbor features in the network features using a graph neural network based on the extracted network features, wherein the neighbor features are basic features of special personnel, risk features of special personnel, and network features of the special personnel graph network on nodes adjacent to the target node;

[0033] The computing module is configured to use the Softmax function to obtain the probability that each node in the graph neural network belongs to each category based on the neighbor features aggregated by the graph neural network, complete the node classification, and obtain the special personnel classification level based on the classification results.

[0034] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, a method for classifying characteristic information of special personnel based on a graph neural network.

[0035] A terminal device includes a processor and a computer-readable storage medium, the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded and executed by the processor to implement a special personnel feature information classification method based on a graph neural network.

[0036] The beneficial effects of the present invention are:

[0037] Compared to existing technologies, this invention provides a method for classifying characteristic information of special personnel based on graph neural networks. It proposes constructing a graph network based on the similarities between individual special personnel, extracting graph network features using complex network analysis, and then evaluating special personnel behavior by aggregating neighbor features through a graph neural network. This reduces computational and model complexity while improving the accuracy and effectiveness of evaluation and prediction. This method meets the needs of special personnel management and improves efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings in the specification, which constitute a part of this application, are used to provide further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute improper limitations on this application.

[0039] Figure 1 is a schematic diagram of a special personnel graph network structure according to an embodiment of the present invention;

[0040] Figure 2 A method for classifying special personnel feature information based on a graph neural network according to an embodiment of the present invention

[0041] Figure 3 2 is a schematic diagram of a graph neural network structure according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The present invention will be further defined below with reference to the accompanying drawings and embodiments, but is not limited thereto.

[0043] Example 1

[0044] like Figure 1 As shown, a special personnel feature information classification method based on graph neural network includes:

[0045] Step 1, data preprocessing:

[0046] 1) Data cleaning: The original data of the special personnel behavior data set contains 44 features, and almost all of the data have missing data. The common methods for processing missing values ​​include deleting columns, deleting rows, and supplementing values. The supplementation of values ​​includes supplementing 0, median, and mean, etc. The present invention sets the missing processing threshold to 70%, that is, the data with more than 70% missing data in the row or column is deleted. Then, the missing values ​​of the continuous features in the retained data are supplemented by the median, and the missing values ​​of the categorical features are set to the "other" category. After data screening, 28 features are retained, and the "classification level" feature is selected as the label, and the special personnel feature information is divided into special personnel basic information and special personnel risk information;

[0047] The basic information and risk information of special personnel are shown in Table 1:

[0048] Table 1

[0049]

[0050]

[0051] 2) Processing numerical data:

[0052] Use the linear function normalization method to perform a linear transformation on the original data so that each numerical data is mapped to the [0,1] interval, which will scale the original data proportionally. The normalization operation formula is:

[0053]

[0054] X represents the original data; X min 、X max Respectively represent the minimum and maximum values ​​of the data; X norm is the result after normalization; after normalization, each numerical data in the data set is mapped to the [0,1] interval.

[0055] 3) Processing categorical data:

[0056] Since the formats of various indicators are not uniform, it is necessary to digitize the categorical text information and uniformly label all discrete fields. The null value in each discrete field is represented by 0, and all attribute values ​​under each discrete field are coded starting from 1, 2, 3, etc. After digitization, all attribute values ​​under each discrete field in the dataset are converted into numerical codes, among which low, moderate, and high risks under the "classification level" field are coded as 1, 2, and 3, respectively.

[0057] Step 2: Build a graph network based on the similarity between special people:

[0058] A graph network is a data structure somewhat similar to a tree, consisting of nodes and edges. A graph can be represented as G = (V, E), where V is the set of nodes and E is the set of edges. Graphs can be categorized as directed or undirected based on whether their edges have directions, and as weighted or unweighted based on whether their edges have weights.

[0059] In the present invention, special personnel are used as nodes, which include the basic characteristics and risk characteristics of special personnel, and the classification level is used as the node label. The similarity between individuals is captured as the edge to construct the special personnel graph structure.

[0060] Specifically, the constructed graph structure is an undirected graph G = (V, E), which has n nodes. The edges connecting the nodes are The dimension of the feature vector contained in each node is C. Given the initial feature matrix Where c (c = 1, ..., C) represents features of different dimensions. The n × n dimensional adjacency matrix A formed by the edges E between each node V is defined as:

[0061]

[0062] Among them, f c and γ c are the similarity measure and distance measure of the c-th feature respectively.

[0063] The similarity measure f is defined as:

[0064]

[0065] Among them, ρ is the correlation distance, σ is the kernel width of the Gaussian function, and e ij =f(x i ,x j );

[0066] The definition of the distance metric γ depends on the feature type. For categorical information, γ is defined as the Kroneckerdelta function δ. For numerical data, γ is defined as a unit step function determined by a threshold θ:

[0067]

[0068] Step 3: Extract graph network features using complex network analysis:

[0069] The extraction of graph network features is based on complex network analysis, and the node centrality identification algorithm used is as follows:

[0070] 1) Degree centrality: Let G = (V, E), an undirected graph network, with N and M nodes and edges respectively. The adjacency matrix is ​​A = {a ij}, node vi The degree is denoted as k i , then the degree centrality of a node is defined as:

[0071]

[0072] 2) Number of cores: The number of cores is a metric that describes the diffusion capacity of a node in a network. The number of cores can be obtained by performing a k-core decomposition of the node. In the k-core decomposition, k is the order of the largest core. For a network, core 0 is the original graph; core 1 is the graph with all isolated points removed; core 2 is the graph that first removes all points with a degree less than 2, then removes all points with a degree less than 2 from the remaining graph; core 3 is the graph that first removes all points with a degree less than 3, then removes all points with a degree less than 3 from the remaining graph, and so on, until no more points can be removed. The order k of the largest core is the number of cores for a node.

[0073] 3) Proximity centrality: Proximity centrality measures the average distance between a node and other nodes, reflecting the degree of proximity between nodes in the network. ij For node v i to v j The shortest path length is , then the closeness centrality is defined as:

[0074]

[0075] 4) Betweenness Centrality: Betweenness centrality is a measure of a node's criticality, using the number of shortest paths passing through a fixed node. It describes the node's ability to manage information in the network. Betweenness centrality is defined as:

[0076]

[0077] Among them, g st and They are node v s , v t The number of shortest paths between and through v s , v t The shortest path through node v i The number of.

[0078] 5) Eigenvector centrality: Eigenvector centrality is an extension of degree centrality. i The eigenvector centrality of x i Defined as:

[0079]

[0080] Among them, c is a constant, a ij is the adjacency matrix, x j is node v j The eigenvector centrality value of .

[0081] Among the eigenvector centrality algorithms, PageRank centrality is the most classic one. The PageRank value is randomly wandered on the network constructed by web pages, and updated iteratively until it stabilizes. is node v j The out-degree of v i The PageRank value at step t is:

[0082]

[0083] Step 4: Aggregate neighbor features using graph neural networks:

[0084] Brunna et al. proposed a graph Laplace method based on the spectral domain, which extended the convolutional neural network to non-Euclidean data. This paper adopts the graph convolutional neural network model in the graph neural network and uses the first-order graph convolutional neural network model based on the spectral domain proposed by Kipf et al. Figure 3 As shown in the following diagram, neighbor features are aggregated based on the obtained network features of the special personnel graph network and the preprocessed basic and risk features of special personnel. When constructing the graph network, nodes only contain the basic and risk features of special personnel. Now, the obtained graph network features, namely degree centrality, core number, closeness centrality, betweenness centrality, and eigenvector centrality, are also added to it, resulting in an N×M-dimensional feature matrix X, where N is the number of nodes and M is the dimension of the eigenvector contained in each node. The propagation method between GCN layers is as follows:

[0085] H (l+1) =σH (l) g θ (10)

[0086] Among them, σ is a nonlinear activation function, H (l) is the feature of the lth layer, H in the input layer (0) =X, is the convolution kernel used in the graph convolutional neural network model, is a first-order adjacency matrix, is the degree matrix, The convolution kernel is a parameter matrix. It is a weighted sum of local information, corresponding to local perception. The principle is that when observing an object, you cannot observe every pixel or the whole object at once, but start from the local information, which corresponds to convolution. The role of the convolution kernel is to extract features. By training the convolution kernel coefficients, the convolution result of the adjacent nodes of the central node and the convolution kernel is calculated, and then combined with the feature matrix H. (l) Multiplying it with the activation function σ and performing a nonlinear transformation yields the matrix H of the aggregated adjacent node features. (l+1) , thereby achieving the aggregation of neighbor features.

[0087] Step 5: Output the special personnel classification results.

[0088] The goal of a spectral domain-based graph convolutional neural network is to learn a function f on graph data. After inputting the adjacency matrix A and the feature matrix X, the probability of each node belonging to each category is obtained through the output network model Z = f(X, A). The loss function is used to measure the quality of the model's predictions and is calculated by the cross-entropy error of all labeled examples. Assuming that the GCN has only two layers, the specific output network model and loss function used in this invention are as follows:

[0089]

[0090]

[0091] in, F is the number of convolution kernels, W (0) , W (1) are the weight matrices from the input layer to the hidden layer and from the hidden layer to the output layer respectively; the softmax activation function is defined as yL is a set of labeled nodes. Represents the features extracted from each node after passing through a GCN layer. The features extracted from each node after the second GCN layer are passed through the Softmax function to obtain the probability that each node belongs to each category, complete the node classification, and obtain the special personnel classification level.

[0092] Example 2

[0093] A special personnel feature information classification system based on graph neural network, including:

[0094] The data acquisition module is configured to acquire a special person behavior data set and special person feature information;

[0095] The preprocessing module is configured to perform data preprocessing based on the acquired special personnel behavior data set and special personnel characteristic information, including supplementing missing values ​​in the data set using the median method; normalizing the numerical data in the data set; digitizing the categorical data in the data set; and dividing the special personnel characteristic information into special personnel basic information and special personnel risk information;

[0096] The graph network module is configured to construct a special person graph network based on the preprocessed special person behavior dataset and special person feature information;

[0097] The network feature module is configured to extract network features based on the special personnel graph network using a complex network, including extracting network features of the special personnel graph network, basic features of special personnel, and risk features of special personnel using the complex network;

[0098] A neighbor feature module is configured to aggregate neighbor features in the network features using a graph neural network based on the extracted network features, wherein the neighbor features are basic features of special personnel, risk features of special personnel, and network features of the special personnel graph network on nodes adjacent to the target node;

[0099] The computing module is configured to use the Softmax function to obtain the probability that each node in the graph neural network belongs to each category based on the neighbor features aggregated by the graph neural network, complete the node classification, and obtain the special personnel classification level based on the classification results.

[0100] Example 3

[0101] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device to provide a method for classifying characteristic information of special personnel based on a graph neural network provided in this embodiment.

[0102] Example 4

[0103] A terminal device includes a processor and a computer-readable storage medium, the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded and executed by the processor. A special personnel feature information classification method based on graph neural network provided in this embodiment.

[0104] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0105] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0106] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0107] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0108] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

[0109] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A special personnel feature information classification method based on graph neural network, characterized by: include: Obtain a dataset containing information on special personnel characteristics; Based on the acquired dataset containing special personnel characteristic information, data preprocessing is performed, including data supplementation using the median method for missing values ​​in the dataset; normalization of the numerical data in the dataset; and numerical processing of the categorical data in the dataset; And divide the characteristic information of special personnel into basic information of special personnel and risk information of special personnel; Construct a special personnel graph network based on the pre-processed special personnel feature information; Based on the special personnel graph network, the complex network is used to extract network features, including using the complex network to extract the network features of the special personnel graph network, the basic features of special personnel, and the risk features of special personnel; Based on the extracted network features, the graph neural network is used to aggregate the neighbor features in the network features. The neighbor features are the basic features of special personnel on the nodes adjacent to the target node, the risk features of special personnel, and the network features of the special personnel graph network; Based on the neighbor features aggregated by the graph neural network, the Softmax function is used to obtain the probability that each node in the graph neural network belongs to each category, and the node classification is completed. The classification level of special personnel is obtained according to the classification results; The special personnel graph network is constructed based on the preprocessed special personnel behavior data set, including taking special personnel as nodes, classification levels as node labels, capturing similarities between special personnel individuals as edges, and constructing the special personnel graph network.

2. The special personnel feature information classification based on graph neural network according to claim 1, characterized in that: In the data preprocessing, the numerical data of the data set is normalized, including using a linear function to normalize the data set and perform a linear transformation so that each numerical data is mapped to the interval [0, 1]. The normalization operation formula is: Where, · represents the original data; X min 、· max Respectively represent the minimum and maximum values ​​of the data; X norm is the normalized result.

3. The special personnel feature information classification based on graph neural network according to claim 1, characterized in that: In the data preprocessing, the categorical data of the data set is digitized, including digitizing the categorical text information in the data set, and uniformly label encoding the discrete fields. The empty value in each discrete field is represented by 0, and all attribute values ​​under each discrete field are encoded in sequence.

4. The special personnel feature information classification based on graph neural network according to claim 1, characterized in that: The method uses a complex network to extract network features based on the special personnel graph network, including using a node centrality identification algorithm to obtain the degree centrality, core number, closeness centrality, betweenness centrality and eigenvector centrality of nodes in the special personnel graph network.

5. The special personnel feature information classification based on graph neural network according to claim 4 is characterized in that: According to the obtained network features, the graph convolutional neural network model in the graph neural network is used to aggregate neighbor features, specifically including calculating the convolution result of the adjacent nodes of the central node in the special personnel graph network and the convolution kernel by training the convolution kernel coefficient of the graph convolutional neural network model, and then multiplying it with the feature matrix and the activation function to perform a nonlinear transformation to obtain a matrix of aggregated adjacent node features, thereby realizing the aggregation of neighbor features. Among them, the convolution kernel g used in the graph convolutional neural network model is θ for: in, is a first-order adjacency matrix, is the degree matrix, and θ is the parameter matrix.

6. The special personnel feature information classification based on graph neural network according to claim 5, characterized in that: According to the neighbor features aggregated by the graph neural network, the Softmax function is used to obtain the probability that each node in the graph neural network belongs to each category, including using the convolutional layer of the graph neural network to obtain the node feature vector, and then passing it through the Softmax function to obtain the probability that each node belongs to each category, thereby completing the node classification.

7. A special personnel feature information classification system based on graph neural network, characterized by: include: The data acquisition module is configured to acquire a special person behavior data set and special person feature information; The preprocessing module is configured to perform data preprocessing based on the acquired special personnel behavior data set and special personnel characteristic information, including data supplementation using the median method for missing values ​​in the data set; normalization of the numerical data in the data set; and numerical processing of the categorical data in the data set; And divide the characteristic information of special personnel into basic information of special personnel and risk information of special personnel; The graph network module is configured to construct a special person graph network based on the preprocessed special person behavior dataset and special person feature information; The network feature module is configured to extract network features based on the special personnel graph network using a complex network, including extracting network features of the special personnel graph network, basic features of special personnel, and risk features of special personnel using the complex network; A neighbor feature module is configured to aggregate neighbor features in the network features using a graph neural network based on the extracted network features, wherein the neighbor features are basic features of special personnel, risk features of special personnel, and network features of the special personnel graph network on nodes adjacent to the target node; The calculation module is configured to use the Softmax function to obtain the probability that each node in the graph neural network belongs to each category based on the neighbor features aggregated by the graph neural network, complete the node classification, and obtain the special personnel classification level based on the classification results; The special personnel graph network is constructed based on the preprocessed special personnel behavior data set, including taking special personnel as nodes, classification levels as node labels, capturing similarities between special personnel individuals as edges, and constructing the special personnel graph network.

8. A computer-readable storage medium, characterized in that Multiple instructions are stored therein, and the instructions are suitable for being loaded by the processor of the terminal device and executed by a special personnel feature information classification method based on a graph neural network as described in any one of claims 1-6.

9. A terminal device, characterized in that: It includes a processor and a computer-readable storage medium, the processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded and executed by the processor. A special personnel feature information classification method based on a graph neural network according to any one of claims 1-6.

Images