Customer data classification method and related device

Abstract

The invention discloses a customer data classification method and a related device, and relates to the technical field of data processing and finance, customer data is acquired and then input into a customer segmentation model to obtain a customer segmentation result, and the customer segmentation model provided by the invention mainly comprises a graph convolutional neural network module and a feature extraction module. The finally obtained customer subdivision result is obtained by clustering topological information obtained through processing of the integrated graph convolutional neural network module and integrated customer information obtained through processing of the feature extraction module and represented by deep semantics, and a dual self-supervision mechanism is adopted in the clustering process. The customer segmentation model provided by the invention can pay attention to self information and topological information of the customer data at the same time, and a dual self-supervision mechanism is adopted for the two modules in the clustering process, so that it is ensured that the accuracy of the output customer segmentation result is relatively high.

Description

Technical Field

[0001] The present application relates to the field of data processing technology and financial technology, and in particular to a method for classifying customer data and related devices. Background Art

[0002] In modern finance, customer segmentation is a crucial task.

[0003] Customer segmentation aims to divide customers into different groups based on their behaviors, needs and characteristics. Through customer segmentation, companies can better understand customer needs and provide personalized services and products, thereby improving customer satisfaction and loyalty.

[0004] Traditional customer segmentation methods usually rely on static characteristics of customers (such as demographic data) and simple clustering algorithms (such as K-means). However, these traditional methods perform poorly when dealing with relationship-intensive data. In addition, with the emergence of new data sources such as social media and financial transaction networks, the interactive relationships between customers have become more complex. Traditional customer segmentation methods can no longer handle today's complex data well, and the accuracy of customer segmentation is low. Summary of the invention

[0005] In view of the above problems, this application provides a customer data classification method and related devices to achieve the purpose of improving the accuracy of customer segmentation. The specific scheme is as follows:

[0006] The first aspect of the present application provides a method for classifying customer data, comprising:

[0007] Access customer data;

[0008] The customer data is input into the customer segmentation model to obtain the customer segmentation result; the customer segmentation model includes a graph convolutional neural network module and a feature extraction module; the customer segmentation model is configured as follows: a customer relationship network is constructed based on the customer data, the customer data is input into the feature extraction module for processing, a deep semantic representation of the customer data is obtained, and the deep semantic representation is sent to the graph convolutional neural network module; the customer relationship network and the customer data are input into the graph convolutional neural network module for processing to obtain the topological information between the customer data, and the topological information and the deep semantic representation are integrated to obtain the integrated customer information; the deep semantic representation and the integrated customer information are clustered based on the dual self-supervision mechanism to obtain the clustering result; the clustering result is used as the customer segmentation result.

[0009] In a possible implementation, building a customer relationship network based on customer data includes:

[0010] Calculate the similarity between every two customers based on the customer data to obtain a similarity set;

[0011] Sort the similarities in the similarity set from high to low, retain the similarities of a previously preset proportion in the similarity sorting sequence, and use the previously preset proportion of similarities as the first similarity subset;

[0012] Sort the similarities between each customer and other customers from high to low to obtain similarity subsets corresponding to each customer;

[0013] Retaining a preset number of similarities in each similarity subset to obtain a second similarity subset corresponding to each customer;

[0014] A customer relationship network is constructed with the retained first similarity subset and each retained second similarity subset as edges.

[0015] In a possible implementation, the feature extraction module includes an autoencoder and a mapping head, and the graph convolutional neural network module includes a graph autoencoder and a graph mapping head. After integrating topological information and deep semantic representation, integrated customer information is obtained, including:

[0016] The deep semantic representation mapped by the mapping head and the topological information mapped by the graph mapping head are weightedly summed to obtain the integrated customer information.

[0017] In one possible implementation, clustering analysis of deep semantic representation and integrated customer information is performed based on a dual self-supervision mechanism, including:

[0018] After processing the deep semantic representation and the integrated customer information based on the k-means clustering algorithm, a first clustering result distribution corresponding to the deep semantic representation and a second clustering result distribution corresponding to the integrated customer information are obtained respectively;

[0019] The target distribution is obtained by normalizing the first clustering result distribution by square and sum;

[0020] Minimize a first divergence loss between the first clustering result distribution and the target distribution, and minimize a second divergence loss between the second clustering result distribution and the target distribution;

[0021] Updating the first clustering result distribution and the second clustering result distribution based on the first divergence loss and the second divergence loss so that the first clustering result distribution and the second clustering result distribution are infinitely close to the target distribution until a preset number of updates is reached;

[0022] The customer segmentation result is calculated based on the updated second clustering result distribution.

[0023] In one possible implementation, both the mapping head and the graph mapping head are multi-layer perceptrons.

[0024] A second aspect of the present application provides a device for classifying customer data, comprising:

[0025] Acquisition unit and customer classification unit; where:

[0026] An acquisition unit, used for acquiring customer data;

[0027] The customer classification unit is used to input customer data into the customer segmentation model to obtain customer segmentation results; the customer segmentation model includes a graph convolutional neural network module and a feature extraction module; the customer segmentation model is configured to: build a customer relationship network based on customer data, input customer data into the feature extraction module for processing, obtain a deep semantic representation of customer data, and send the deep semantic representation to the graph convolutional neural network module; input the customer relationship network and customer data into the graph convolutional neural network module for processing, obtain topological information between customer data, integrate the topological information and the deep semantic representation to obtain integrated customer information; perform cluster analysis on the deep semantic representation and the integrated customer information based on a dual self-supervision mechanism to obtain a clustering result; and use the clustering result as the customer segmentation result.

[0028] In a possible implementation, building a customer relationship network based on customer data in the customer classification unit includes:

[0029] Calculate the similarity between every two customers based on the customer data to obtain a similarity set;

[0030] Sort the similarities in the similarity set from high to low, retain the similarities of a previously preset proportion in the similarity sorting sequence, and use the previously preset proportion of similarities as the first similarity subset;

[0031] Sort the similarities between each customer and other customers from high to low to obtain similarity subsets corresponding to each customer;

[0032] Retaining a preset number of similarities in each similarity subset to obtain a second similarity subset corresponding to each customer;

[0033] A customer relationship network is constructed with the retained first similarity subset and each retained second similarity subset as edges.

[0034] A third aspect of the present application provides a classification device for customer data, comprising at least one processor and a memory connected to the processor, wherein:

[0035] The memory is used to store computer programs;

[0036] The processor is used to execute the computer program so that the customer data classification device can implement any one of the customer data classification methods described above.

[0037] A fourth aspect of the present application provides a computer program product, including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any one of the customer data classification methods described above.

[0038] A fifth aspect of the present application provides a computer storage medium, which carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the customer data classification methods described above.

[0039] By means of the above technical scheme, the customer data classification method and related devices provided by the present application obtain customer data and input them into the customer segmentation model to obtain customer segmentation results. The customer segmentation model provided by the present application mainly includes a graph convolutional neural network module and a feature extraction module. The final customer segmentation result is obtained by clustering the topological information obtained by the integrated graph convolutional neural network module and the integrated customer information represented by deep semantics obtained by the feature extraction module, and a dual self-supervision mechanism is adopted in the clustering process. It can be seen that the customer segmentation model provided by the present application can pay attention to the customer data's own information and topological information at the same time, and adopts a dual self-supervision mechanism for the two modules in the clustering process, thereby ensuring that the output customer segmentation results have a high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.

[0041] Figure 1 A flow chart of a method for classifying customer data provided for this application;

[0042] Figure 2 An example flow chart of the customer relationship network construction algorithm provided for this application;

[0043] Figure 3 A diagram showing how the customer segmentation model provided for this application works;

[0044] Figure 4 An example flow chart of a method for classifying customer data based on a graph convolutional neural network provided in this application;

[0045] Figure 5 A schematic diagram of the structure of a client data classification device provided in this application;

[0046] Figure 6 A schematic diagram of the structure of the customer data classification device provided for this application. DETAILED DESCRIPTION

[0047] The following describes the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. The terms used in the implementation method section of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0048] The embodiments of the present application are described below in conjunction with the accompanying drawings. Those skilled in the art will appreciate that, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0049] The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, which is only to describe the distinction mode adopted by the objects of the same attributes when describing in the embodiments of the present application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment comprising a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0050] Traditional customer data classification methods usually use static customer features and simple clustering algorithms. However, this method cannot handle today's increasingly complex customer data well, resulting in lower accuracy in customer segmentation.

[0051] In order to solve the above problems, the present application provides a method and related device for classifying customer data.

[0052] Optional, see Figure 1 , a flow chart of a method for classifying customer data provided in this application.

[0053] like Figure 1 As shown, the classification method of customer data includes the following steps:

[0054] Step 101: Obtain customer data.

[0055] Customer data generally includes basic customer information, such as the customer's name, gender, age, education, occupation, etc., which is used to understand the customer's basic background and demographic characteristics; contact information, such as telephone number, email address, social media account, etc., which is used to communicate and contact with customers; geographic location, such as the customer's place of residence, place of work, etc.; purchase records, such as the customer's past purchase records, including the products or services purchased, purchase time, purchase amount, etc., which are used to reflect the customer's consumption habits and preferences; interaction records between customers and companies, such as customer service inquiries, complaints, feedback, etc., which are used to understand customer needs and satisfaction; behavioral data, such as customers' browsing history, click behavior, visit duration, page dwell time, etc., which are used to reflect Customers' interests and concerns; loyalty data, customer loyalty indicators, such as repeat purchase rate, membership points, number of recommendations, etc., are used to evaluate customer loyalty and value; social media behavior, that is, customer behavior on social media, such as likes, comments, shares, etc., can reflect the customer's social influence and brand awareness; financial information, that is, the customer's financial status, such as income level, credit rating, etc.; risk preference, used to better recommend suitable financial products; transaction frequency and amount, important indicators for evaluating customer activity and contribution; account status and amount, that is, the account opening date, current amount, highest balance, lowest balance and other information, used to reflect the customer's asset status and capital flow; authorization level and customer status; etc.

[0056] From the specific content of customer data, we can know that customer data includes internal data, external data and real-time data. Internal data in customer data can be obtained from internal systems, such as core business systems, credit management systems, customer relationship management systems, financial systems and other systems; external data involved in customer data can be obtained from credit reporting agencies, social media, public data, and partners; real-time data in customer data can be obtained through real-time monitoring systems.

[0057] Optionally, obtain customer data from internal systems, external organizations, etc.

[0058] Step 102: input the customer data into the customer segmentation model to obtain the customer segmentation result; the customer segmentation model includes a graph convolutional neural network module and a feature extraction module; the customer segmentation model is configured to: build a customer relationship network based on the customer data, input the customer data into the feature extraction module for processing, obtain the deep semantic representation of the customer data, and send the deep semantic representation to the graph convolutional neural network module; input the customer relationship network and the customer data into the graph convolutional neural network module for processing, obtain the topological information between the customer data, integrate the topological information and the deep semantic representation to obtain the integrated customer information; perform cluster analysis on the deep semantic representation and the integrated customer information based on the dual self-supervision mechanism to obtain the clustering result; use the clustering result as the customer segmentation result.

[0059] After obtaining the customer data, it will be input into the customer segmentation model provided in this application. The customer segmentation model mainly includes two modules, namely a feature extraction module and a graph convolutional neural network module, and a self-supervised training module that can integrate the two modules together. This module can also make the final output results of the feature extraction module and the graph convolutional neural network module converge to the same; among them, the feature extraction module mainly includes an autoencoder and a mapping head, which are mainly used to extract deep semantic representations in customer data; the graph convolutional neural network module includes a graph autoencoder and a graph mapping head, which are used to capture the topological information between customer data.

[0060] When customer data is input into the customer segmentation model, the customer relationship network will be constructed using the customer relationship network construction algorithm. The steps for constructing the customer relationship network are as follows:

[0061] First, the customer data will be processed for dimensionality reduction. Specifically, the PCA (Principal Component Analysis) dimensionality reduction method can be used to reduce the dimensionality and obtain the reduced customer data.

[0062] Step 1: Calculate the similarity between every two customers based on the customer data to obtain a similarity set.

[0063] The cosine similarity between every two customer data is calculated to obtain a similarity set, which may be in the form of a similarity matrix.

[0064] Step 2: Sort the similarities in the similarity set from high to low, retain the similarities of a previously preset proportion in the similarity sorting sequence, and use the similarities of the previously preset proportion as the first similarity subset.

[0065] The similarity matrix is ​​expanded and sorted from high to low to obtain a similarity sorting sequence from high to low, and the similarities of the previous preset ratio in the similarity sorting sequence are retained. The previous preset ratio is a preset ratio value, for example, the preset ratio can be 20%, that is, the first 20% of the similarity values ​​in the similarity sorting sequence are retained, and the retained similarities constitute a first similarity subset. It can be understood that the two customer data corresponding to each similarity in the retained similarities are recognized as having a complex relationship, and an edge is used to connect the two customer nodes in the customer relationship network.

[0066] It should be noted that the customer data corresponding to each similarity retained may not cover all customer data, that is, there may be isolated nodes in the preliminary customer relationship network formed by the connection at this time. In order to solve this problem and reduce the number of isolated nodes in the preliminary customer relationship network, steps three and four are taken to continue to improve the preliminary customer relationship network.

[0067] Step 3: Sort the similarities between each customer and other customers from high to low to obtain similarity subsets corresponding to each customer.

[0068] Each customer data has similarity with other customer data. The similarity between each customer data and other customers is sorted from high to low to obtain a sorted similarity subset corresponding to each customer.

[0069] Step 4: retain the previously preset number of similarities in each similarity subset to obtain a second similarity subset corresponding to each customer.

[0070] The first preset number of similarities in the similarity subset of each customer data is retained to obtain a second similarity subset corresponding to each customer, wherein the preset number may be the first topk customer data in each second similarity subset.

[0071] Step 5: construct a customer relationship network using the retained first similarity subset and each retained second similarity subset as edges.

[0072] Each pair of customer data corresponding to each similarity in the retained first similarity subset is connected, and other customer data that do not belong to the first similarity subset among the top k customer data with the highest similarity corresponding to each customer data are connected to obtain a customer relationship network. It should be noted that the customer relationship network is an undirected graph.

[0073] For example, suppose that the customer data set contains N customers. Before building the customer relationship network, the customer data is transformed to obtain customer data X, and the cosine similarity between the customer data is calculated to obtain a similarity matrix S. The similarity matrix S is expanded and sorted to obtain a similarity sorting sequence. The preset ratio is r, that is, the number of retained similarities is , which corresponds to the number of edges of the initial customer relationship network to be constructed. The retained similarities form the first similarity subset. The last minimum similarity value in the first similarity subset can be used as the similarity threshold T. Suppose the similarity between customers is , represents the similarity value between customer i and customer j, when When the similarity is greater than or equal to the similarity threshold T, it is considered that there is a complex relationship between the two customers. In the graph, an edge is used to connect the two customer nodes. In order to reduce the number of isolated nodes in the customer relationship network, each customer node is also connected to its corresponding topk most similar customers. Finally, an undirected graph is constructed. express.

[0074] Optional, see Figure 2 , an example flow chart of the customer relationship network construction algorithm provided in this application.

[0075] like Figure 2 As shown in the figure, the customer relationship network construction algorithm processes the customer data by PCA dimensionality reduction to obtain the vector {X 1 , X 2 , …, X N}, the output customer relationship network G.

[0076] The number of samples is N, the ratio of retained edges is r, the k nearest neighbors is k, the similarity matrix is ​​S, and the similarity value between customer i and customer j is , the kth largest value among the similarities between node i and other nodes is Top(i,k), and the kth largest value among the similarities between node j and other nodes is Top(j,k).

[0077] First, initialize i=0, obtain the similarity threshold T according to the similarity matrix and the ratio r of the retained edges, and initialize j=i+1. >T, connect node i to node j. ≥Top(i,k)or ≥Top(j,k), connect node i to node j until i equals N and j equals N.

[0078] After obtaining the customer relationship network, the graph convolutional neural network module and the feature extraction module are used to process the customer relationship network and customer data. Among them, the feature extraction module processes the customer data, and the graph convolutional neural network module processes the customer relationship network and customer data.

[0079] Specifically, the feature extraction module is used to learn the deep semantic representation of customer data, which is mainly composed of two parts: the autoencoder and the mapping head. Assume that the autoencoder has a total of L layers, where l represents the specific number of layers. The content learned by the lth layer of the autoencoder can be expressed as:

[0080] ;

[0081] in, Represents activation functions, such as Relu and Sigmoid, represents the weight of the encoder at layer l, represents the bias of the encoder at layer l, represents the input data X. It should be noted that the mapping head of the feature extraction module can be a multi-layer perceptron, which is mainly used to map the features extracted by the autoencoder. The output of the autoencoder is recorded as , the final output of the feature extraction module can be expressed as:

[0082] .

[0083] Next, the graph convolutional neural network module is introduced. This module learns the complex topological information in the customer relationship network based on GCN (Graph Convolutional Network). Traditional GCN will pay too much attention to the topological information between customers and ignore some information about the customers themselves. In this application, in order to enable the graph convolutional neural network module to capture the topological information between customers while paying attention to some information about the customers themselves, a message passing mechanism is set up, and the feature extraction module integrates the learned deep semantic representation of customer data into the graph convolutional neural network module.

[0084] Specifically, the graph convolutional neural network module is mainly composed of two parts: the graph autoencoder and the graph mapping head. The graph autoencoder is stacked by multiple layers of GCN. In order to facilitate the message transmission between the feature extraction module and the graph convolutional neural network module, the mapping head in the feature extraction module corresponds to the graph mapping head in the graph convolutional neural network module. For a GCN module stacked by multiple layers of GCN, the content learned by the lth layer can be expressed as:

[0085] ;

[0086] in, represents the adjacency matrix of the customer relationship network G, represents the adjacency matrix of the customer relationship network G, is the matrix of the l-1th layer, is the weight of the l-1th layer autoencoder, , , for The value of the point on the i-th row and j-th column in , for The value of the point at the i-th row and j-th column in .

[0087] In order to prevent the GCN module from paying too much attention to the topological information on the customer relationship network, this application sets up two message passing mechanisms in the graph autoencoder and the graph mapping header, which can be expressed as:

[0088] ;

[0089] ;

[0090] in, represents the output of the graph autoencoder, represents the output of the autoencoder, represents the output of the graph mapping head, Represents the output of the mapping header, is a balance coefficient used to weigh the importance of the representation learned in the graph convolutional neural network module and the representation learned in the feature extraction module.

[0091] It can be understood that the setting of the feature extraction module and the message passing mechanism can prevent the graph convolutional neural network module from focusing too much on the topological information in the customer relationship network and ignoring the customer's own data information, so that the entire customer segmentation model can learn a more complete and meaningful deep semantic representation of customer data.

[0092] To summarize, the graph convolutional neural network module in the customer segmentation model receives the deep semantic representation extracted by the feature extraction module, integrates the deep semantic representation and the topological information captured by itself, and obtains integrated customer information.

[0093] Afterwards, the deep semantic representation and integrated customer information will be clustered and analyzed based on the dual self-supervision mechanism to obtain the clustering result, which is the customer segmentation result.

[0094] Firstly, after processing the deep semantic representation and the integrated customer information based on the k-means clustering algorithm, a first clustering distribution result corresponding to the deep semantic representation and a second clustering result distribution corresponding to the integrated customer information are obtained respectively.

[0095] Assume that the distribution of the first clustering result is Q and the distribution of the second clustering result is Z. Specifically, the customer segmentation model uses the Student t distribution to measure the data representation of the i-th sample. and the jth cluster center vector The similarity between them is as follows, assuming that the formula is Formula 1:

[0096] ;

[0097] in, is the output of the autoencoder in the feature extraction module The i-th sample of is the split center initialized by the k-means clustering algorithm, t represents the degrees of freedom of the Student's t distribution, represents the probability of the i-th customer being assigned to the j-th cluster. In order to make the data closer to the cluster center, the target distribution P is obtained by normalizing the square and sum of each distribution in Q using the following formula, which is assumed to be Formula 2:

[0098] ;

[0099] in, , is the overall frequency of all users assigned to the jth cluster.

[0100] Subsequently, by minimizing the KL divergence loss between the first clustering result distribution Q and the target distribution P, the target distribution P helps the feature extraction module learn a better clustering task representation, i.e., move the customer representation closer to the cluster center.

[0101] It can be seen that the target distribution P is calculated through the first clustering result distribution Q, and the target distribution P can supervise the update of the first clustering result distribution Q, which is the first self-supervision mechanism.

[0102] The specific formula can be as follows, assuming that the formula is Formula 3:

[0103] ;

[0104] The second clustering result distribution is recorded as Z, which can be obtained by the following formula, assuming that the formula is Formula 4:

[0105] ;

[0106] Among them, softmax() is the activation function and the matrix of the Lth layer of the autoencoder.

[0107] Then, the second divergence loss between the second clustering result distribution Z and the target distribution P is minimized, so that the target distribution P supervises the update of the clustering result distribution Z. This is the second self-supervision mechanism formula, which can be as follows, assuming that the formula is Formula 5:

[0108] ;

[0109] Through the dual self-supervision mechanism, the customer segmentation model can consider the topological information between customers while considering the information of the customer itself. At the same time, since the feature extraction module and the GCN module are both close to the target distribution P, the results of the two modules after training tend to be consistent. The overall loss function of the customer segmentation model is as follows, assuming that the formula is Formula 6:

[0110] ;

[0111] Among them, L is the overall loss function of the customer segmentation model, and α is the weight parameter.

[0112] Optionally, the first clustering result distribution and the second clustering result distribution are updated based on the first divergence loss and the second divergence loss, so that the first clustering result distribution and the second clustering result distribution are infinitely close to the target distribution until a preset number of updates is reached. Finally, the customer segmentation result is calculated based on the updated second clustering result distribution.

[0113] To summarize, the customer data classification method provided in the present application obtains customer data and then inputs it into the customer segmentation model to obtain customer segmentation results. The customer segmentation model provided in the present application mainly includes a graph convolutional neural network module and a feature extraction module. The final customer segmentation result is obtained by clustering the topological information obtained by the integrated graph convolutional neural network module and the integrated customer information represented by deep semantics obtained by the feature extraction module, and a dual self-supervision mechanism is adopted in the clustering process. It can be seen that the customer segmentation model provided in the present application can pay attention to the customer data's own information and topological information at the same time, and adopts a dual self-supervision mechanism for the two modules in the clustering process, thereby ensuring that the output customer segmentation results have a high degree of accuracy.

[0114] Optional, see Figure 3 , a working principle diagram of the customer segmentation model provided in this application.

[0115] like Figure 3As shown, first, the customer data is processed into a customer relationship network, the customer relationship network and the customer data are input into the GCN module, and the customer data is input into the autoencoder module. It should be noted that the autoencoder module is the feature extraction module. The deep semantic representation extracted by the feature extraction module is message-transmitted and input into the GCN module. The GCN module combines the topological information obtained by itself with the deep semantic representation to obtain the integrated customer information, and calculates the clustering result distribution Z. The feature extraction module calculates the clustering result distributions Q and P. The losses are calculated for each module respectively, and the parameters of the customer segmentation model are updated according to the inspections in the two modules. Finally, the segmentation results of the customer data are calculated based on the updated clustering result distribution Z.

[0116] In summary, the present application learns the inherent information of customer data through the feature extraction module, and enables the customer segmentation model to pay attention to the topological information in the customer relationship network and the inherent information of customer data through the message transmission mechanism.

[0117] Correspondingly, see Figure 4 , an example flow chart of a method for classifying customer data based on a graph convolutional neural network provided in this application.

[0118] Among them, the vector obtained by PCA dimensionality reduction is {X 1 , X 2 , …, X N}, the customer relationship network dimension is represented by an undirected graph , where E is the number of training times, K is the number of clusters, and the cluster assignment is R.

[0119] First, the K-means algorithm is used to initialize the cluster center μ, and the customer data is input into the autoencoder module and the mapping head to obtain and According to Formula 1 and Formula 2 above, the first clustering result distribution Q and target distribution P are calculated. The autoencoder module transmits messages to the GCN module. The GCN module calculates the second clustering result distribution Z. The loss L is calculated using Formula 3 and Formula 5 above. clu and L gcn , based on these two losses, update the parameters in the classification method of customer data until the number of training times E is reached, and calculate the cluster allocation result R according to Formula 6 above.

[0120] A method for classifying customer data provided in an embodiment of the present application is introduced above. A device for executing the method for classifying customer data is introduced below.

[0121] See also Figure 5 , Figure 5 This is a schematic diagram of the structure of a customer data classification device provided by this application. Figure 5As shown, the device comprises:

[0122] Acquisition unit 10 and customer classification unit 20; wherein:

[0123] An acquisition unit 10, used to acquire customer data;

[0124] The customer classification unit 20 is used to input customer data into the customer segmentation model to obtain customer segmentation results; the customer segmentation model includes a graph convolutional neural network module and a feature extraction module; the customer segmentation model is configured to: build a customer relationship network based on customer data, input the customer data into the feature extraction module for processing, obtain a deep semantic representation of the customer data, and send the deep semantic representation to the graph convolutional neural network module; input the customer relationship network and customer data into the graph convolutional neural network module for processing, obtain topological information between the customer data, integrate the topological information and the deep semantic representation to obtain integrated customer information; perform cluster analysis on the deep semantic representation and the integrated customer information based on a dual self-supervision mechanism to obtain a clustering result; and use the clustering result as the customer segmentation result.

[0125] In one embodiment, the customer classification unit 20 is specifically used for:

[0126] Calculate the similarity between every two customers based on the customer data to obtain a similarity set;

[0127] Sort the similarities in the similarity set from high to low, retain the similarities of a previously preset proportion in the similarity sorting sequence, and use the previously preset proportion of similarities as the first similarity subset;

[0128] Sort the similarities between each customer and other customers from high to low to obtain similarity subsets corresponding to each customer;

[0129] Retaining a preset number of similarities in each similarity subset to obtain a second similarity subset corresponding to each customer;

[0130] A customer relationship network is constructed with the retained first similarity subset and each retained second similarity subset as edges.

[0131] In one embodiment, the feature extraction module in the customer classification unit 20 includes an autoencoder and a mapping head, and the graph convolutional neural network module includes a graph autoencoder and a graph mapping head, which are specifically used for:

[0132] The deep semantic representation mapped by the mapping head and the topological information mapped by the graph mapping head are weightedly summed to obtain the integrated customer information.

[0133] In one embodiment, the customer classification unit 20 is specifically used for:

[0134] After processing the deep semantic representation and the integrated customer information based on the k-means clustering algorithm, a first clustering result distribution corresponding to the deep semantic representation and a second clustering result distribution corresponding to the integrated customer information are obtained respectively;

[0135] The target distribution is obtained by normalizing the first clustering result distribution by square and sum;

[0136] Minimize a first divergence loss between the first clustering result distribution and the target distribution, and minimize a second divergence loss between the second clustering result distribution and the target distribution;

[0137] Updating the first clustering result distribution and the second clustering result distribution based on the first divergence loss and the second divergence loss so that the first clustering result distribution and the second clustering result distribution are infinitely close to the target distribution until a preset number of updates is reached;

[0138] The customer segmentation result is calculated based on the updated second clustering result distribution.

[0139] In one embodiment, the mapping head and the graph mapping head in the customer classification unit 20 are both multi-layer perceptrons.

[0140] The present application also provides a customer data classification device. Figure 6 As shown, it shows a schematic diagram of the structure of a classification device for customer data provided by the present application. The classification device for customer data in the embodiment of the present application may include but is not limited to fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 6 The customer data classification device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0141] like Figure 6 As shown, the client data classification device may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 to a random access memory (RAM) 603. When the client data classification device is powered on, various programs and data required for the operation of the client data classification device are also stored in the RAM 603. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0142] Typically, the following devices may be connected to the I / O interface 605: input devices 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 608 including, for example, a memory card, a hard disk, etc.; and communication devices 609. The communication devices 609 may allow the classification device of the customer data to communicate with other devices wirelessly or by wire to exchange data. Although Figure 6 The classification device of customer data having various devices is shown, but it should be understood that it is not required to implement or have all the devices shown. More or fewer devices may be implemented or have instead.

[0143] An embodiment of the present application also provides a computer program product including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any one of the customer data classification methods provided in the embodiments of the present application.

[0144] A computer-readable storage medium is also provided in an embodiment of the present application. The storage medium carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any customer data classification method provided in the embodiment of the present application.

[0145] It should also be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the device embodiments provided by the present application, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines.

[0146] Through the description of the above implementation mode, the technicians in the field can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a ROM, a RAM, a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a training device, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0147] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0148] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, a computer, a training device or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, training device or data center. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a training device, a data center, etc. that includes one or more available media integration. Available media can be magnetic media, (e.g., floppy disk, hard disk, tape), optical media (e.g., DVD), or semiconductor media (e.g., solid-state hard disk (SSD)), etc.

Claims

1. A method for classifying customer data, characterized in that: include: Access customer data; Inputting the customer data into a customer segmentation model to obtain a customer segmentation result; The customer segmentation model includes a graph convolutional neural network module and a feature extraction module; The customer segmentation model is configured as follows: constructing a customer relationship network based on the customer data, inputting the customer data into the feature extraction module for processing to obtain a deep semantic representation of the customer data, and sending the deep semantic representation to the graph convolutional neural network module; inputting the customer relationship network and the customer data into the graph convolutional neural network module for processing to obtain topological information between the customer data, integrating the topological information and the deep semantic representation to obtain integrated customer information; performing cluster analysis on the deep semantic representation and the integrated customer information based on a dual self-supervision mechanism to obtain a clustering result; and using the clustering result as the customer segmentation result.

2. The method for classifying customer data according to claim 1, characterized in that: The step of building a customer relationship network based on the customer data includes: Calculate the similarity between every two customers based on the customer data to obtain a similarity set; Sort the similarities in the similarity set from high to low, retain a previously preset ratio of similarities in the similarity sorting sequence, and use the previously preset ratio of similarities as a first similarity subset; Sort the similarities between each customer and other customers from high to low to obtain similarity subsets corresponding to each customer; Retaining a preset number of similarities in each similarity subset to obtain a second similarity subset corresponding to each customer; A customer relationship network is constructed with the retained first similarity subset and each of the retained second similarity subsets as edges.

3. The method for classifying customer data according to claim 1, characterized in that: The feature extraction module includes an autoencoder and a mapping head, the graph convolutional neural network module includes a graph autoencoder and a graph mapping head, and the integrated customer information is obtained by integrating the topological information and the deep semantic representation, including: The deep semantic representation mapped by the mapping head and the topological information mapped by the graph mapping head are weightedly summed to obtain the integrated customer information.

4. The method for classifying customer data according to claim 1, characterized in that: The clustering analysis of the deep semantic representation and the integrated customer information based on the dual self-supervision mechanism includes: After processing the deep semantic representation and the integrated customer information based on a k-means clustering algorithm, a first clustering result distribution corresponding to the deep semantic representation and a second clustering result distribution corresponding to the integrated customer information are obtained respectively; Performing square and normalization processing on the first clustering result distribution to obtain a target distribution; Minimize a first divergence loss between the first clustering result distribution and the target distribution, and minimize a second divergence loss between the second clustering result distribution and the target distribution; updating the first clustering result distribution and the second clustering result distribution based on the first divergence loss and the second divergence loss, so that the first clustering result distribution and the second clustering result distribution are infinitely close to the target distribution until a preset number of updates is reached; The customer segmentation result is calculated based on the updated second clustering result distribution.

5. The method for classifying customer data according to claim 3, characterized in that: The mapping head and the graph mapping head are both multi-layer perceptrons.

6. A classification device for customer data, characterized in that: include: Acquisition unit and customer classification unit; where: The acquisition unit is used to acquire customer data; The customer classification unit is used to input the customer data into the customer segmentation model to obtain customer segmentation results; the customer segmentation model includes a graph convolutional neural network module and a feature extraction module; the customer segmentation model is configured to: build a customer relationship network based on the customer data, input the customer data into the feature extraction module for processing, obtain a deep semantic representation of the customer data, and send the deep semantic representation to the graph convolutional neural network module; input the customer relationship network and the customer data into the graph convolutional neural network module for processing to obtain topological information between the customer data, integrate the topological information and the deep semantic representation to obtain integrated customer information; perform cluster analysis on the deep semantic representation and the integrated customer information based on a dual self-supervision mechanism to obtain a clustering result; and use the clustering result as the customer segmentation result.

7. The customer data classification device according to claim 6, characterized in that: The step of constructing a customer relationship network based on the customer data in the customer classification unit includes: Calculate the similarity between every two customers based on the customer data to obtain a similarity set; Sort the similarities in the similarity set from high to low, retain a previously preset ratio of similarities in the similarity sorting sequence, and use the previously preset ratio of similarities as a first similarity subset; Sort the similarities between each customer and other customers from high to low to obtain similarity subsets corresponding to each customer; Retaining a preset number of similarities in each similarity subset to obtain a second similarity subset corresponding to each customer; A customer relationship network is constructed with the retained first similarity subset and each of the retained second similarity subsets as edges.

8. A classification device for customer data, characterized in that: The method comprises at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is configured to execute the computer program so that the customer data classification device can implement the customer data classification method according to any one of claims 1 to 5.

9. A computer program product, characterized in that The method comprises computer-readable instructions, and when the computer-readable instructions are executed on an electronic device, the electronic device implements the method for classifying customer data as claimed in any one of claims 1 to 5.

10. A computer storage medium, characterized in that: The storage medium carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device can implement the customer data classification method as described in any one of claims 1 to 5.

Images