Group classification method and apparatus

Abstract

Embodiments of the present application provide a group classification method and device, relate to the field of data processing, and can also be used in the field of finance. The method comprises the following steps: obtaining training data, determining corresponding nodes and edges according to the training data, and constructing a graph according to the nodes and edges; determining local influence and global influence of the nodes according to the graph, and aggregating the local influence and the global influence to form a super connection graph; inputting the super connection graph into a preset convolutional neural network to obtain a corresponding feature vector, inputting a classification label of the node and the feature vector as a node sample into a preset classification model for model training to obtain a classification model after the model training, and determining a group classification result according to the classification model. The present application can effectively improve the accuracy of user classification.

Description

Technical Field

[0001] This application relates to the field of data processing, and can also be used in the financial field, specifically to a group classification method and apparatus. Background Technology

[0002] With the digitalization of finance, new financial products and services are constantly emerging, and the scale of transaction data is rapidly increasing. Commercial banks generate massive amounts of financial data during their operations. This data contains customer consumption behavior information from many transaction scenarios. This behavioral data includes structured, unstructured, or semi-structured data, forming a large-scale network of data with complexity and irregular characteristics, making information mining difficult. If we use manual labeling to classify hundreds of millions of transaction data into scenarios one by one, there are problems such as inconsistent classification standards, high costs, and low efficiency. Therefore, community classification systems for bank customer behavior have emerged. These systems utilize historical customer behavior data (including static data such as customer assets, income, credit history, and repayment behavior, as well as dynamic behavioral data such as personal customer clothing, food, housing, transportation, entertainment, and wealth management) to deeply mine user behavioral preferences, enabling marketers to conduct marketing based on customer categories. This allows for better understanding of customer needs, meeting potential financial and non-financial service needs, helping banks to identify high-quality customers, and achieving precise marketing.

[0003] In classification systems, the core component is mining user information to extract as much useful information as possible, thereby improving the accuracy of subsequent classification. This core component is commonly referred to as a representation algorithm in academic circles. Therefore, a good representation algorithm can significantly improve the accuracy of classification tasks at downstream nodes.

[0004] The inventors discovered that existing dynamic graph representation learning techniques can learn not only the structural information of the current graph but also its changes over time. Most current graph representation learning methods employ transductive learning, training the entire network to generate node embeddings at time intervals using continuously generated bank transaction data. This model training is time-consuming, leading to a decrease in both accuracy and efficiency. Inductive learning automatically generates node embeddings over time, typically ignoring the influence of the entire network on the relationships between nodes. Summary of the Invention

[0005] To address the problems in the prior art, this application provides a group classification method and apparatus that can effectively improve the accuracy of user classification.

[0006] To solve at least one of the above problems, this application provides the following technical solution:

[0007] Firstly, this application provides a method for group classification, including:

[0008] Obtain the training data, determine the corresponding nodes and edges based on the training data, and construct a graph based on the nodes and edges;

[0009] The local and global effects of the nodes are determined based on the graph, and a hyperconnection graph is formed by aggregating the local and global effects.

[0010] The hyperconnected graph is input into a pre-defined convolutional neural network to obtain the corresponding feature vectors. The classification labels of the nodes and the feature vectors are used as node samples to input into a preset classification model for model training, resulting in a classification model after model training. The group classification result is then determined based on the classification model.

[0011] Furthermore, determining the local influence of the node based on the graph includes:

[0012] Determine the node weights of the neighboring nodes corresponding to the node in the graph;

[0013] The node weights of the neighboring nodes determine the local influence of the node.

[0014] Further, determining the node weights of the neighboring nodes corresponding to the node in the graph includes:

[0015] Determine the current and historical interaction times between nodes and their corresponding neighboring nodes in the graph, and determine the corresponding time influence factors;

[0016] The affinity between the node and its neighboring nodes is determined based on the time influence factor, and the node weight of the neighboring nodes is determined based on the affinity.

[0017] Furthermore, determining the global impact of the node based on the graph includes:

[0018] The activity level of the node is determined by comparing the cumulative value of the affinity within a set time period with the value of a preset threshold.

[0019] The global impact of a node is determined based on its activity level.

[0020] Furthermore, the step of aggregating the local and global effects to form a hyperconnection graph includes:

[0021] The corresponding node information is determined based on the local influence of the node.

[0022] A hyperconnection graph is obtained by establishing a connection between the two nodes in the graph based on the node information and the global influence.

[0023] Further, the step of using the classification label of the node and the feature vector as node samples to input into a preset classification model for model training, to obtain a classification model after model training, includes:

[0024] The classification labels and feature vectors of the nodes are used as node samples to input into a preset classification model for model training.

[0025] The model is optimized based on the preset binary cross-entropy loss function and the training results of the model to obtain the classification model after training.

[0026] Secondly, this application provides a group classification device, comprising:

[0027] The graph construction module is used to acquire the training data, determine the corresponding nodes and edges based on the training data, and construct a graph based on the nodes and edges.

[0028] An aggregation module is used to determine the local and global effects of the nodes based on the graph, and to aggregate the local and global effects to form a hyperconnection graph.

[0029] The model classification module is used to input the hyperconnection graph into a predefined convolutional neural network to obtain the corresponding feature vectors, and to input the node classification labels and the feature vectors as node samples into a preset classification model for model training to obtain a classification model after model training, and to determine the group classification result based on the classification model.

[0030] Furthermore, the aggregation module includes:

[0031] A node weight determination unit is used to determine the node weights of the neighboring nodes corresponding to the node in the graph.

[0032] A local influence determination unit is used to determine the local influence of a node based on the node weights of its neighboring nodes.

[0033] Furthermore, the node weight determination unit includes:

[0034] The influence factor determination subunit is used to determine the current interaction time and historical interaction time between the nodes in the graph and their corresponding neighbor nodes, and to determine the corresponding time influence factor.

[0035] The affinity determination subunit is used to determine the affinity between the node and the neighboring nodes based on the time influence factor, and to determine the node weight of the neighboring nodes based on the affinity.

[0036] Furthermore, the aggregation module also includes:

[0037] An activity determination unit is used to determine the activity of a node based on a comparison between the cumulative value of affinity within a set time period and a preset threshold value.

[0038] A global impact determination unit is used to determine the global impact of a node based on its activity level.

[0039] Furthermore, the aggregation module also includes:

[0040] A node information acquisition unit is used to determine the corresponding node information based on the local influence of the node.

[0041] The hyperlink establishment unit is used to establish a connection based on the node information of two nodes in the graph and the global influence, thereby obtaining a hyperlink graph.

[0042] Furthermore, the model classification module includes:

[0043] The model training unit is used to train the model by inputting the classification label of the node and the feature vector as node samples into a preset classification model.

[0044] The model optimization unit is used to optimize the model based on the preset binary cross-entropy loss function and the model training results to obtain the classification model after the model training.

[0045] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the group classification method.

[0046] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the described group classification method.

[0047] Fifthly, this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the described group classification method.

[0048] As can be seen from the above technical solution, this application provides a group classification method and apparatus. It acquires training data, determines corresponding nodes and edges based on the training data, and constructs a graph based on the nodes and edges. It determines the local and global influences of the nodes based on the graph, and aggregates the local and global influences to form a hyperconnected graph. The hyperconnected graph is input into a pre-defined convolutional neural network to obtain corresponding feature vectors. The classification labels of the nodes and the feature vectors are used as node samples and input into a preset classification model for model training. This yields a classification model after model training, and the group classification result is determined based on the classification model. This effectively improves the accuracy of user classification. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is one of the flowcharts illustrating the group classification method in the embodiments of this application;

[0051] Figure 2 This is a second flowchart illustrating the group classification method in the embodiments of this application;

[0052] Figure 3 This is the third flowchart illustrating the group classification method in the embodiments of this application;

[0053] Figure 4 This is the fourth flowchart illustrating the group classification method in the embodiments of this application;

[0054] Figure 5 This is the fifth flowchart illustrating the group classification method in the embodiments of this application;

[0055] Figure 6 This is the sixth flowchart illustrating the group classification method in the embodiments of this application;

[0056] Figure 7 This is one of the structural diagrams of the group classification device in the embodiments of this application;

[0057] Figure 8 This is the second structural diagram of the group classification device in the embodiments of this application;

[0058] Figure 9 This is the third structural diagram of the group classification device in the embodiments of this application;

[0059] Figure 10 This is the fourth structural diagram of the group classification device in the embodiments of this application;

[0060] Figure 11 This is the fifth structural diagram of the group classification device in the embodiments of this application;

[0061] Figure 12 This is the sixth structural diagram of the group classification device in the embodiments of this application;

[0062] Figure 13 This is a diagram of the classifier structure in a specific embodiment of this application;

[0063] Figure 14 This is a schematic diagram of the structure of the electronic device in the embodiments of this application. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0065] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0066] Considering that existing dynamic graph representation learning technologies can learn not only the structural information of the current graph but also its changes over time, most current graph representation learning methods employ transductive learning, which trains the entire network to generate node embeddings at time intervals based on continuously generated bank transaction data. This model training is time-consuming, leading to a decrease in model accuracy and efficiency. Inductive learning automatically generates node embeddings over time, often ignoring the influence of the entire network on the nodes. This application provides a group classification method and apparatus that acquires training data, determines corresponding nodes and edges based on the training data, and constructs a graph based on the nodes and edges; determines the local and global influences of the nodes based on the graph, and aggregates these local and global influences to form a hyperconnected graph; inputs the hyperconnected graph into a pre-defined convolutional neural network to obtain corresponding feature vectors, and uses the node classification labels and the feature vectors as node samples to train a preset classification model, obtaining a classification model after training; and determines the group classification result based on the classification model, thereby effectively improving the accuracy of user classification.

[0067] To effectively improve the accuracy of user classification, this application provides an embodiment of a group classification method, see [link to embodiment]. Figure 1 The group classification method specifically includes the following:

[0068] Step S101: Obtain the training data, determine the corresponding nodes and edges based on the training data, and construct a graph based on the nodes and edges.

[0069] Step S102: Determine the local and global effects of the nodes based on the graph, and aggregate the local and global effects to form a hyperconnection graph.

[0070] Understandably, there are local effects, capturing information about neighboring nodes within a specified distance. Global effects, on the other hand, occur as the overall network environment changes, impacting interactions between nodes.

[0071] Optionally, in the local influence calculation, the impact of neighbors on the target node is calculated, with different types and distances of neighbor nodes having different weights. The greater the weight of a neighbor node, the greater its influence on the target node, and vice versa. Compared to other methods, we can learn the importance of different neighbor nodes, thus more accurately capturing influential neighbor nodes. The calculation method is as follows:

[0072] Optionally, this application can calculate the affinity of neighboring nodes through the interaction between two nodes. At the same time, a time-related factor is introduced, and we use... Indicating the influencing factor at a specific moment. The definition of the impact factor is as follows.

[0073]

[0074] t is a time-dependent learnable parameter. c This represents the current time, t. i This represents a historical moment prior to the current moment. As shown in Formula 4, for example, t... i The closer to t c ,So The larger this value, the greater the total weight of local influences after incorporating the time factor. The local influence embedding is obtained as follows:

[0075] Where i represents all neighboring nodes. When the length is limited to L, the generated neighborhood graph is:

[0076] For example, in a large information network, if node A and node B interact, such as frequently transferring funds between them, it indicates that A and B may have a very close relationship. In this case, we can use affinity to calculate the closeness between the two customers. If they only had a transaction 10 years ago, then over time, the degree of influence between them will decrease, and conversely, the shorter the interaction time, the greater the possibility of them influencing each other.

[0077] Optionally, in the context of global influence, local modules can capture information about neighboring nodes within a specified distance, but cannot cover the global impact on nodes. However, the global network environment also affects interactions between nodes. For example, when a node first joins a network, it is not sensitive to changes in the global network environment. Only when the accumulated affinity exceeds a certain threshold does it become sensitive to global network changes, and the node becomes more active. Therefore, for any two nodes u... i and u j We calculate their affinity to obtain a new graph structure. ε is a value between 0 and 1, representing a threshold for the activity of nodes.

[0078]

[0079] Step S103: Input the hyperconnection graph into a pre-defined convolutional neural network to obtain the corresponding feature vectors. Then, use the classification labels of the nodes and the feature vectors as node samples to input into a pre-defined classification model for model training to obtain a classification model after model training. Finally, determine the group classification result based on the classification model.

[0080] Optionally, this application can utilize the above strategy to capture neighbor node information within a specified distance through neighbor influence, and then use global influence to establish connections for nodes with high affinity but large distances, thereby expanding the information of the graph. Subsequently, we use an aggregation method to merge the domain graph and the global graph to obtain a graph containing influential neighbor nodes, which is called a hyperconnection graph. Here, w represents the learnable weight coefficient.

[0081]

[0082] Optionally, this application can input the hyperconnected graph into a graph convolutional neural network to obtain node representations. This means that when given a new node, it is not necessary to retrain the entire network to generate its node embedding; instead, it is only necessary to compute the network input model of its local and global effects, without retraining the entire network.

[0083] As described above, the group classification method provided in this application can obtain training data, determine corresponding nodes and edges based on the training data, and construct a graph based on the nodes and edges; determine the local and global influences of the nodes based on the graph, and aggregate the local and global influences to form a hyperconnection graph; input the hyperconnection graph into a predefined convolutional neural network to obtain corresponding feature vectors, and input the classification labels of the nodes and the feature vectors as node samples into a preset classification model for model training to obtain a classification model after model training; and determine the group classification result based on the classification model, thereby effectively improving the accuracy of user classification.

[0084] Optionally, this application may use a binary cross-entropy loss function to optimize the model. The function is defined as follows.

[0085]

[0086] σ is the sigmoid function. P represents the set of neighboring nodes centered at node u at time t that interact with node v at time t. t (v) is the set of negative samples of node v at time t (negative samples refer to nodes that have not interacted with node v), w n It is a hyperparameter that can be manually adjusted to balance positive and negative samples.

[0087] Finally, the node representation and classification label are input into the classifier, which then classifies the node into the corresponding category, determining the customer's classification based on banking and financial products. Examples include wealth management, fixed-term deposits, funds, bonds, and entertainment products.

[0088] In one embodiment of the group classification method of this application, see [link to embodiment]. Figure 2 It can also specifically include the following:

[0089] Step S201: Determine the node weights of the neighboring nodes corresponding to the node in the graph;

[0090] Step S202: The node weights of the neighboring nodes determine the local influence of the node.

[0091] Optionally, this application can calculate the affinity of neighboring nodes through the interaction between two nodes. At the same time, a time-related factor is introduced, and we use... Indicating the influencing factor at a specific moment. The definition of the impact factor is as follows.

[0092]

[0093] t is a time-dependent learnable parameter. c This represents the current time, t. i This represents a historical moment prior to the current moment. As shown in Formula 4, for example, t... i The closer to t c ,So The larger this value, the greater the total weight of local influences after incorporating the time factor. The local influence embedding is obtained as follows:

[0094] Where i represents all neighboring nodes. When the length is limited to L, the generated neighborhood graph is:

[0095] In one embodiment of the group classification method of this application, see [link to embodiment]. Figure 3 It can also specifically include the following:

[0096] Step S301: Determine the current interaction time and historical interaction time between the node in the graph and its corresponding neighbor node, and determine the corresponding time influence factor;

[0097] Step S302: Determine the affinity between the node and the neighboring nodes based on the time influence factor, and determine the node weight of the neighboring nodes based on the affinity.

[0098] For example, in a large information network, if node A and node B interact, such as frequently transferring funds between them, it indicates that A and B may have a very close relationship. In this case, we can use affinity to calculate the closeness of the two customers. If they only had one transaction 10 years ago, then over time, the degree of influence between them will decrease, and conversely, the shorter the interaction time, the greater the possibility of them influencing each other.

[0099] In one embodiment of the group classification method of this application, see [link to embodiment]. Figure 4 It can also specifically include the following:

[0100] Step S401: Determine the activity level of the node by comparing the cumulative value of the affinity within a set time period with the value of a preset threshold.

[0101] Step S402: Determine the global impact of the node based on its activity level.

[0102] Optionally, in the context of global influence, local modules can capture information about neighboring nodes within a specified distance, but cannot cover the global impact on nodes. However, the global network environment also affects interactions between nodes. For example, when a node first joins a network, it is not sensitive to changes in the global network environment. Only when the accumulated affinity exceeds a certain threshold does it become sensitive to global network changes, and the node becomes more active. Therefore, for any two nodes u... i and u j We calculate their affinity to obtain a new graph structure. ε is a value between 0 and 1, representing a threshold for the activity of nodes.

[0103]

[0104] In one embodiment of the group classification method of this application, see [link to embodiment]. Figure 5 It can also specifically include the following:

[0105] Step S501: Determine the corresponding node information based on the local influence of the node;

[0106] Step S502: Establish a connection based on the node information and global influence of the two nodes in the graph to obtain a hyperconnection graph.

[0107] Optionally, this application can utilize the above strategy to capture neighbor node information within a specified distance through neighbor influence, and then use global influence to establish connections for nodes with high affinity but large distances, thereby expanding the information of the graph. Subsequently, we use an aggregation method to merge the domain graph and the global graph to obtain a graph containing influential neighbor nodes, which is called a hyperconnection graph. Here, w represents the learnable weight coefficient.

[0108]

[0109] In one embodiment of the group classification method of this application, see [link to embodiment]. Figure 6 It can also specifically include the following:

[0110] Step S601: Input the node's classification label and feature vector as node samples into a preset classification model for model training;

[0111] Step S602: Optimize the model based on the preset binary cross-entropy loss function and the training results of the model to obtain the classification model after training.

[0112] Optionally, this application can utilize the above strategy to capture neighbor node information within a specified distance through neighbor influence, and then use global influence to establish connections for nodes with high affinity but large distances, thereby expanding the information of the graph. Subsequently, we use an aggregation method to merge the domain graph and the global graph to obtain a graph containing influential neighbor nodes, which is called a hyperconnection graph. Here, w represents the learnable weight coefficient.

[0113]

[0114] Optionally, this application can input the hyperconnected graph into a graph convolutional neural network to obtain node representations. This means that when given a new node, it is not necessary to retrain the entire network to generate its node embedding; instead, it is only necessary to compute the network input model of its local and global effects, without retraining the entire network.

[0115] To effectively improve the accuracy of user classification, this application provides an embodiment of a group classification device for implementing all or part of the aforementioned group classification method. See [link to embodiment]. Figure 7 The group classification device specifically includes the following components:

[0116] Graph construction module 10 is used to acquire training data, determine corresponding nodes and edges based on the training data, and construct a graph based on the nodes and edges.

[0117] Aggregation module 20 is used to determine the local and global effects of the nodes according to the graph, and to aggregate the local and global effects to form a hyperconnection graph.

[0118] The model classification module 30 is used to input the hyperconnection graph into a preset convolutional neural network to obtain the corresponding feature vector, and to input the node's classification label and the feature vector as node samples into a preset classification model for model training to obtain a classification model after model training, and to determine the group classification result based on the classification model.

[0119] As described above, the group classification device provided in this application embodiment can acquire training data, determine corresponding nodes and edges based on the training data, and construct a graph based on the nodes and edges; determine the local and global influences of the nodes based on the graph, and aggregate the local and global influences to form a hyperconnection graph; input the hyperconnection graph into a preset convolutional neural network to obtain the corresponding feature vectors, and input the classification labels of the nodes and the feature vectors as node samples into a preset classification model for model training to obtain a classification model after model training, and determine the group classification result based on the classification model, thereby effectively improving the accuracy of user classification.

[0120] In one embodiment of the group classification device of this application, see [reference needed]. Figure 8 The aggregation module 20 includes:

[0121] The node weight determination unit 21 is used to determine the node weight of the neighboring nodes corresponding to the node in the graph.

[0122] The local influence determination unit 22 is used to determine the local influence of the node based on the node weights of the neighboring nodes.

[0123] In one embodiment of the group classification device of this application, see [reference needed]. Figure 9 The node weight determination unit 21 includes:

[0124] The influence factor determination subunit 211 is used to determine the current interaction time and historical interaction time between the node and its corresponding neighbor node in the graph, and to determine the corresponding time influence factor.

[0125] Affinity determination subunit 212 is used to determine the affinity between the node and its neighboring nodes based on the time influence factor, and to determine the node weight of the neighboring nodes based on the affinity. In one embodiment of the group classification device of this application, see... Figure 10 The aggregation module 20 further includes:

[0126] The activity determination unit 23 is used to determine the activity of the node based on the comparison between the cumulative value of affinity within a set time period and a preset threshold value.

[0127] The global impact determination unit 24 is used to determine the global impact of the node based on the node's activity level.

[0128] In one embodiment of the group classification device of this application, see [reference needed]. Figure 11 The aggregation module 20 further includes:

[0129] The node information acquisition unit 25 is used to determine the corresponding node information based on the local influence of the node.

[0130] The hyperlink establishment unit 26 is used to establish a connection based on the node information of two nodes in the graph and the global influence, thereby obtaining a hyperlink graph. In one embodiment of the group classification device of this application, see... Figure 12 The model classification module 30 includes:

[0131] The model training unit 31 is used to train the model by inputting the classification label of the node and the feature vector as node samples into a preset classification model.

[0132] The model optimization unit 32 is used to optimize the model based on the preset binary cross-entropy loss function and the model training results to obtain the classification model after the model training.

[0133] To further illustrate this solution, this application also provides a specific application example of using the above-mentioned group classification device to implement a group classification method, which specifically includes the following:

[0134] See Figure 13 It mainly consists of three modules: the first module (influence module) is composed of local and global parts; the second module (aggregation module) forms a hyperconnection graph; and the third module (classification module) performs classification. For example... Figure 13 This model allows for customer classification based on historical behavior, enabling corresponding follow-up marketing efforts based on customer category. Specifically, this includes:

[0135] Module 1 consists of two parts: local impact and global impact. Local impact captures information about neighboring nodes within a specified distance. Global impact affects interactions between nodes as the global network environment changes.

[0136] In local influence analysis, the impact of neighbors on the target node is calculated. Different types and distances of neighbor nodes have different weights; the greater the weight of a neighbor node, the greater its influence on the target node, and vice versa. Compared to other methods, we can learn the importance of different neighbor nodes, thus more accurately capturing influential neighbor nodes. The calculation method is as follows:

[0137] The affinity between neighboring nodes is calculated through the interaction between two nodes. At the same time, a time-related factor is introduced, and we use... Indicating the influencing factor at a specific moment. The definition of the impact factor is as follows.

[0138]

[0139] t is a time-dependent learnable parameter. c This represents the current time, t. i This represents a historical moment prior to the current moment. As shown in the formula, for example, t... i The closer to t c ,So The larger this value, the greater the total weight of local influences after incorporating the time factor. The local influence embedding is obtained as follows:

[0140] Where i represents all neighboring nodes. When the length is limited to L, the generated neighborhood graph is:

[0141] For example, in a large information network, if node A and node B interact, such as frequently transferring funds between them, it indicates that A and B may have a very close relationship. In this case, we can use affinity to calculate the closeness between the two customers. If they only had one transaction 10 years ago, then over time, their influence on each other will decrease, and conversely, the shorter the interaction time, the greater the likelihood of their mutual influence.

[0142] In the context of global influence, local modules can capture information about neighboring nodes within a specified distance, but they cannot cover the global impact on nodes. However, the global network environment also affects interactions between nodes. For example, when a node first joins a network, it is not sensitive to changes in the global network environment. Only when the accumulated affinity exceeds a certain threshold does it become sensitive to global network changes, and the node becomes more active. Therefore, for any two nodes u... i and u j We calculate their affinity to obtain a new graph structure. ε is a value between 0 and 1, representing a threshold for the activity of nodes.

[0143]

[0144] In aggregation module 2, we utilize the aforementioned strategy to capture neighbor node information within a specified distance based on neighbor influence. Then, we employ global influence to establish connections between nodes with high affinity but large distances, thereby expanding the graph's information. Subsequently, we use aggregation to merge the domain graph and the global graph, resulting in a graph containing influential neighbor nodes, called the hyperconnection graph. Here, 'w' represents the learnable weight coefficient.

[0145]

[0146] Representation module 3 inputs the hyperconnected graph into the graph convolutional neural network to obtain the node representations. This means that when given a new node, it is not necessary to retrain the entire network to generate its node embedding; instead, it is only necessary to compute the network input model of its local and global effects, without retraining the entire network.

[0147] For model optimization 4, we use a binary cross-entropy loss function to optimize the model. The function is defined as follows.

[0148]

[0149] σ is the sigmoid function. P represents the set of neighboring nodes centered at node u at time t that interact with node v at time t. t (v) is the set of negative samples of node v at time t (negative samples refer to nodes that have not interacted with node v), w n It is a hyperparameter that can be manually adjusted to balance positive and negative samples.

[0150] Finally, in classification module 5, the node representation and classification label are input into the classifier, allowing the node to be classified into the corresponding category, thus determining the customer's classification based on banking and financial products. Examples include wealth management, fixed-term deposits, funds, bonds, and entertainment products.

[0151] From a hardware perspective, in order to effectively improve the accuracy of user classification, this application provides an embodiment of an electronic device for implementing all or part of the content of the aforementioned group classification method. The electronic device specifically includes the following components:

[0152] The system comprises a processor, memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to realize information transmission between the group classification device and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the group classification method and the group classification device in the embodiments, the contents of which are incorporated herein, and repeated details will not be described again.

[0153] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0154] In practical applications, the group classification method can be partially executed on the electronic device side as described above, or all operations can be completed in the client device. The choice can be made based on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed in the client device, the client device may further include a processor.

[0155] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0156] Figure 14 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 14 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 14 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0157] In one embodiment, the population classification method functionality can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following control:

[0158] Step S101: Obtain the training data, determine the corresponding nodes and edges based on the training data, and construct a graph based on the nodes and edges.

[0159] Step S102: Determine the local and global effects of the nodes based on the graph, and aggregate the local and global effects to form a hyperconnection graph.

[0160] Step S103: Input the hyperconnection graph into a pre-defined convolutional neural network to obtain the corresponding feature vectors. Then, use the classification labels of the nodes and the feature vectors as node samples to input into a pre-defined classification model for model training to obtain a classification model after model training. Finally, determine the group classification result based on the classification model.

[0161] As described above, the electronic device provided in this application embodiment acquires training data, determines corresponding nodes and edges based on the training data, and constructs a graph based on the nodes and edges; determines the local and global influences of the nodes based on the graph, and aggregates the local and global influences to form a hyperconnected graph; inputs the hyperconnected graph into a predefined convolutional neural network to obtain corresponding feature vectors, and inputs the node classification labels and the feature vectors as node samples into a preset classification model for model training, thereby obtaining a classification model after model training, and determining the group classification result based on the classification model, thereby effectively improving the accuracy of user classification.

[0162] In another embodiment, the group classification device can be configured separately from the central processing unit 9100. For example, the group classification device can be configured as a chip connected to the central processing unit 9100, and the group classification method function can be implemented through the control of the central processing unit.

[0163] like Figure 14 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 14 All components shown; in addition, the electronic device 9600 may also include Figure 14 For components not shown, please refer to existing technologies.

[0164] like Figure 14 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0165] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0166] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0167] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0168] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device's communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0169] The communication module 9110 is a transmitter / receiver 9110 that transmits and receives signals via the antenna 9111. The communication module (transmitter / receiver) 9110 is coupled to the central processing unit 9100 to provide input signals and receive output signals, which can be the same as in a conventional mobile communication terminal.

[0170] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module (transmitter / receiver) 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Additionally, the audio processor 9130 is coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored sound via the speaker 9131.

[0171] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the group classification method with a server or client as the execution subject in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the group classification method with a server or client as the execution subject in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0172] Step S101: Obtain the training data, determine the corresponding nodes and edges based on the training data, and construct a graph based on the nodes and edges.

[0173] Step S102: Determine the local and global effects of the nodes based on the graph, and aggregate the local and global effects to form a hyperconnection graph.

[0174] Step S103: Input the hyperconnection graph into a pre-defined convolutional neural network to obtain the corresponding feature vectors. Then, use the classification labels of the nodes and the feature vectors as node samples to input into a pre-defined classification model for model training to obtain a classification model after model training. Finally, determine the group classification result based on the classification model.

[0175] As described above, the computer-readable storage medium provided in this application embodiment acquires training data, determines corresponding nodes and edges based on the training data, and constructs a graph based on the nodes and edges; determines the local and global influences of the nodes based on the graph, and aggregates the local and global influences to form a hyperconnected graph; inputs the hyperconnected graph into a predefined convolutional neural network to obtain corresponding feature vectors, and uses the classification labels of the nodes and the feature vectors as node samples to input into a preset classification model for model training, thereby obtaining a classification model after model training, and determining the group classification result based on the classification model, thereby effectively improving the accuracy of user classification.

[0176] Embodiments of this application also provide a computer program product capable of implementing all steps of the group classification method in the above embodiments, where the execution subject is a server or a client. When executed by a processor, this computer program / instruction implements the steps of the group classification method. For example, the computer program / instruction implements the following steps:

[0177] Step S101: Obtain the training data, determine the corresponding nodes and edges based on the training data, and construct a graph based on the nodes and edges.

[0178] Step S102: Determine the local and global effects of the nodes based on the graph, and aggregate the local and global effects to form a hyperconnection graph.

[0179] Step S103: Input the hyperconnection graph into a pre-defined convolutional neural network to obtain the corresponding feature vectors. Then, use the classification labels of the nodes and the feature vectors as node samples to input into a pre-defined classification model for model training to obtain a classification model after model training. Finally, determine the group classification result based on the classification model.

[0180] As described above, the computer program product provided in this application obtains training data, determines corresponding nodes and edges based on the training data, and constructs a graph based on the nodes and edges; determines the local and global influences of the nodes based on the graph, and aggregates the local and global influences to form a hyperconnected graph; inputs the hyperconnected graph into a predefined convolutional neural network to obtain corresponding feature vectors, and inputs the classification labels of the nodes and the feature vectors as node samples into a preset classification model for model training, thereby obtaining a classification model after model training, and determining the group classification result based on the classification model, thereby effectively improving the accuracy of user classification.

[0181] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0182] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0183] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0184] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0185] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method of classifying a population, characterized by, The method comprises: acquiring training data, determining corresponding nodes and edges according to the training data, and constructing a graph according to the nodes and edges; determining local influence and global influence of the nodes according to the graph, and aggregating the local influence and the global influence to form a super connection graph; inputting the super connection graph into a preset convolutional neural network to obtain a corresponding feature vector, inputting a classification label of the node and the feature vector as a node sample into a preset classification model for model training to obtain a classification model after the model training, and determining a group classification result according to the classification model; the determination of the local influence and the global influence of the nodes according to the graph comprises: determining current interaction time and historical interaction time of a node and corresponding neighbor nodes in the graph, determining a corresponding time influence factor, determining affinity between the node and the neighbor nodes according to the time influence factor, determining node weights of the neighbor nodes according to the affinity, and determining the local influence of the node according to the node weights of the neighbor nodes; determining the activity of the node according to a numerical comparison relationship between an accumulated value of the affinity in a set time period and a preset threshold, and determining the global influence of the node according to the activity of the node.

2. The population classification method of claim 1, wherein, the aggregation of the local influence and the global influence to form the super connection graph comprises: determining corresponding node information according to the local influence of the node; establishing a connection according to the node information and the global influence of two nodes in the graph to obtain a super connection graph.

3. The population classification method of claim 1, wherein, the inputting of the classification label of the node and the feature vector as the node sample into the preset classification model for the model training to obtain the classification model after the model training comprises: inputting the classification label of the node and the feature vector as the node sample into the preset classification model for the model training; performing model optimization according to a preset binary cross-entropy loss function and a result of the model training to obtain the classification model after the model training.

4. A population classification device, characterized by The device for performing the group classification method of any one of claims 1-3 comprises: a graph construction module configured to acquire training data, determine corresponding nodes and edges according to the training data, and construct a graph according to the nodes and edges; an aggregation module configured to determine local influence and global influence of the nodes according to the graph, and aggregate the local influence and the global influence to form a super connection graph; a model classification module configured to input the super connection graph into a preset convolutional neural network to obtain a corresponding feature vector, input a classification label of the node and the feature vector as a node sample into a preset classification model for model training to obtain a classification model after the model training, and determine a group classification result according to the classification model.

5. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps of the group classification method of any one of claims 1-3 when executing the program.

6. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program implements the steps of the group classification method of any one of claims 1-3 when executed by the processor.

7. A computer program product comprising computer programs / instructions, characterized in that, The computer program / instructions, when executed by the processor, implement the steps of the population classification method of any one of claims 1 to 3.

Images