CN117611224A

CN117611224A - Client grouping method, device, equipment, medium and program product

Info

Publication number: CN117611224A
Application number: CN202311656902.6A
Authority: CN
Inventors: 田吉龙; 王凯; 周洪菊; 曾文华
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2023-12-05
Filing date: 2023-12-05
Publication date: 2024-02-27

Abstract

The invention provides a client grouping method, a client grouping device, a client grouping equipment, a client grouping medium and a client grouping program product, which can be applied to the technical field of big data and the financial field. The method comprises the following steps: collecting attribute information and behavior data of m clients to form a data set; converting data in the dataset into first graph structure data comprising a first node list, first node features, a first edge list and first edge weights; constructing a customer relation graph according to the first graph structure data, wherein the customer relation graph comprises m nodes and n sides, the m nodes correspond to a first node list and first node characteristics, and the n sides correspond to a first side list and first side weights; importance sampling is carried out on the client relationship graph based on the weights of the n sides, so that a plurality of client relationship subgraphs are obtained; inputting a plurality of client relationship subgraphs into a graph rolling network to perform feature extraction to obtain second node features, wherein the dimensions of the second node features are lower than those of the first node features; and processing the second node characteristic by using a clustering algorithm to group m clients.

Description

Client grouping method, device, equipment, medium and program product

Technical Field

The present invention relates to the field of big data technologies, and in particular, to a client grouping method, device, apparatus, medium, and program product.

Background

With the increase of market competition, enterprises pay more and more attention to the fine operation of clients. Customer grouping is an effective operation strategy, and accurate operation is performed for different types of customers by grouping the customers, so that customer satisfaction and loyalty are improved. In customer clustering, enterprises need to be clustered according to different characteristics and behaviors of customers. These characteristics may include customer's purchase preferences, frequency of purchase, amount of purchase, channel of purchase, time of purchase, etc. By analyzing these features, an enterprise may divide a customer into different groups, e.g., high value customers, low value customers, potential customers, etc. Customer clustering can also help enterprises better understand customer needs and market trends. Through deep analysis on different types of clients, enterprises can better understand the consumption psychology and behavior habit of the clients, so that the client demands and market trends are better met.

The existing technical schemes of client clustering mainly comprise two kinds, one is based on the traditional machine learning and statistics method, and the client clustering task is usually realized by using manually extracted features and a classical clustering algorithm. Common clustering algorithms include K-means clustering, hierarchical clustering, density clustering, and the like. These algorithms may divide clients into different groups according to their characteristics. The traditional machine learning method only considers the characteristics of the nodes, ignores the relation among the nodes, and possibly causes information loss in the graph data. Moreover, for large-scale graph data, the conventional method may have high computational complexity, and is difficult to apply to processing of large-scale graph data. The other customer grouping technical scheme is based on a traditional graph neural network (Graph Neural Networks, GNN), wherein the GNN is a deep learning model specially used for processing graph structure data, and can perform feature representation learning on nodes and edges and perform information transfer on a graph. However, the traditional graph neural network technical scheme has large calculation overhead and is easy to cause neighbor explosion. Meanwhile, the conventional GNN generally adopts fixed weight parameters in information aggregation, which means that the weight of the neighboring node of each node on the influence of the neighboring node is the same, thereby causing limitation of information aggregation. Such an information aggregation approach may not be suitable for handling complex graph structures because the importance and relationships between nodes may be dynamically changing, while GNNs cannot assign different importance to different neighbor nodes of a node.

Disclosure of Invention

In view of the foregoing, embodiments of the present invention provide a client grouping method, apparatus, device, medium, and program product.

According to a first aspect of the present invention, there is provided a client clustering method comprising: under the condition of obtaining authorization of m clients, collecting attribute information and behavior data of the m clients to form a data set, wherein m is a positive integer greater than or equal to 3; converting data in the dataset into first graph structure data, wherein the first graph structure data comprises a first node list, first node features, a first edge list and first edge weights; constructing a customer relation graph according to the first graph structure data, wherein the customer relation graph comprises m nodes and n sides, the m nodes correspond to a plurality of nodes in the first node list and the first node characteristics, the n sides correspond to a plurality of first sides in the first side list and the first side weights, and n is a positive integer greater than or equal to 3; based on the weights of the n sides, importance sampling is carried out on the client relationship graph, and a plurality of client relationship subgraphs are obtained; inputting the plurality of client relationship subgraphs into a graph convolution network to perform feature extraction to obtain a second node feature, wherein the dimension of the second node feature is lower than that of the first node feature; and processing the second node characteristic by using a clustering algorithm to group the m clients.

According to an embodiment of the present invention, the converting the data in the dataset into the first graph structure data includes: constructing the first node list and the first node characteristics based on the attribute information and the behavior data of the m clients; constructing the first edge list based on transaction data among the m clients; and determining the first edge weight based on the transaction amount and the number of transactions in the transaction data.

According to an embodiment of the present invention, the performing importance sampling on the client relationship graph to obtain a plurality of client relationship subgraphs includes: a random path acquisition step: starting from any node in the client relation graph, performing random walk for a plurality of times to obtain a plurality of random paths, wherein the node corresponding to the starting point is used as a central node; sampling the neighbor nodes: sampling neighbor nodes in each step according to the set wander length and the number of neighbor nodes for each random path in the plurality of random paths, and constructing a client relationship subgraph corresponding to each random path where the center node walks randomly; and repeating the random path acquisition step and the neighbor node sampling step for each of the m nodes to obtain the plurality of customer relationship subgraphs.

According to an embodiment of the present invention, the sampling the neighbor node in each step includes: and selecting a part of neighbor nodes for sampling according to a first edge weight between the center node and the neighbor nodes, and obtaining a client relationship subgraph corresponding to each random path of the random walk of the center node, wherein the larger the first edge weight is, the higher the probability that the corresponding neighbor node is sampled is.

According to the embodiment of the invention, each client relation sub-graph comprises a first node list and a first edge list, and the first node list and the first edge list corresponding to the client relation sub-graphs are used as feature matrixes of the m nodes.

According to an embodiment of the present invention, the feature extraction of the plurality of client relationship subgraphs input into the graph convolution network to obtain a second node feature includes: performing linear transformation on the feature matrix of the m nodes, and mapping node features to a new representation space; splicing the initial node characteristics and the end node characteristics of each edge in the plurality of client relationship subgraphs to obtain an attention input matrix containing twice the output dimension; calculating an attention score for each edge in the plurality of customer relationship subgraphs using the full connection layer; performing edge normalization on the attention scores, wherein the sum of weights of all edges corresponding to each node is 1; and multiplying the neighbor node characteristics of each node by the corresponding attention weight to obtain second node characteristics.

According to an embodiment of the invention, the graph convolution network includes two graph convolution layers and two attention layers.

A second aspect of the present invention provides a client clustering apparatus comprising: the acquisition module is used for collecting attribute information and behavior data of m clients under the condition of acquiring authorization of the m clients to form a data set, wherein m is a positive integer greater than or equal to 3; the conversion module is used for converting the data in the data set into first graph structure data, wherein the first graph structure data comprises a first node list, first node characteristics, a first edge list and first edge weights; the construction module is used for constructing a customer relation graph according to the first graph structure data, wherein the customer relation graph comprises m nodes and n sides, the m nodes correspond to a plurality of nodes in the first node list and the first node characteristics, the n sides correspond to a plurality of first sides and the first side weights in the first side list, and n is a positive integer greater than or equal to 3; the sampling module is used for sampling importance of the client relationship graph based on the weights of the n sides to obtain a plurality of client relationship subgraphs; the feature extraction module is used for carrying out feature extraction on the plurality of client relationship subgraphs input into a graph convolution network to obtain a second node feature, wherein the dimension of the second node feature is lower than that of the first node feature; and the classification module is used for processing the second node characteristics by using a clustering algorithm so as to group the m clients.

A third aspect of the present invention provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform the client clustering method described above.

A fourth aspect of the present invention also provides a computer readable storage medium having stored thereon executable instructions which, when executed by a processor, cause the processor to perform the above-described client clustering method.

The fifth aspect of the present invention also provides a computer program product comprising a computer program which, when executed by a processor, implements the client clustering method described above.

According to the embodiment of the invention, the client grouping method, the device, the equipment, the medium and the program product provided by the invention have at least the following beneficial effects: by converting the original data into the graph structure data and sampling the importance of the graph structure data, the subgraph is extracted from the original graph structure data, so that the scale of the graph is reduced, and the calculation and memory overhead is reduced. Through sampling, the graph neural network can be trained more efficiently, so that the training model can process large-scale graph data and converge rapidly. By introducing an attention mechanism, different attention weights can be dynamically given to the relations between the nodes, so that the importance and the relevance between the nodes can be more flexibly learned, the structural information of the graph can be better captured, a more complex relation mode can be found in the client clustering process, and the accuracy and the efficiency of client clustering are improved.

Drawings

For a more complete understanding of the present invention, and the advantages thereof, reference is now made to the embodiments of the invention described in detail with reference to the accompanying drawings, wherein:

FIG. 1 schematically illustrates an application scenario diagram of a client clustering method according to an embodiment of the invention;

FIG. 2 schematically illustrates an effect diagram of customer groupings in accordance with an embodiment of the present invention;

FIG. 3 schematically illustrates a flow chart of a client grouping method according to an embodiment of the invention;

FIG. 4 schematically illustrates a flow chart of the conversion of a dataset into diagram structural data according to an embodiment of the present invention;

FIG. 5 schematically illustrates a flow chart of sampling according to an embodiment of the invention;

FIG. 6 schematically illustrates a block diagram of a graph rolling network in accordance with an embodiment of the invention;

FIG. 7 schematically illustrates a flow diagram of graph roll-up network feature extraction, in accordance with an embodiment of the invention;

FIG. 8 schematically illustrates a block diagram of a client grouping apparatus according to an embodiment of the present invention;

fig. 9 schematically shows a block diagram of an electronic device adapted to implement customer clustering in accordance with an embodiment of the invention.

In the drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

Detailed Description

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. It should be understood that the description is only illustrative and is not intended to limit the scope of the invention. In the following detailed description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the invention. It may be evident, however, that one or more embodiments may be practiced without these specific details. In addition, in the following description, descriptions of well-known structures and techniques are omitted so as not to unnecessarily obscure the present invention.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The terms "comprises," "comprising," and/or the like, as used herein, specify the presence of stated features, steps, operations, and/or components, but do not preclude the presence or addition of one or more other features, steps, operations, or components.

All terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. It should be noted that the terms used herein should be construed to have meanings consistent with the context of the present specification and should not be construed in an idealized or overly formal manner.

Where expressions like at least one of "A, B and C, etc. are used, the expressions should generally be interpreted in accordance with the meaning as commonly understood by those skilled in the art (e.g.," a system having at least one of A, B and C "shall include, but not be limited to, a system having a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.).

In the technical scheme of the invention, the related user information (including but not limited to user personal information, user image information, user equipment information, such as position information and the like) and data (including but not limited to data for analysis, stored data, displayed data and the like) are information and data authorized by a user or fully authorized by all parties, and the related data are collected, stored, used, processed, transmitted, provided, invented, applied and the like, and all processed according to the related laws and regulations and standards of related countries and regions, necessary security measures are adopted, the public welfare is not violated, and corresponding operation inlets are provided for the user to select authorization or rejection.

Technical terms involved in the present invention are briefly described below in order to better understand the present solution by relevant persons.

Neighbor explosion: in the graph neural network, a single training sample is a plurality of nodes, and the nodes are interdependent. Each layer of calculation is to aggregate the neighbor node information of the current node, so that the node to be calculated after the layer number is increased can exponentially increase along with the increase of the layer number, thereby causing the calculated amount of the multi-layer node to explode.

Graph sampling: graph sampling is a method of extracting subgraphs from a graph for processing large-scale graph data so that the graph neural network can be trained more efficiently. In processing large graph data, the entire graph may contain a large number of nodes and edges, which can result in the training process of the graph neural network being very time consuming. The goal of graph sampling is to extract sub-graphs from the original graph, making the sub-graphs smaller in size, thereby reducing computation and memory overhead, and preserving as much as possible the structural information of the original graph.

Edge weight: in a complex network, edge weights are used to represent the strength of the relationship between nodes at both ends of an edge, and in a weightless network, the relationship between nodes represented by each edge is the same. In an actual network, the relation between nodes is different due to different application scenes, so that the edge weights between different nodes are also different.

Importance sampling: when sampling neighbor nodes of a center node to construct a subgraph, determining the probability of selecting the neighbor nodes according to the edge weights, wherein the probability of selecting the nodes corresponding to the edges with larger weights is higher. This allows important nodes and edges to be preserved, while some less important nodes and edges are ignored.

Attention score: typically generated by an attention model, there may be different arrangements of scores depending on the implementation and application scenario. For example, the scores are arranged in order of magnitude from high to low, indicating that the attention to each element in the input sequence is high to low.

In order to better serve clients and improve the service quality, client grouping is an effective operation strategy. By grouping the clients, accurate operation is performed for different types of clients, thereby improving the satisfaction and loyalty of the clients. According to the embodiment of the invention, through collecting the attribute information and the behavior data of the authorized clients, carrying out graph structure conversion, importance sampling and graph neural network training on the data of the clients, and adopting a clustering algorithm to automatically group the clients, the importance and the relevance among the clients can be more flexibly learned, and the structure information of the graph can be better captured, so that a more complex relation mode can be found in the process of automatically grouping the clients, and the accuracy and the efficiency of the client grouping are improved.

Fig. 1 schematically shows an application scenario diagram of a client clustering method according to an embodiment of the invention.

As shown in fig. 1, an application scenario 100 according to this embodiment may include m clients, where a direct association relationship exists between some of the m clients, for example, in a financial system, and a direct transaction relationship exists between some of the clients. According to the association relation between the clients, n sides between m clients are constructed, for example, any transaction in transaction records existing between m clients can obtain two clients, namely an initiator and a receiver of the transaction. The two clients are used as a starting node and an ending node of the edge to form an edge. Similarly, for each transaction record, one edge is constructed, thereby constructing n edges representing the interrelationship between m clients.

In embodiments of the present invention, the user's consent or authorization may be obtained prior to obtaining the user's information. For example, a request may be issued to the user to obtain user information. In the event that the user agrees or authorizes that user information is available, information is collected about the customers, such as identity information and transaction records between the customers.

Fig. 2 schematically shows an effect diagram of customer clustering according to an embodiment of the invention.

As shown in fig. 2, after the client grouping apparatus 800 collects attribute information and behavior information of m clients to form a data set, the data set may be converted into graph structure data, so that correspondence between m clients and m nodes and n edges in the graph structure data may be established. The graph structure data is further processed and analyzed, such as importance sampling, feature extraction and the like, and then clients are clustered according to analysis results, so that similar clients can be clustered together more accurately, and more personalized and accurate service and marketing strategies can be provided. In this embodiment, m clients may be divided into a different client groups, a being a positive integer greater than or equal to 3. The number of people in each customer group may be the same or different from each other. The clients in the same client group may or may not have directly related edges. For example, referring to fig. 1 and 2 in combination, there is no directly associated edge between the client C1 and the client C5, but the clusters are grouped into the same client group G1 according to the similarity. There may also be a directly associated edge between two clients in two different client groups, for example, there may be a directly associated edge between client C1 in client group G1 and client C2 in the other client group.

Through converting the attribute information and behavior information of the clients into nodes and edges in the graph data structure, then carrying out importance sampling and feature extraction on the nodes and edges, and then clustering and grouping, large-scale graph data can be efficiently processed, meanwhile, the personal information and behavior mode of the clients are fully utilized, and similar clients are clustered together more accurately, so that more personalized and accurate service and marketing strategies are provided.

Fig. 3 schematically shows a flow chart of a client grouping method according to an embodiment of the invention.

As shown in fig. 3, the client clustering method of this embodiment includes operations S310 to S360.

The method may be applied to a computer device, which may be a server or a service terminal having a certain data processing capability, etc., and the server may be a server providing various services, such as a background management server (merely an example) providing support for a website browsed by a user using the service terminal device. The embodiment of the invention does not limit the product type of the computer equipment.

In operation S310, in the case of acquiring authorization of m clients, attribute information and behavior data of the m clients are collected to form a data set, where m is a positive integer greater than or equal to 3.

Taking a financial system as an example, after obtaining the authorization of the client, historical behavior data and attribute information of the client can be collected from the embedded point logs of a mobile banking system, an internet banking system and a customer service system, and the data can be stored in a database so as to facilitate subsequent processing and analysis of the data.

In operation S320, data in the dataset is converted into first graph structure data, wherein the first graph structure data includes a first node list, a first node feature, a first edge list, and a first edge weight.

In this embodiment, the attribute information and behavior data of the client are combined together as a feature vector of the node. These nodes may be represented using unique identifiers (customer IDs). Taking a financial system as an example, historical behavioral data of a customer may include, but is not limited to:

recent mobile bank login times, for example, statistics of the number of times a customer has logged in to a mobile bank in the last 5 years;

recent mobile banking queries, such as counting the number of times a customer has queried balance and transaction records on a mobile bank for the last 5 years;

recent mobile banking transaction frequency, such as counting the number of transactions of the last 5 years of customers; recent mobile banking transaction time, such as counting the customer's recent transaction time;

The recent mobile banking transaction amount average, for example, calculates the average transaction amount of the last 5 years of the customer;

recent mobile banking transaction number average, for example, calculate the average per transaction amount for the last 5 years of customers;

recent online banking login times, such as statistics of the last 5 years of online banking login times of a client, reflect the use frequency of online banking service of the client;

the number of purchased products, such as counting the number of financial products purchased by the customer;

purchase product types, such as recording specific financial product types purchased by customers, such as financial products, loan products, etc.;

the attribute information of the client may include: age, income, sex, region, occupation, marital status, etc.

According to the transaction data of the clients, the transaction relationship between the clients, namely the edges between the clients, is constructed. For each transaction, two clients, namely an initiator and a receiver of the transaction, can be obtained, and the two clients are used as a starting node and an ending node of the edge to form an edge.

In operation S330, a customer relationship graph is constructed according to the first graph structure data, where the customer relationship graph includes m nodes and n edges, the m nodes correspond to a plurality of nodes in the first node list and the first node features, the n edges correspond to a plurality of first edges in the first edge list and the first edge weights, and n is a positive integer greater than or equal to 3.

Through establishing the corresponding relation between the graph structure data and the client relation, the clients can be clustered and clustered conveniently by extracting the characteristics in the graph structure data, and the efficiency and accuracy of client clustering are improved.

In operation S340, importance sampling is performed on the client relationship graph based on the weights of the n edges, so as to obtain a plurality of client relationship subgraphs.

In the embodiment of the invention, in order to improve the efficiency of client grouping, the client relationship graph can be sampled, so that a representative client relationship subgraph is obtained. The structural information of the original graph can be reserved as much as possible through importance sampling, meanwhile, the calculation and memory expenditure is reduced, and the calculation efficiency is improved, so that more efficient customer clustering is realized.

In operation S350, the plurality of client relationship subgraphs are input into a graph rolling network to perform feature extraction, so as to obtain a second node feature, where the dimension of the second node feature is lower than that of the first node feature.

In the embodiment of the invention, after importance sampling is completed, a plurality of client relationship subgraphs can be obtained, and then the feature extraction is carried out on the client relationship subgraphs through a training graph convolution network, so that the client nodes are represented as the second node features of the low-dimensional vectors, and the clients can be conveniently and accurately clustered according to the second node features of the clients.

In operation S360, the second node characteristic is processed using a clustering algorithm to cluster the m clients.

In the embodiment of the invention, the node representation can be divided into the groups with the self-defined number by using a K-means clustering algorithm, so that the automatic grouping of clients is realized.

In an embodiment of the present invention, the second node feature may be further processed by an algorithm selected from one or more of a hierarchical clustering algorithm, a DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm, a OPTICS (Ordering Points To Identify the Clustering Structure) algorithm, and a GMM (Gaussian Mixture Model) algorithm, so as to cluster the m clients.

Hierarchical clustering is a bottom-up or top-down clustering method that organizes data points by constructing a tree-like hierarchical structure of clusters. This hierarchy may be represented as a tree diagram.

DBSCAN is a density-based clustering algorithm that divides clusters by finding the density reachability of data points and can automatically handle noise.

OPTICS is an extended density-based clustering algorithm that can identify clusters of different densities and generate an ordered list of clusters.

GMM is a probabilistic model that models data points as a mixture of gaussian distributions. Through maximum likelihood estimation, the GMM may assign data points to different gaussian distribution clusters.

Different clustering algorithms can be adopted according to the types of the client groups to automatically group the clients, so that the efficiency and accuracy of client grouping are improved.

Fig. 4 schematically shows a flow chart of the conversion of a data set into graph structure data according to an embodiment of the invention.

In an embodiment of the present invention, converting the data in the data set into the first graph structure data in operation S320 specifically further includes operations S410 to S430.

In operation S410, the first node list and the first node characteristics are constructed based on the attribute information and behavior data of the m clients.

Constructing the first edge list based on transaction data among the m clients in operation S420;

in operation S430, the first side weight is determined based on the transaction amount and the number of transactions in the transaction data.

By establishing the corresponding relation between the transaction amount and the transaction times in the transaction data and the first side weight, the size or importance of the transaction can be reflected, so that important nodes and sides can be reserved as much as possible during subsequent sampling, unimportant nodes and sides are omitted, and the accuracy of sampling is improved.

Fig. 5 schematically shows a flow chart of sampling according to an embodiment of the invention.

In the embodiment of the present invention, the importance sampling of the client relationship graph to obtain a plurality of client relationship subgraphs includes operations S510 to S530.

In operation S510, a random path acquisition step: and starting from any node in the client relation graph, performing random walk for a plurality of times to obtain a plurality of random paths, wherein the node corresponding to the starting point is used as a central node.

In this embodiment, a path may be randomly selected from any node representing a client to walk, and the length of the walk and the number of neighboring nodes may be specifically set according to actual needs. Multiple random paths may be selected for the same node.

In operation S520, the neighbor node sampling step: and sampling neighbor nodes in each step according to the set wander length and the number of the neighbor nodes for each random path in the plurality of random paths, and constructing a client relationship subgraph corresponding to each random path which is randomly walked by the central node.

In an embodiment of the present invention, the sampling the neighbor node in each step includes: and selecting a part of neighbor nodes for sampling according to a first edge weight between the center node and the neighbor nodes, and obtaining a client relationship subgraph corresponding to each random path of the random walk of the center node, wherein the larger the first edge weight is, the higher the probability that the corresponding neighbor node is sampled is. Taking a financial system as an example, the first side weight may be assigned accordingly according to the total amount of transactions between clients within a period of time, for example, 5 transactions within a year between client a and client B, the total amount of transactions being 10 tens of thousands, the first side weight between client a and client B being assigned 10 tens of thousands, 8 transactions within a year between client C and client D, the total amount of transactions being 1 tens of thousands, the first side weight between client C and client D being assigned 1 tens of thousands.

In the embodiment of the invention, graph saint can be utilized to sample and train graph data, subgraphs are extracted from the original graph, the scale of the graph is reduced, and the calculation and memory overhead is reduced. GraphSAINT is a graph neural network training method based on a sampling subgraph, and the calculation on each small-batch training sample batch is performed on the sampling subgraph, so that the phenomenon of neighbor explosion does not occur.

Importance sampling based on the first edge weights may preserve important nodes and edges while omitting some less important nodes and edges. The importance and the relevance among the clients can be more flexibly learned, and the structural information of the graph can be better captured, so that more complex relation modes can be found in the process of automatic client clustering, and the accuracy and the efficiency of client clustering are improved.

In operation S530, the random path acquisition step and the neighbor node sampling step are repeatedly performed for each of the m nodes to obtain the plurality of customer relationship subgraphs.

In an embodiment of the present invention, each of the client relationship subgraphs includes a first node list and a first edge list, and the first node list and the first edge list corresponding to the plurality of client relationship subgraphs are used as feature matrices of the m nodes.

By sampling each of the m nodes to obtain a plurality of client relationship subgraphs, important information about the m nodes can be reserved as much as possible, and reliability of subsequent feature extraction of the m nodes is guaranteed, so that accuracy of client grouping is improved.

Fig. 6 schematically shows a block diagram of a graph rolling network according to an embodiment of the invention.

In an embodiment of the invention, feature extraction is performed on the sampled graph data of the plurality of customer relationship subgraphs through a graph convolution network. As shown in fig. 6, the graph convolution network 600 includes two graph convolution layers and two attention layers, such as a first graph convolution layer 610, a first attention layer 620, a second graph convolution layer 630, and a second attention layer 640.

Fig. 7 schematically illustrates a flow diagram of graph roll-up network feature extraction according to an embodiment of the invention.

In an embodiment of the present invention, the feature extraction is performed on the plurality of client relationship subgraphs input into the graph convolution network, and obtaining the second node feature includes operations S710 to S750.

In operation S710, the feature matrix of the m nodes is linearly transformed to map the node features to a new representation space. Operation S710 may be performed by the first graph convolutional layer 610.

In operation S720, the respective start node feature and end node feature of each edge in the plurality of client relationship subgraphs are spliced to obtain an attention input matrix including twice the output dimension. Operation S720 may be performed by the first attention layer 620.

In operation S730, an attention score of each edge in the plurality of customer relationship subgraphs is calculated using the full connection layer. Operation S730 may be performed by the second graph convolution layer 630.

In operation S740, the attention score is edge normalized, where the sum of the weights of all edges corresponding to each node is 1. Operation S740 may be performed by the second attention layer 640. For example, there are 4 sides corresponding to one node, and the respective attention scores calculated in S730 are 40, 30, 20 and 10, respectively. After edge normalization, the respective attention scores are calculated according to the ratio of the corresponding attention score to the total attention score corresponding to the node, for example, the normalized attention scores of the four edges corresponding to the node are respectively 0.4,0.3,0.2 and 0.1.

In operation S750, the neighboring node features of each node are multiplied by the corresponding attention weight to obtain a second node feature.

In the embodiment of the invention, the dimension of the second node characteristic is lower than that of the first node characteristic, so that the more concise and more prominent node characteristic is obtained, and the subsequent clustering algorithm is convenient for clustering.

By introducing an attention mechanism in training the graph convolutional network, different attention weights can be dynamically assigned to the relationships between nodes. Such a mechanism of attention allows the training model of the graph rolling network to learn more about the importance and relevance between nodes, thereby better capturing the structural information of the graph. The attention mechanism helps to find more complex relational patterns in the client's automatic clustering, improving the performance and behavior of the model.

Fig. 8 schematically shows a block diagram of a client grouping apparatus according to an embodiment of the present invention.

As shown in fig. 8, the client clustering apparatus 800 of this embodiment includes an acquisition module 810, a conversion module 820, a construction module 830, a sampling module 840, a feature extraction module 850, and a classification module 860.

The obtaining module 810 is configured to collect attribute information and behavior data of m clients to form a data set in a case of obtaining authorization of the m clients, where m is a positive integer greater than or equal to 3. In an embodiment, the obtaining module 810 may be configured to perform the operation S310 described above, which is not described herein.

The conversion module 820 is configured to convert data in the dataset into first graph structure data, where the first graph structure data includes a first node list, a first node feature, a first edge list, and a first edge weight. In an embodiment, the conversion module 820 may be used to perform the operation S320 described above, which is not described herein.

In an embodiment of the present invention, the converting module 820 converts the data in the dataset into the first graph structure data includes: operations S410 to S430.

in operation S430, the first side weight is determined based on the transaction amount and the number of transactions in the transaction data. The building module 830 is configured to build a customer relationship graph according to the first graph structure data, where the customer relationship graph includes m nodes and n edges, the m nodes correspond to a plurality of nodes in the first node list and the first node feature, the n edges correspond to a plurality of first edges in the first edge list and the first edge weight, and n is a positive integer greater than or equal to 3. In an embodiment, the construction module 830 may be configured to perform the operation S330 described above, which is not described herein.

The sampling module 840 is configured to sample importance of the client relationship graph based on the weights of the n edges, so as to obtain a plurality of client relationship subgraphs. In an embodiment, the sampling module 840 may be configured to perform the operation S340 described above, which is not described herein.

In an embodiment of the present invention, the sampling module 840 performs importance sampling including S510-S530.

In operation S520, the neighbor node sampling step: sampling neighbor nodes in each step of each random path in the plurality of random paths according to the set wandering length and the number of neighbor nodes, and constructing a client relationship subgraph corresponding to each random path randomly walked by the center node, wherein the sampling of the neighbor nodes in each step comprises:

And selecting a part of neighbor nodes for sampling according to a first edge weight between the center node and the neighbor nodes, and obtaining a client relationship subgraph corresponding to each random path of the random walk of the center node, wherein the larger the first edge weight is, the higher the probability that the corresponding neighbor node is sampled is.

Each client relation sub-graph comprises a first node list and a first edge list, and the first node list and the first edge list corresponding to the client relation sub-graphs are used as feature matrixes of the m nodes.

The feature extraction module 850 is configured to perform feature extraction on the plurality of client relationship subgraphs input into the graph convolution network, so as to obtain a second node feature, where a dimension of the second node feature is lower than a dimension of the first node feature. In an embodiment, the feature extraction module 850 may be used to perform the operation S350 described above, which is not described herein.

In an embodiment of the present invention, the feature extraction module 850 performs feature extraction on the plurality of client relationship subgraphs input graph convolution networks, and obtaining the second node feature includes operations S7 1 0 to S750.

The classification module 860 is configured to process the second node characteristic by using a clustering algorithm to group the m clients. In an embodiment, the classification module 860 may be configured to perform the operation S360 described above, which is not described herein. By employing importance sampling, important node information and important edges in the graph structure data can be preserved while reducing the data size of the graph structure data. By employing an attention mechanism in the graph neural network training, relationships between nodes can be dynamically weighted differently. Such a mechanism of attention allows the training model of the graph rolling network to learn more about the importance and relevance between nodes, thereby better capturing the structural information of the graph. The attention mechanism helps to find more complex relational patterns in the client's automatic clustering, improving the performance and behavior of the model.

Any of the acquisition module 810, the conversion module 820, the construction module 830, the sampling module 840, the feature extraction module 850, and the classification module 860 may be combined in one module to be implemented, or any of the modules may be split into a plurality of modules, according to an embodiment of the present invention. Alternatively, at least some of the functionality of one or more of the modules may be combined with at least some of the functionality of other modules and implemented in one module. At least one of the acquisition module 810, the conversion module 820, the construction module 830, the sampling module 840, the feature extraction module 850, and the classification module 860 may be implemented, at least in part, as hardware circuitry, such as a Field Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a system-on-chip, a system-on-substrate, a system-on-package, an Application Specific Integrated Circuit (ASIC), or in hardware or firmware in any other reasonable manner of integrating or packaging the circuitry, or in any one of or a suitable combination of three of software, hardware, and firmware. Alternatively, at least one of the acquisition module 810, the conversion module 820, the construction module 830, the sampling module 840, the feature extraction module 850, and the classification module 860 may be at least partially implemented as computer program modules that, when executed, perform the corresponding functions.

It should be noted that, the implementation manner, the technical problem to be solved, the function to be realized, and the technical effect to be achieved of each module in the embodiment of the apparatus portion are the same as or similar to the implementation manner, the technical problem to be solved, the function to be realized, and the technical effect to be achieved of each corresponding step in the embodiment of the method portion, respectively, and are not described herein again.

As shown in fig. 9, an electronic device 900 according to an embodiment of the present invention includes a processor 901 that can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM) 902 or a program loaded from a storage section 908 into a Random Access Memory (RAM) 903. The processor 901 may include, for example, a general purpose microprocessor (e.g., a CPU), an instruction set processor and/or an associated chipset and/or a special purpose microprocessor (e.g., an Application Specific Integrated Circuit (ASIC)), or the like. Processor 901 may also include on-board memory for caching purposes. Processor 901 may include a single processing unit or multiple processing units for performing the different actions of the method flows according to embodiments of the invention.

In the RAM 903, various programs and data necessary for the operation of the electronic device 900 are stored. The processor 901, the ROM 902, and the RAM 903 are connected to each other by a bus 904. The processor 901 performs various operations of the method flow according to an embodiment of the present invention by executing programs in the ROM 902 and/or the RAM 903. Note that the program may be stored in one or more memories other than the ROM 902 and the RAM 903. The processor 901 may also perform various operations of the method flow according to embodiments of the present invention by executing programs stored in one or more memories.

According to an embodiment of the invention, the electronic device 900 may also include an input/output (I/O) interface 905, the input/output (I/O) interface 905 also being connected to the bus 904. The electronic device 900 may also include one or more of the following components connected to the I/O interface 905: an input section 906 including a keyboard, a mouse, and the like; an output portion 907 including a display such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and a speaker; a storage portion 908 including a hard disk or the like; and a communication section 909 including a network interface card such as a LAN card, a modem, or the like. The communication section 909 performs communication processing via a network such as the internet. The drive 910 is also connected to the I/O interface 905 as needed. A removable medium 911 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is installed as needed on the drive 910 so that a computer program read out therefrom is installed into the storage section 908 as needed.

The present invention also provides a computer-readable storage medium that may be embodied in the apparatus/device/system described in the above embodiments; or may exist alone without being assembled into the apparatus/device/system. The computer-readable storage medium carries one or more programs which, when executed, implement methods in accordance with embodiments of the present invention.

According to embodiments of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, which may include, for example, but is not limited to: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. For example, according to embodiments of the invention, the computer-readable storage medium may include ROM 902 and/or RAM 903 and/or one or more memories other than ROM 902 and RAM 903 described above.

Embodiments of the present invention also include a computer program product comprising a computer program containing program code for performing the method shown in the flowcharts. The program code means for causing a computer system to carry out the method for automatic client clustering as provided by the embodiments of the present invention when the computer program product is run on the computer system.

The above-described functions defined in the system/apparatus of the embodiment of the present invention are performed when the computer program is executed by the processor 901. The systems, apparatus, modules, units, etc. described above may be implemented by computer program modules according to embodiments of the invention.

In one embodiment, the computer program may be based on a tangible storage medium such as an optical storage device, a magnetic storage device, or the like. In another embodiment, the computer program may also be transmitted, distributed, and downloaded and installed in the form of a signal on a network medium, via communication portion 909, and/or installed from removable medium 911. The computer program may include program code that may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the foregoing.

In such an embodiment, the computer program may be downloaded and installed from the network via the communication portion 909 and/or installed from the removable medium 911. The above-described functions defined in the system of the embodiment of the present invention are performed when the computer program is executed by the processor 901. The systems, devices, apparatus, modules, units, etc. described above may be implemented by computer program modules according to embodiments of the invention.

According to embodiments of the present invention, program code for carrying out computer programs provided by embodiments of the present invention may be written in any combination of one or more programming languages, and in particular, such computer programs may be implemented in high-level procedural and/or object-oriented programming languages, and/or in assembly/machine languages. Programming languages include, but are not limited to, such as Java, c++, python, "C" or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of remote computing devices, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., connected via the Internet using an Internet service provider).

The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

Those skilled in the art will appreciate that the features recited in the various embodiments of the invention and/or in the claims may be combined in various combinations and/or combinations even if such combinations and/or combinations are not explicitly recited in the invention. In particular, the features recited in the various embodiments of the invention and/or in the claims can be combined in various combinations and/or combinations without departing from the spirit and teachings of the invention. All such combinations and/or combinations fall within the scope of the invention.

The embodiments of the present invention are described above. However, these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Although the embodiments are described above separately, this does not mean that the measures in the embodiments cannot be used advantageously in combination. The scope of the invention is defined by the appended claims and equivalents thereof. Various alternatives and modifications can be made by those skilled in the art without departing from the scope of the invention, and such alternatives and modifications are intended to fall within the scope of the invention.

Claims

1. A method of customer clustering, wherein the method comprises:

under the condition of obtaining authorization of m clients, collecting attribute information and behavior data of the m clients to form a data set, wherein m is a positive integer greater than or equal to 3;

converting data in the dataset into first graph structure data, wherein the first graph structure data comprises a first node list, first node features, a first edge list and first edge weights;

constructing a customer relation graph according to the first graph structure data, wherein the customer relation graph comprises m nodes and n sides, the m nodes correspond to a plurality of nodes in the first node list and the first node characteristics, the n sides correspond to a plurality of first sides in the first side list and the first side weights, and n is a positive integer greater than or equal to 3;

Based on the weights of the n sides, importance sampling is carried out on the client relationship graph, and a plurality of client relationship subgraphs are obtained;

inputting the plurality of client relationship subgraphs into a graph convolution network to perform feature extraction to obtain a second node feature, wherein the dimension of the second node feature is lower than that of the first node feature; and

and processing the second node characteristic by using a clustering algorithm to group the m clients.

2. The method of claim 1, wherein the converting the data in the dataset into first graph structure data comprises:

constructing the first node list and the first node characteristics based on the attribute information and the behavior data of the m clients;

constructing the first edge list based on transaction data among the m clients; and

the first edge weight is determined based on the transaction amount and the number of transactions in the transaction data.

3. The method of claim 2, wherein the importance sampling the customer relationship graph to obtain a plurality of customer relationship subgraphs comprises:

a random path acquisition step: starting from any node in the client relation graph, performing random walk for a plurality of times to obtain a plurality of random paths, wherein the node corresponding to the starting point is used as a central node;

Sampling the neighbor nodes: sampling neighbor nodes in each step according to the set wander length and the number of neighbor nodes for each random path in the plurality of random paths, and constructing a client relationship subgraph corresponding to each random path where the center node walks randomly; and

and repeating the random path acquisition step and the neighbor node sampling step for each node in the m nodes to obtain the plurality of client relationship subgraphs.

4. A method according to claim 3, wherein said sampling neighbor nodes in each step comprises:

selecting a part of neighbor nodes for sampling according to a first edge weight between the center node and the neighbor nodes to obtain a client relationship subgraph corresponding to each random path of the random walk of the center node, wherein,

the larger the first edge weight is, the higher the probability that the corresponding neighbor node is sampled.

5. The method of claim 3, wherein each of the customer relationship subgraphs includes a first node list and a first edge list, and the first node list and the first edge list corresponding to the plurality of customer relationship subgraphs are used as feature matrices of the m nodes.

6. The method of claim 5, wherein the feature extraction of the plurality of customer relationship subgraphs into the graph convolution network to obtain a second node feature comprises:

performing linear transformation on the feature matrix of the m nodes, and mapping node features to a new representation space;

splicing the initial node characteristics and the end node characteristics of each edge in the plurality of client relationship subgraphs to obtain an attention input matrix containing twice the output dimension;

calculating an attention score for each edge in the plurality of customer relationship subgraphs using the full connection layer;

performing edge normalization on the attention scores, wherein the sum of weights of all edges corresponding to each node is 1; the method comprises the steps of,

multiplying the neighbor node characteristics of each node by the corresponding attention weight to obtain second node characteristics.

7. The method of claim 6, wherein the graph convolution network comprises two graph convolution layers and two attention layers.

8. A customer grouping apparatus comprising:

the acquisition module is used for collecting attribute information and behavior data of m clients under the condition of acquiring authorization of the m clients to form a data set, wherein m is a positive integer greater than or equal to 3;

The conversion module is used for converting the data in the data set into first graph structure data, wherein the first graph structure data comprises a first node list, first node characteristics, a first edge list and first edge weights;

the construction module is used for constructing a customer relation graph according to the first graph structure data, wherein the customer relation graph comprises m nodes and n sides, the m nodes correspond to a plurality of nodes in the first node list and the first node characteristics, the n sides correspond to a plurality of first sides and the first side weights in the first side list, and n is a positive integer greater than or equal to 3;

the sampling module is used for sampling importance of the client relationship graph based on the weights of the n sides to obtain a plurality of client relationship subgraphs;

the feature extraction module is used for carrying out feature extraction on the plurality of client relationship subgraphs input into a graph convolution network to obtain a second node feature, wherein the dimension of the second node feature is lower than that of the first node feature; and

and the classification module is used for processing the second node characteristics by using a clustering algorithm so as to group the m clients.

9. An electronic device, comprising:

One or more processors; and

a memory for storing one or more programs,

wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of any of claims 1 to 7.

10. A computer readable storage medium storing computer executable instructions which, when executed, are adapted to carry out the method of any one of claims 1 to 7.

11. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 7.