CN116883069A

CN116883069A - Information pushing method, device, computer equipment and storage medium

Info

Publication number: CN116883069A
Application number: CN202311066040.1A
Authority: CN
Inventors: 陈灏然; 董琳珏; 李蓝林; 宋林忆
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2023-08-23
Filing date: 2023-08-23
Publication date: 2023-10-13

Abstract

The application relates to an information pushing method, an information pushing device, a computer device, a storage medium and a computer program product, wherein the information pushing method comprises the following steps: acquiring first context information of a customer to be tested, acquiring second context information of two attribute graphs based on the two attribute graphs of the customer to be tested and a merchant, inputting the first context information and the second context information of the customer to be tested into a pre-trained graph translation model, and determining a probability value of the customer to be tested belonging to a bank card information pushing customer by the graph translation model through the first context information and the second context information; under the condition that the probability value output by the graph translation model is larger than a preset value, adding the client to be tested to a target client list; pushing bank card information to the clients in the target client list; and whether the client to be tested belongs to the target client or not is obtained based on the big data and the pre-trained graph translation model, and then information pushing is carried out on the target client, so that the requirement of screening the target client on manpower is reduced, and the accuracy of information pushing is improved.

Description

Information pushing method, device, computer equipment and storage medium

Technical Field

The present application relates to the field of big data technology, and in particular, to an information pushing method, an apparatus, a computer device, a storage medium, and a computer program product.

Background

In the conventional technology, for example, in the aspect of using a financial card, a plurality of target clients are screened manually based on personal information of the clients and the situation of using the card mainly through expert knowledge, and then the clients are informed of taking the financial card related activities through a specific information pushing mode. However, as the number of customers of financial cards increases, the screening work consumes huge manpower and time costs, and the efficiency is also difficult to meet the actual application demands. With the development of big data technology, it is considered that a method for screening a customer group can be provided based on big data so as to screen out corresponding customer groups according to limiting conditions; therefore, it is desirable to provide an information pushing method based on big data to push information to a target guest group.

Disclosure of Invention

In view of the foregoing, it is desirable to provide an information push method, apparatus, computer device, computer readable storage medium, and computer program product that can screen out a desired customer group based on big data.

In a first aspect, the present application provides an information pushing method, where the method includes:

acquiring first context information of a customer to be tested; the first context information comprises at least one of basic characteristic information and bank card transaction information of the client to be tested and clients associated with the client to be tested;

acquiring second context information of the two attribute graphs based on the two attribute graphs of the client to be tested and the merchant, which are constructed in advance; the graph elements of the two attribute graphs comprise nodes and connecting edges, the nodes comprise two types of clients and merchants, and the connecting edges exist between the nodes of different types and are used for representing bank card transaction information between the corresponding clients and merchants; the second context information comprises attribute information of each graph element in the graph; the attribute information comprises corresponding graph element category information, corresponding association relation between nodes and continuous edges and basic characteristic information of each graph element;

inputting the first context information of the client to be tested and the second context information of the two attribute graphs into a pre-trained graph translation model, wherein the graph translation model determines a probability value of the client to be tested belonging to a bank card information pushing client through the first context information and the second context information;

Adding the client to be tested to a target client list under the condition that the probability value output by the graph translation model is larger than a preset value; and pushing the bank card information to the clients in the target client list.

In one embodiment, the step of obtaining the second context information includes:

respectively acquiring attribute information of customer class nodes of the customer to be tested in the two attribute diagrams, attribute information of merchant class nodes of all merchants transacted with the customer to be tested with a bank card, and attribute information of all continuous edges corresponding to the customer to be tested and the merchants, and respectively representing the attribute information as vectorsAnd->Wherein c is a customer identifier, b is a merchant identifier, t is a continuous edge identifier, m represents an mth merchant, k represents a kth continuous edge, and v represents a customer to be tested;

splicing the vectors to obtain the second context information; the second context information comprises vectors of M merchants and vectors of K continuous edges, M is more than or equal to 2, K is more than or equal to 2, M is more than or equal to 1 and less than or equal to M, and K is more than or equal to 1 and less than or equal to K.

In one embodiment, the obtaining the attribute information of the customer class node of the customer to be tested, the attribute information of the merchant class node of each merchant having a bank card transaction with the customer to be tested, and the attribute information of each connecting edge corresponding to the customer to be tested and the merchant in the two attribute graphs includes:

The basic characteristic information of the customer to be tested, the merchant and the continuous edge is obtained and respectively expressed as initial vectorsAnd->

Adding corresponding category information to the initial vectors to obtain respective primary augmented vectors And->

Respectively toAdding the corresponding association relation information of the nodes and the connecting edges to obtain respective secondary augmentation vectors +.>

The corresponding category information is one of a client node category, a merchant node category and a connecting edge category.

In one embodiment, the adding of the corresponding category information to the initial vector respectively obtains respective primary augmentation vectorsAnd->Comprising the following steps:

three category identification vectors are obtained and respectively correspond to the client node category, the merchant node category and the link category;

splicing the category identification vector corresponding to the category of the client node with the initial vector of the client to be tested to obtain the primary augmentation vector of the client to be tested

Splicing the class identification vector corresponding to the class of the merchant node with the initial vector of the merchant to obtain the primary augmentation vector of the merchant

Splicing the class identification vector corresponding to the continuous edge class with the initial vector of the continuous edge to obtain the primary augmentation vector of the continuous edge

In one embodiment, the respective pairsAdding the corresponding association relation information of the nodes and the connecting edges to obtain respective secondary augmentation vectors +.>Comprising the following steps:

acquiring a first orthogonal node identification matrix and a second orthogonal node identification matrix, wherein the first orthogonal node identification matrix is used for identifying the association relation information of the client class node and the corresponding connecting edge, and the second orthogonal node is used for identifying the association relation information of the merchant class node and the corresponding connecting edge;

the first orthogonal node identification matrix is matched with the client to be testedThe one-time augmentation vectorSplicing to obtain the secondary augmentation vector ++ ++of the client to be tested>

Combining the second orthogonal node identification matrix with the one-time augmented vector of the merchantSplicing to obtain the secondary augmentation vector of the merchant>

The first orthogonal node identification matrix and the second orthogonal node identification matrix are connected with the primary augmentation vector of the connecting edgeSplicing to obtain the secondary augmentation vector ++ ++of the continuous edge>

In one embodiment, the graph translation model includes an encoder and a decoder;

the inputting the first context information of the customer to be tested and the second context information of the two-part attribute graph into a pre-trained graph translation model comprises the following steps:

Inputting the first context information of the customer under test to the decoder of the pre-trained graph translation model, and inputting the second context information of the two attribute graphs of the customer under test to the encoder of the pre-trained graph translation model;

the pre-trained graph translation model is specifically used for processing the second context information through the encoder, calculating the processed information based on an attention mechanism, outputting a first calculation result to a forward propagation layer of the encoder, and performing nonlinear change calculation and mapping on the first calculation result to obtain a first low-dimensional vector; and inputting the first low-dimensional vector and the first context information into a decoder, outputting a second calculation result to a forward propagation layer of the decoder based on an attention mechanism, performing nonlinear change calculation on the second calculation result, mapping to obtain a second low-dimensional vector, and obtaining a probability value of the customer to be tested belonging to a bank card information pushing customer based on the output of the decoder.

In one embodiment, the graph translation model is trained by:

Constructing a sample two-part attribute map of a sample client and a sample merchant, and acquiring third context information of the sample two-part attribute map;

inputting the third context information into a graph translation model to be trained, and outputting a prediction result of the sample client belonging to a bank card information pushing client by the graph translation model to be trained;

determining a current loss function value of the model; the loss function value is determined based on a preconfigured two-class cross entropy loss function, the prediction result currently output by a model, a first label of the sample client participating in the information push or a second label of the sample client not participating in the information push;

and adjusting model parameters based on the current loss function value until training is finished, and obtaining the pre-trained graph translation model.

In a second aspect, the present application further provides an information pushing apparatus, where the apparatus includes:

the information acquisition module is used for acquiring first context information of the client to be tested; the first context information comprises at least one of basic characteristic information and bank card transaction information of the client to be tested and clients associated with the client to be tested; acquiring second context information of the two attribute graphs based on the two attribute graphs of the client to be tested and the merchant, which are constructed in advance; the graph elements of the two attribute graphs comprise nodes and connecting edges, the nodes comprise two types of clients and merchants, and the connecting edges exist between the nodes of different types and are used for representing bank card transaction information between the corresponding clients and merchants; the second context information comprises attribute information of each graph element in the graph; the attribute information comprises corresponding graph element category information, corresponding association relation between nodes and continuous edges and basic characteristic information of each graph element;

The result calculation module is used for inputting the first context information of the client to be tested and the second context information of the two attribute graphs into a pre-trained graph translation model, and the graph translation model determines a probability value of the client to be tested belonging to a bank card information pushing client through the first context information and the second context information;

the recommendation module is used for adding the clients to be tested to a target client list under the condition that the probability value output by the graph translation model is larger than a preset value; and pushing the bank card information to the clients in the target client list.

In a third aspect, the present application also provides a computer device comprising a memory and a processor, the memory storing a computer program, the processor implementing the following steps when executing the computer program:

In a fourth aspect, the present application also provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:

In a fifth aspect, the application also provides a computer program product comprising a computer program which, when executed by a processor, performs the steps of:

The information pushing method, the information pushing device, the computer equipment, the storage medium and the computer program product are characterized in that first context information of a client to be tested is obtained, second context information is obtained based on two attribute diagrams of the client to be tested and a merchant, the first context information of the client to be tested and the second context information of the two attribute diagrams are input into a pre-trained diagram translation model, the diagram translation model determines and obtains the probability value of the client to be tested belonging to a bank card information pushing client through the first context information and the second context information, and the client to be tested is added into a target client list under the condition that the probability value is larger than a preset value, so that the bank card information is pushed to the client in the target client list; according to the method, whether the client to be tested belongs to the target client or not can be obtained based on the big data and the pre-trained graph translation model, so that information pushing is carried out on the target client, the requirement of screening the target client on manpower is reduced, and the accuracy of information pushing is improved.

Drawings

FIG. 1 is an application environment diagram of an information pushing method in an embodiment of the present application;

fig. 2 is a flowchart of a first embodiment of an information pushing method according to an embodiment of the present application;

fig. 3 is a flowchart of a second embodiment of an information pushing method according to an embodiment of the present application;

fig. 4 is a flowchart of a third embodiment of an information pushing method according to an embodiment of the present application;

fig. 5 is a flowchart of a fourth embodiment of an information pushing method according to an embodiment of the present application;

fig. 6 is a flowchart of a fifth embodiment of an information pushing method according to an embodiment of the present application;

fig. 7 is a flowchart of a sixth embodiment of an information pushing method according to an embodiment of the present application;

FIG. 8 is a block diagram of a graph translation model provided by an embodiment of the present application;

FIG. 9 is a flowchart of a training process of the graph translation model according to an embodiment of the present application;

FIG. 10 is a schematic diagram of two attribute map context information according to an embodiment of the present application;

FIG. 11 is a schematic diagram of an information pushing device according to an embodiment of the present application;

fig. 12 is an internal structure diagram of a computer device according to an embodiment of the present application.

Detailed Description

The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.

With the continuous progress and development of society, the consumption patterns of people are also changing increasingly. Among them, financial cards such as bank cards and credit cards have become a credit product commonly used in daily life, and are mainly used in the process of paying for goods. The credit card is different from the traditional consumption mode, and the consumption mode based on the social credit system allows customers to carry out overdraft consumption in a certain limit range, so that the consumption requirements of people can be met, the consumption capability of people can be greatly stimulated, and the development of economy and society is accelerated.

In order to enable more people to accept and use different kinds of financial cards, a good card utilization ecology is built, a card issuing mechanism needs to adopt a plurality of marketing modes, and the utilization rate of the different kinds of financial cards in daily life of people is improved, wherein issuing of the card utilization benefits is a method capable of directly pushing the card utilization of clients, a target guest group is firstly screened out from the total clients, and then a notice for acquiring the card utilization benefits is pushed to the target guest group through a specific channel (such as a short message); because these benefits may offer certain benefits to customers in certain consumption scenarios, customers may be directed to use more of the financial cards.

However, because the consumption concept and the demand of each customer are inconsistent, a specific consumption scenario cannot stimulate the consumption ability of all customers, so it is very important how to accurately screen out the marketing guest group in order to ensure that the marketing campaign is taken with the card equity to achieve the best effect. With the development of big data technology, it is considered that a method for screening a customer group can be provided based on big data so as to screen out corresponding customer groups according to limiting conditions; therefore, it is desirable to provide an information pushing method based on big data to push information to a target guest group.

The information pushing method provided by the embodiment of the application can be applied to an application environment shown in fig. 1. The terminal 12 communicates with the server 14 through a network, and the data storage system may store data that needs to be processed by the server 14, for example, the data storage system may be used to store customer information, merchant information, bank card transaction information between a customer and a merchant, and the like, and the data storage system may be integrated on the server 14, or may be placed on a cloud or other network servers. The terminal 12 may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, internet of things devices, and portable wearable devices, where the internet of things devices may be smart speakers, smart televisions, smart air conditioners, smart car-mounted devices, and the portable wearable devices may be smart watches, smart bracelets, headsets, and the like. The server 14 may be implemented as a stand-alone server or as a cluster of servers.

Fig. 2 is a flowchart of a first embodiment of an information pushing method according to an embodiment of the present application, in an embodiment, as shown in fig. 2, an information pushing method 100 is provided, and the method is applied to the server 14 for illustration, and the method includes:

step 101, acquiring first context information of a customer to be tested; the first context information comprises at least one of basic characteristic information and bank card transaction information of a customer to be tested and a customer associated with the customer to be tested.

Specifically, the information pushing method 100 provided by the present application at least includes steps 101-104, where step 101 is configured to obtain first context information of a to-be-tested client, where the first context information may include at least basic feature information of the to-be-tested client, bank card transaction information thereof, and further may include basic feature information of a client having an association relationship with the to-be-tested client, and bank card transaction information thereof, where the client having an association relationship with the to-be-tested client is, for example, friends, relatives, colleagues, etc. of the to-be-tested client. It should be noted that, the first context information may only obtain basic feature information of the customer to be tested and the customer associated with the customer according to the requirement, where the basic feature information may include, but is not limited to, occupation, income, consumption habit, age; the first context information can also only acquire the bank card transaction information of the customer to be tested and the customer related to the customer to be tested according to the requirement, wherein the bank card transaction information can include, but is not limited to, transaction amount, transaction time and type of purchased goods; the first context information can also obtain the basic feature information and bank card transaction information of the customer to be tested, the customer related to the customer to be tested and the customer related to the customer at the same time according to the requirement, and the application is not limited in particular. By acquiring the information of the client to be tested and the related clients, the consumption capability, consumption habit, consumption place and the like of the client to be tested can be predicted more accurately through the information of the client to be tested and the information of the client related to the client to be tested, so that the client to be tested can push more accurate information for the client to be tested later.

It should be further noted that the first context information may further include basic information features of the merchant consumed by the customer to be tested, such as a class of the merchant, a business condition, and the like, and may further include basic information of other customers related to the merchant consumed by the customer to be tested.

Step 102, based on a pre-constructed two-part attribute diagram of a customer to be tested and a merchant, obtaining second context information of the two-part attribute diagram; the graph elements of the two attribute graphs comprise nodes and connecting edges, wherein the nodes comprise two types of clients and merchants, and the connecting edges exist between the nodes of different types and are used for representing bank card transaction information between the corresponding clients and merchants; the second context information comprises attribute information of each picture element in the picture; the attribute information comprises corresponding graph element category information, corresponding association relation between nodes and continuous edges and basic characteristic information of each graph element.

Specifically, after the first context information of the customer to be tested is obtained in step 101, the second context information may be further obtained in step 102, where the second context information is obtained based on a two-part attribute map, and the two-part attribute map is specifically a two-part attribute map of the customer to be tested and the merchant; it should be noted that, the graph elements of the two attribute graphs include nodes and connecting edges, the nodes include two types of clients and merchants, the connecting edges exist between the nodes of different types and are used for representing transaction information between the clients and merchants which are correspondingly connected; the second context information comprises attribute information of each picture element in the picture, wherein the attribute information comprises corresponding picture element category information, corresponding association relation between nodes and continuous edges and basic characteristic information of each picture element; the category information of the graph elements can be used for distinguishing clients, merchants and continuous edges; the basic information features of each drawing element include, for example, occupation, income, consumption habit, age, etc. of the customer, merchant category, business condition, etc. of the merchant, transaction amount, transaction time, type of commodity purchased, etc. on the side. The second context information obtained in the step 102 can be used for realizing information between the customer to be detected and the merchant with the consumption relationship, so that more accurate prediction on the consumption capability, consumption habit, consumption place and the like of the customer to be detected is improved, and more accurate information is pushed to the customer to be detected in the follow-up process.

Step 103, inputting the first context information of the customer to be tested and the second context information of the two attribute graphs into a pre-trained graph translation model, and determining the probability value of the customer to be tested belonging to the bank card information pushing customer by the graph translation model through the first context information and the second context information.

Specifically, after the first context information and the related second context information of the to-be-tested client are obtained, the first context information and the second context information are further input into a pre-trained (already trained) graph translation model through step 103, the probability value of the to-be-tested client belonging to the bank card information pushing client is calculated through the graph translation model, and the advanced prediction of the subsequent consumption of the to-be-tested client based on the information of the to-be-tested client, the information of the merchant associated with the to-be-tested client, the specific transaction condition and other information is realized, so that the bank system corresponding to the bank card can push the related activity information more interested by the to-be-tested client for the to-be-tested client, the related activity information pushed by the bank system corresponding to the bank card for the to-be-tested client is more likely to be participated by the client, and the utilization rate of the bank card is improved through the pushed activity information.

Step 104, adding the client to be tested to the target client list under the condition that the probability value output by the graph translation model is larger than a preset value; and pushing the bank card information to the clients in the target client list.

Specifically, after the probability value of the client to be tested belonging to the bank card information pushing client is obtained in step 103, the probability value obtained in step 104 is compared with the preset value, and when the probability value output by the graph translation model is larger than the preset value, the client to be tested is added to the target client list, and the possibility that the client in the target client list receives and uses relevant activity information is larger than that of other clients, so that the bank card information pushing is performed to the client in the target client list, the pushing accuracy of the activity can be improved, and the cost input of manpower and time required by the activity pushing is reduced.

That is, the application takes the influence of individual behaviors of clients, information of other clients related to the clients, relevant transaction merchants and other factors into consideration, and rapidly and accurately screens a certain item or a plurality of target clients suitable for activities from a whole number of clients from consideration of various factors, and further carries out the pushing of the activity information to the clients in the target client list in a specific information pushing mode, so that the activity has a great probability of bringing a certain benefit to the target users, and is beneficial to improving the participation probability of the relevant activity information by the clients, thereby improving the utilization rate of bank cards, and simultaneously the whole pushing process of the activity information can reduce the cost investment of manpower and time.

It is also necessary to supplement that if only the individual behaviors of the clients are considered, because the consumption concept and the demand of each client are inconsistent, a certain specific activity does not necessarily stimulate the participation enthusiasm of the client, so that on the basis of considering the individual behaviors of the clients, the application further considers the information of other clients related to the clients and factors such as related transaction merchants, and the like, thereby avoiding the influence of small probability events in the individual behaviors of the clients on the probability of the clients to participate in a certain activity, and further improving the screening accuracy of a target client suitable for a certain activity or multiple activities.

It should be added that the present application provides a value selectable by the preset value of 0.5, but the present application is not limited thereto, and the preset value may be selected according to actual requirements, for example, 0.4, 0.48, 0.55, 0.6, 0.63, 0.65, etc.

And (3) repeating the steps 101-104 for all clients, so that target client groups needing information pushing can be screened, and based on the target client list, information pushing can be performed in a specific mode to inform the clients to acquire related activity information.

Fig. 3 is a flowchart of a second embodiment of the information pushing method 100 according to the embodiment of the present application, referring to fig. 3, in one embodiment, the step of obtaining the second context information in step 102 includes:

Step 201, respectively obtaining attribute information of customer class nodes of customers to be tested in the two attribute diagrams, attribute information of merchant class nodes of merchants having bank card transactions with the customers to be tested, attribute information of each connecting edge corresponding to the customers to be tested and the merchants, and respectively representing the attribute information as vectorsAnd->Wherein c is a customer identifier, b is a merchant identifier, t is a continuous edge identifier, m represents an mth merchant, k represents a kth continuous edge, and v represents a customer to be tested;

step 202, vector stitching to obtain second context information; the second context information comprises vectors of M merchants and vectors of K continuous edges, M is more than or equal to 2, K is more than or equal to 2, M is more than or equal to 1 and less than or equal to M, and K is more than or equal to 1 and less than or equal to K.

In particular, the present application also provides an alternative embodiment, that, regarding the acquisition of the second context information, may be implemented through step 201 and step 202,wherein the content executed in step 201 includes respectively acquiring attribute information of customer class nodes of customers to be tested in the two attribute diagrams, attribute information of merchant class nodes of merchants having bank card transactions with the customers to be tested, attribute information of each continuous edge corresponding to the customers to be tested and the merchants, and respectively representing the attribute information as vectorsAnd->The merchant associated with the same customer may include multiple merchants and multiple types, so that for one customer to be tested, the corresponding merchant may include M merchants, the connecting edge between the customer to be tested and the merchant with a consumption relationship may include K merchants. And then, the three types of vectors are spliced in step 202 to obtain second context information, wherein the second context information can comprise multiple groups of consumption relation data, and each group of consumption relation data comprises attribute information vectors of a customer to be tested, a merchant and related edges.

The second context information can be used for realizing information between the client to be detected and the merchant with the consumption relation, so that more accurate prediction on the consumption capacity, consumption habit, consumption place and the like of the client to be detected is improved, and more accurate information is pushed to the client to be detected.

Fig. 4 is a flowchart of a third embodiment of the information pushing method 100 provided by the embodiment of the present application, referring to fig. 4, in one embodiment, in step 201, attribute information of a client class node of a client to be tested in the two attribute diagrams, attribute information of a merchant class node of each merchant transacted with a bank card of the client to be tested, attribute information of each connecting edge corresponding to the client to be tested and the merchant are respectively obtained, and the method includes the following steps:

step 301, obtaining basic feature information of clients, merchants and continuous edges to be detected, and respectively representing the basic feature information as initial vectorsAnd->

Step 302, adding corresponding category information to the initial vectors to obtain respective primary augmented vectorsAnd->

Step 303, respectively toAdding the corresponding association relation information of the nodes and the connecting edges to obtain respective secondary augmentation vectors +.>

The corresponding category information is one of a customer node category, a merchant node category and a connecting edge category.

Specifically, the present application also provides an alternative implementation manner, in which the process of respectively obtaining the attribute information of the customer class node of the customer to be tested in the two attribute diagrams, the attribute information of the merchant class node of each merchant transacted with the bank card of the customer to be tested, and the attribute information of each connecting edge corresponding to the customer to be tested and the merchant includes obtaining the basic feature information of the customer to be tested through step 301 and representing the basic feature information as an initial vectorAcquiring basic characteristic information of a merchant with a transaction with a customer to be tested and representing the basic characteristic information as an initial vector +.>Basic characteristic information of a connecting edge between a relevant merchant and a client to be tested is acquired and respectively expressed as an initial vector +.>Further, the initial vectors of the clients to be tested are respectively +.>Adding corresponding category information to obtain primary augmentation vector ++of the client to be tested>Initial vector for merchant->Adding corresponding category information to obtain a primary augmentation vector +.>Initial vector of the opposite edge->Adding corresponding category information to obtain a primary augmentation vector of the continuous edgeFurther, the first augmentation vector of the customer to be tested is respectively +.>Adding the association relation information to obtain a secondary augmentation vector of the client to be tested>One augmentation vector for merchant >Adding the association relation information to obtain a secondary augmentation vector of the merchant>One-time augmentation vector of opposite edges>Adding the association relation information to obtain a secondary augmentation vector ++>And then, the three types of secondary augmentation vectors are spliced in step 202 to obtain second context information, wherein the second context information can comprise multiple groups of consumption relationship data, and each group of consumption relationship data comprises secondary augmentation vectors of clients to be tested, merchants and related edges.

The initial vector obtained in the step 301 is added with category information through the step 302, so that the judgment that attribute vectors belong to clients, merchants and continuous edges can be realized, and the primary augmentation vector obtained in the step 302 is added with association relation information through the step 303, so that the association relation information between nodes and continuous edges can be obtained; after adding the category information and the association relation information to the initial vector obtained in step 301 through step 302 and step 303, the model can be focused on the information of the current input node connecting edge or node by the influence of the attention mechanism in the graph translation model, so that the accuracy of the output result of the graph translation model is improved, the accurate prediction of the graph translation model on the activities interested by the clients to be tested is improved, the follow-up pushing of more accurate information for the clients to be tested is facilitated, the utilization rate of a bank card is improved, and meanwhile, the cost input of manpower and time can be reduced in the whole process of pushing the activity information.

Fig. 5 is a flowchart of a fourth embodiment of the information pushing method 100 according to the embodiment of the present application, referring to fig. 5, in one embodiment, the step 302 is to add corresponding category information to the initial vectors to obtain respective primary augmented vectorsAnd->The method comprises the following steps:

step 3021, obtaining three category identification vectors corresponding to a client node category, a merchant node category and a link category respectively;

step 3022, splicing the category identification vector corresponding to the category of the client node with the initial vector of the client to be tested to obtain a primary augmented vector of the client to be tested

Step 3023, splicing the class identification vector corresponding to the class of the merchant node with the initial vector of the merchant to obtain a primary augmentation vector of the merchant

Step 3024, splicing the class identification vector corresponding to the connected edge class with the initial vector of the connected edge to obtain a primary augmented vector of the connected edge

Specifically, the present application provides an alternative implementation manner, in which three kinds of identification vectors corresponding to a client node category, a merchant node category and a link category are obtained through step 3021, namely three kinds of identification vectors are obtained, specifically, a category identification vector corresponding to a client node category, a category identification vector corresponding to a merchant node category and a category identification vector corresponding to a link category, and then the category identification vector corresponding to a client node category is spliced with an initial vector of a client to be tested through steps 3022, 3023 and 3024, respectively, to obtain a primary augmentation vector of the client to be tested Splicing the class identification vector corresponding to the class of the merchant node with the initial vector of the merchant to obtain a primary augmentation vector of the merchant>Splicing the class identification vector corresponding to the class of the continuous edge with the initial vector of the continuous edge to obtain a primary augmentation vector +.>The identification vectors of the three categories are used for adding the category information to the initial vector respectively, so that the attribute vector can be judged to belong to a client, a merchant and a continuous edge, and the accuracy of the graph translation model on data processing is improved.

Fig. 6 is a flowchart of a fifth embodiment of the information pushing method 100 according to the embodiment of the present application, referring to fig. 6, in one embodiment, the step 303 is performed separatelyAdding the corresponding association relation information of the nodes and the connecting edges to obtain respective secondary augmentation vectors +.>The method comprises the following steps:

step 3031, a first orthogonal node identification matrix and a second orthogonal node identification matrix are obtained, wherein the first orthogonal node identification matrix is used for identifying the association relation information of the client class node and the corresponding connecting edge, and the second orthogonal node is used for identifying the association relation information of the merchant class node and the corresponding connecting edge;

step 3032, combining the first orthogonal node identification matrix with the primary augmented vector of the customer to be tested Splicing to obtain a secondary augmentation vector of the client to be tested>

Step 3033, combining the second orthogonal node identification matrix with the primary augmentation vector of the merchantSplicing to obtain a secondary augmentation vector of the merchant>

Step 3034, adding the first orthogonal node identification matrix, the second orthogonal node identification matrix and the primary augmentation vector of the continuous edgeSplicing to obtain a secondary augmentation vector +.>

Specifically, the present application provides an alternative embodiment, in which, in step 3031, a first orthogonal node identification matrix for identifying association information of a customer node and a corresponding connection edge and a second orthogonal node identification matrix for identifying association information of a merchant node and a corresponding connection edge are obtained, and in turn, in step 3023, a primary augmentation vector of the first orthogonal node identification matrix and a customer to be tested is adoptedSplicing to obtain the secondary augmentation vector of the customer to be testedStep 3033 is performed using the second orthogonal node identification matrix and the merchant's primary augmentation vector +.>Splicing to obtain a secondary augmentation vector of the merchant>First augmentation vector +.A first orthogonal node identification matrix and a second orthogonal node identification matrix are adopted simultaneously with the continuous edge through 3034 >Splicing to obtain a secondary augmentation vector +.>The first orthogonal node identification matrix and the second orthogonal node identification matrix are used for adding the association relation information to the primary augmentation vector of the client, the primary augmentation vector of the merchant and the primary augmentation vector of the connecting edge, so that the association relation information between the nodes and the connecting edge can be obtained, and the processing accuracy of the graph translation model to the data is improved.

Fig. 7 is a flowchart of a sixth embodiment of an information pushing method 100 according to an embodiment of the present application, and fig. 8 is a block diagram of a graph translation model according to an embodiment of the present application, referring to fig. 7 and 8, in which the graph translation model includes an encoder and a decoder;

inputting the first context information of the customer to be tested and the second context information of the two attribute graphs into the pre-trained graph translation model in step 103, comprising the following steps:

step 1031, inputting first context information of a customer to be tested to a decoder of a pre-trained graph translation model, and inputting second context information of two attribute graphs of the customer to be tested to an encoder of the pre-trained graph translation model;

step 1032, the pre-trained graph translation model is specifically configured to process the second context information through the encoder, calculate the processed information based on the attention mechanism, output a first calculation result to a forward propagation layer of the encoder, and perform nonlinear variation calculation and mapping on the first calculation result to obtain a first low-dimensional vector; the first low-dimensional vector and the first context information are input to a decoder, a second calculation result is output to a forward propagation layer of the decoder based on an attention mechanism, nonlinear change calculation is carried out on the second calculation result, the second low-dimensional vector is obtained through mapping, and a probability value that a customer to be tested belongs to a bank card information pushing customer is obtained based on the output of the decoder.

Specifically, the application provides an alternative implementation mode, which comprises an encoder and a decoder in a graph translation model, wherein the encoder comprises a cascade attention mechanism and a forward propagation layer, and the decoder also comprises a cascade attention mechanism and a forward propagation layer, and the forward propagation layer of the encoder is connected with the attention mechanism layer of the decoder. The step 103 of inputting the first context information of the client to be tested and the second context information of the two attribute graphs into the pre-trained graph translation model may be specifically implemented through the steps 1031 and 1032, where the step 1031 is used to input the first context information of the client to be tested into the decoder of the pre-trained graph translation model, input the second context information into the encoder of the pre-trained graph translation model, that is, process the first context information and the second context information through the decoder and the encoder respectively, specifically, as in the step 1032, process the second context information through the encoder, and calculate the processed information based on the attention mechanism of the encoder, including the longitudinal splicing of the secondary augmentation vector correlation matrix, where, because the lengths of the client, the merchant and the edge correlation attribute vectors are different, the length of the longest attribute vector may be selected, the attribute vector less than the length is filled with 0 element, and the input element in the normalized function is normalized, and the input element in the matrix is between each line and 1 to 1; the attention mechanism outputs a first calculation result to a forward propagation layer of the encoder, and performs nonlinear variation calculation on the first calculation result and maps the first calculation result to obtain a first low-dimensional vector; the first low-dimensional vector and the first context information are input into the decoder together, a second calculation result is output to a forward propagation layer of the decoder through an attention mechanism of the decoder, nonlinear change calculation is carried out on the second calculation result, the second low-dimensional vector is obtained through mapping, and then the probability value that the customer to be measured belongs to the information pushing customer of the bank card is calculated based on the output of the decoder.

According to the application, the first context information and the second context information are input into the pre-trained graph translation model, the probability value of the client to be tested, which belongs to the bank card information pushing client, is calculated through the graph translation model, and the advanced prediction of the subsequent consumption of the client to be tested based on the information of the client to be tested, the information of the merchant associated with the client to be tested, the specific transaction condition and the like is realized, so that the bank system corresponding to the bank card can push the relevant activity information more interested by the client to be tested for the client to be tested, and the probability that the bank system corresponding to the bank card pushes the relevant activity information more likely to be participated by the client to be tested is improved, thereby the use ratio of the bank card is correspondingly improved through the pushed activity information.

Besides basic characteristic information of the clients, learning of the client relationship data is added in the model, and more accurate and reasonable evaluation results can be obtained. In addition, by means of the advantage of the model structure, the method can receive the original data collected in multiple aspects and multiple types, fully utilize the customer base and behavior information in the original data, does not depend on the characteristic engineering of manual work as a main part, reduces the manual evaluation investment of marketers based on expert knowledge, and improves the working efficiency of the whole business process.

Fig. 9 is a flowchart of a training process of a graph translation model provided in an embodiment of the present application, please refer to fig. 9, and in an embodiment, the graph translation model is obtained through training in the following step 200:

step 401, constructing a sample two-part attribute map of a sample client and a sample merchant, and acquiring third context information of the sample two-part attribute map;

step 402, inputting third context information into a graph translation model to be trained, and outputting a prediction result of a sample client belonging to a bank card information pushing client by the graph translation model to be trained;

step 403, determining a current loss function value of the model; the loss function value is determined based on a preconfigured two-class cross entropy loss function, a prediction result currently output by the model, a first label of the sample client participating in information pushing or a second label of the sample client not participating in information pushing;

and step 404, adjusting model parameters based on the current loss function value until training is finished, and obtaining a pre-trained graph translation model.

Specifically, the application also provides a training process of the graph translation model, which specifically comprises a step 401-a step 404, wherein the step 401 is used for constructing a sample two-part attribute graph of a sample client and a sample merchant, obtaining third context information of the sample two-part attribute graph, and the third context information can comprise a secondary augmentation vector of the sample client, a secondary augmentation vector of the sample merchant and a secondary augmentation vector of a connecting edge of the sample client and the sample merchant with a transaction relationship; the third context information is further input into a graph translation model to be trained through step 402, and the graph translation model to be trained outputs a prediction result of the sample client belonging to the bank card information pushing client through calculation of the third context information; then determining the current loss function value of the model through step 403; step 404 is used to adjust model parameters based on the current loss function value until training is completed, and a pre-trained graph translation model is obtained.

In addition, the application also provides an optional training step of the graph translation model, which comprises the following steps:

fig. 10 is a schematic diagram of context information of a two-part attribute map provided in an embodiment of the present application, please refer to fig. 10, in which two-part attribute maps corresponding to sample clients are obtained, first, from an initial node, depth-first traversal is performed; in particular, if the sample clients, sample merchants and transaction attributes are respectively used as matrix Representing that each row of the matrix corresponds to a sample client, a sample merchant and an attribute vector of a transaction, then after the depth-first traversal algorithm is finished, an original attribute vector sequence +_> Representing sample clients, transactions and sample merchants, respectivelyThe nth, k and m row vectors of the family attribute matrix; wherein c is a customer identifier, b is a merchant identifier, t is a continuous edge identifier, and the original attribute vector sequence comprises N sample customers, M merchant vectors and K continuous edge vectors, wherein N is more than or equal to 2, M is more than or equal to 2, K is more than or equal to 2, N is more than or equal to 1 and less than or equal to N, M is more than or equal to 1 and less than or equal to M, and K is more than or equal to 1 and less than or equal to K.

Because the original attribute vector sequence can not provide the distinguishing information of whether a certain attribute vector belongs to a sample client, a sample merchant or a transaction, a trainable category identification vector can be obtained For marking the sequence of attribute vectors. Specifically, the category marking rule is: attribute vector for a sample client n>Marking the category as ++>Attribute vector for sample merchant m>Marking the category asAttribute vector for a transaction t>Marking the category as ++>Wherein, [ ·, ]]And the transverse splicing of the two matrixes is shown, and e is more than or equal to 1.

On the other hand, since the above-mentioned attribute vector sequence S cannot provide the association information between the nodes and the edges, two orthogonal node identification matrices are also requiredAnd->The method is used for further marking the augmentation vector after the category marking is completed. Specifically, the node marking rule is: n-attribute augmentation vector for a sample client>Marking the category as ++>Attribute vector for sample merchant m>Marking the category as ++>Attribute vector for a transaction t>If the two end nodes are respectively a sample client node and a sample merchant node, marking the class mark as +.>Wherein p is greater than or equal to 1 and q is greater than or equal to 1.

After the orthogonal matrix is used for marking, the model can be more focused on the information of the edge or the node connected with the current input node under the influence of the attention mechanism in the translation model. After the original attribute vectors are sequentially subjected to category and node marking, a context information sequence of a two-part attribute map can be finally obtained

Further, training a graph translation model is executed, and in a model training stage, firstly, context information of two graph attribute graphs is input into an encoder in sequence in full quantity; specifically, the context information may first be entered into the attention mechanism for the following calculations:

wherein, the liquid crystal display device comprises a liquid crystal display device,[·；·]and representing the longitudinal splicing of the two matrixes, namely respectively splicing the sample clients, the sample merchants and the amplified matrixes of the transaction attribute matrixes after twice marking (category and node). Since the sample customer, sample merchant, and transaction attributes differ in length, the longest attribute vector length is used here to fill in attribute vectors that are less than this length with 0 elements. softmax (·) is a normalization function for normalizing the elements in each row of vectors in the matrix input thereto to an interval between 0 and 1, and making the sum of all elements of the row vector 1; wherein T is not null.

The output of this attention mechanism is then passed through a forward propagation layer, where it is calculated and mapped into a new low-dimensional vector space as follows:

FFN ^encode (Y″)＝max(0,Attention ^encode (Y″,Y″,Y″)W ₁ +b ₁ )W ₂ +b ₂

wherein W is ₁ 、b ₁ And W is ₂ 、b ₂ Are trainable model parameters. This calculated low-dimensional vector FFN ^encode (Y') is the final output of the encoder.

At the output FFN of this encoder ^encode After (Y "), it can be input with the sample data into the attention mechanism of the decoder for the following calculation:

then, similar to the encoder's calculation, the output Attention of this Attention mechanism is presented ^decode Is input into a forward propagation layer to obtain FFN ^decode (Y ₍ ″ _i,·) ) This is taken as the output of the decoder. And adding the following sigmoid function at the output layer of the decoder as the final output of the graph translation model:

where w and b are both trainable model parameters. Finally, in order to train the graph translation model, different sample clients can be respectively marked with different rights and interests labels according to whether the sample clients get and use the rights and interests of the credit card, and based on the labels, the following two classification cross entropy loss functions are constructed, and the definition is as follows:

wherein s' _i E {1,0} represents a set of tags that are whether or not the credit card equity was picked up and used, where 1 represents picked up and used and 0 represents picked up and unused. After the construction of the loss function is finished, defining an initialization learning rate, and then adopting a gradient descent method of Adam (random optimization algorithm) to minimize the loss function, thus finishing the training of model parameters.

For a certain client v, its context information X', is added _(v,·) And two-part attribute map context informationRespectively input into a decoder and an encoder to obtain an output s of the decoder _v If the output value is greater than 0.5, the client is considered to be required to receive the rights and interests, otherwise, the client is not required to receive the rights and interests. Repeating the steps for all clients to screen out target client groups needing to receive credit card rights; based on the group list, message pushing can be performed through a specific channel, clients are informed of acquiring rights and interests, and rights and interests marketing activities are completed.

Fig. 11 is a schematic diagram of an information pushing device according to an embodiment of the present application, please refer to fig. 11 in conjunction with fig. 2-9, and based on the same inventive concept, the present application further provides an information pushing device 300, which includes:

an information obtaining module 91, configured to obtain first context information of a client to be tested; the first context information comprises at least one of basic characteristic information and bank card transaction information of a customer to be tested and a customer associated with the customer to be tested; acquiring second context information of the two attribute graphs based on the two attribute graphs of the client to be tested and the merchant, which are constructed in advance; the graph elements of the two attribute graphs comprise nodes and connecting edges, wherein the nodes comprise two types of clients and merchants, and the connecting edges exist between the nodes of different types and are used for representing bank card transaction information between the corresponding clients and merchants; the second context information comprises attribute information of each picture element in the picture; the attribute information comprises corresponding graph element category information, corresponding association relation between nodes and connecting edges and basic characteristic information of each graph element;

The result calculation module 92 is configured to input first context information of a customer to be tested and second context information of the two attribute maps into a pre-trained map translation model, where the map translation model determines a probability value that the customer to be tested belongs to a bank card information push customer through the first context information and the second context information;

a recommendation module 93, configured to add a client to be tested to the target client list when the probability value output by the graph translation model is greater than a preset value; and pushing the bank card information to the clients in the target client list.

Specifically, the present application further provides an information pushing device 300, where the information pushing device 300 includes an information obtaining module 91, a result calculating module 92, and a recommending module 93, where the information obtaining module 91 is configured to obtain first context information of a to-be-tested client, where the first context information may include at least basic feature information of the to-be-tested client, bank card transaction information thereof, and may further include basic feature information of a client related to the to-be-tested client, and bank card transaction information thereof, where the client related to the to-be-tested client is, for example, friends, relatives, colleagues, etc. of the to-be-tested client. It should be noted that, the first context information may only obtain basic feature information of the customer to be tested and the customer associated with the customer according to the requirement, where the basic feature information may include, but is not limited to, occupation, income, consumption habit, age; the first context information can also only acquire the bank card transaction information of the customer to be tested and the customer related to the customer to be tested according to the requirement, wherein the bank card transaction information can include, but is not limited to, transaction amount, transaction time and type of purchased goods; the first context information can also obtain the basic feature information and bank card transaction information of the customer to be tested, the customer related to the customer to be tested and the customer related to the customer at the same time according to the requirement, and the application is not limited in particular. By acquiring the information of the client to be tested and the related clients, the consumption capability, consumption habit, consumption place and the like of the client to be tested can be predicted more accurately through the information of the client to be tested and the information of the client related to the client to be tested, so that the client to be tested can push more accurate information for the client to be tested later.

The information obtaining module 91 is further configured to obtain second context information, where the second context information is obtained based on a two-part attribute map, and the two-part attribute map is specifically a two-part attribute map of a customer and a merchant to be tested; it should be noted that, the graph elements of the two attribute graphs include nodes and connecting edges, the nodes include two types of clients and merchants, the connecting edges exist between the nodes of different types and are used for representing transaction information between the clients and merchants which are correspondingly connected; the second context information comprises attribute information of each picture element in the picture, wherein the attribute information comprises corresponding picture element category information, corresponding association relation between nodes and continuous edges and basic characteristic information of each picture element; the category information of the graph elements can be used for distinguishing clients, merchants and continuous edges; the basic information features of each drawing element include, for example, occupation, income, consumption habit, age, etc. of the customer, merchant category, business condition, etc. of the merchant, transaction amount, transaction time, type of commodity purchased, etc. on the side. The acquired second context information can be used for realizing information between the client to be detected and the merchant with the consumption relation, so that more accurate prediction on the consumption capacity, consumption habit, consumption place and the like of the client to be detected is improved, and more accurate information is pushed to the client to be detected in the follow-up process.

The result calculation module 92 is configured to input the first context information and the second context information into a pre-trained graph translation model, calculate, through the graph translation model, a probability value that a to-be-tested client belongs to a bank card information pushing client, and implement early prediction of subsequent consumption of the to-be-tested client based on information of the to-be-tested client, information of a merchant associated with the to-be-tested client, specific transaction conditions, and other information, so that a bank system corresponding to a bank card can push relevant activity information more interested in the to-be-tested client for the to-be-tested client, and the relevant activity information pushed by the bank system corresponding to the bank card for the to-be-tested client is more likely to be participated by the client, thereby correspondingly improving the use rate of the bank card through the pushed activity information.

The recommendation module 93 is configured to compare the probability value obtained by the result calculation module 92 with a preset value, and add the to-be-tested client to the target client list when the probability value output by the graph translation model is greater than the preset value, where the likelihood that the client in the target client list retrieves and uses relevant activity information is greater than that of other clients, so that the bank card information is pushed to the client in the target client list, so that the activity pushing accuracy can be improved, and the cost input of manpower and time required by the activity pushing is reduced.

In one embodiment, the acquiring of the second context information performed in the information acquiring module 91 includes:

respectively acquiring attribute information of customer class nodes of customers to be tested in the two attribute diagrams, attribute information of merchant class nodes of merchants with bank card transactions of the customers to be tested, and attribute information of each continuous edge corresponding to the customers to be tested and the merchants, and respectively representing the attribute information as vectorsAnd->Wherein c is a customer identifier, b is a merchant identifier, t is a continuous edge identifier, m is an mth merchant, k is a kth continuous edge, and v is a customer to be tested；

Vector stitching to obtain second context information; the second context information comprises vectors of M merchants and vectors of K continuous edges, M is more than or equal to 2, K is more than or equal to 2, M is more than or equal to 1 and less than or equal to M, and K is more than or equal to 1 and less than or equal to K. Reference is made in particular to fig. 3, and to the description above in relation to fig. 3.

In one embodiment, the information obtaining module 91 obtains attribute information of a customer class node of a customer to be tested in the two attribute graphs, attribute information of a merchant class node of each merchant having a bank card transaction with the customer to be tested, and attribute information of each connecting edge corresponding to the customer to be tested and the merchant respectively, and includes the following steps:

Basic characteristic information of a customer to be tested, a merchant and a connecting edge is obtained and respectively expressed as initial vectors And/>

adding corresponding category information to the initial vectors to obtain respective primary augmented vectorsAnd

The corresponding category information is one of a customer node category, a merchant node category and a connecting edge category. Reference is made in particular to fig. 4, and to the description above in relation to fig. 4.

In one embodiment, the information obtaining module 91 adds corresponding category information to the initial vectors to obtain respective primary augmented vectorsAnd->The method comprises the following steps:

three category identification vectors are obtained and respectively correspond to a client node category, a merchant node category and a connecting edge category;

splicing the category identification vector corresponding to the category of the client node with the initial vector of the client to be detected to obtain a primary augmentation vector of the client to be detected

Splicing the class identification vector corresponding to the class of the merchant node with the initial vector of the merchant to obtain a primary augmentation vector of the merchant

Splicing the class identification vector corresponding to the class of the continuous edge with the initial vector of the continuous edge to obtain a primary augmented vector of the continuous edge Reference is made in particular to fig. 5 and to the description above in relation to fig. 5.

In one embodiment, the respective pairs performed in the information acquisition module 91Adding the corresponding association relation information of the nodes and the connecting edges to obtain respective secondary augmentation vectors +.>The method comprises the following steps:

acquiring a first orthogonal node identification matrix and a second orthogonal node identification matrix, wherein the first orthogonal node identification matrix is used for identifying association relation information of client class nodes and corresponding connecting edges, and the second orthogonal node is used for identifying association relation information of merchant class nodes and corresponding connecting edges;

first orthogonal node identification matrix and one-time augmentation vector of customer to be testedSplicing to obtain a secondary augmentation vector of the client to be tested>

One-time augmentation vector of second orthogonal node identification matrix and commercial tenantSplicing to obtain a secondary augmentation vector of the merchant>

First augmentation vectors for connecting the first orthogonal node identification matrix and the second orthogonal node identification matrixSplicing to obtain a secondary augmentation vector +.>Reference is made in particular to fig. 6, and to the description above in relation to fig. 6.

the information obtaining module 91 inputs the first context information of the client to be tested and the second context information of the two attribute maps to the pre-trained map translation model, and includes the following steps:

Inputting first context information of a client to be tested to a decoder of a pre-trained graph translation model, and inputting second context information of two attribute graphs of the client to be tested to an encoder of the pre-trained graph translation model;

the pre-trained graph translation model is specifically used for processing second context information through an encoder, calculating the processed information based on an attention mechanism, outputting a first calculation result to a forward propagation layer of the encoder, and carrying out nonlinear change calculation and mapping on the first calculation result to obtain a first low-dimensional vector; the first low-dimensional vector and the first context information are input to a decoder, a second calculation result is output to a forward propagation layer of the decoder based on an attention mechanism, nonlinear change calculation is carried out on the second calculation result, the second low-dimensional vector is obtained through mapping, and a probability value that a customer to be tested belongs to a bank card information pushing customer is obtained based on the output of the decoder. Reference is made in particular to fig. 7 and to the description above in relation to fig. 7.

In one embodiment, the graph translation model associated in the information acquisition module 91 is trained by:

Inputting third context information into a graph translation model to be trained, and outputting a prediction result of a sample client belonging to a bank card information pushing client by the graph translation model to be trained;

determining a current loss function value of the model; the loss function value is determined based on a preconfigured two-class cross entropy loss function, a prediction result currently output by the model, a first label of the sample client participating in information pushing or a second label of the sample client not participating in information pushing;

and adjusting model parameters based on the current loss function value until training is finished, and obtaining a pre-trained graph translation model. Reference is made in particular to fig. 9, and to the description above in relation to fig. 9.

Fig. 12 is a schematic diagram of a computer device according to an embodiment of the present application, please refer to fig. 12 in conjunction with fig. 2-9, and based on the same inventive concept, the present application further provides a computer device 400, including a memory and a processor, where the memory stores a computer program, and the processor implements the foregoing information pushing method when executing the computer program, and the information pushing method is any one of the information pushing methods mentioned in the embodiments of the present application.

Fig. 12 is an internal structure diagram of a computer device 400 according to an embodiment of the present application, and in an embodiment, a computer device 400 is provided, where the computer device 400 may be a server or a terminal device, and the internal structure diagram may be as shown in fig. 12. The computer device 400 includes a processor, a memory, an Input/Output interface (I/O) and a transmitter, a receiver. The processor, the memory and the input/output interface are connected through a system bus, and the communication interface is connected to the system bus through the input/output interface. Wherein the processor of the computer device 400 is configured to provide computing and control capabilities. The memory of the computer device 400 includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device 400 is used to store relevant information such as customers, merchants, transaction relationships between customers and merchants, and the like. The input/output interface of the computer device 400 is used to exchange information between the processor and external devices. The communication interface of the computer device 400 is used for communication with an external terminal through a network connection. The computer program, when executed by a processor, implements any of a number of information push methods.

It will be appreciated by those skilled in the art that the structure shown in FIG. 12 is merely a block diagram of some of the structures associated with the present inventive arrangements and does not constitute a limitation of the computer device 400 to which the present inventive arrangements may be applied, and that a particular computer device 400 may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.

Based on the same inventive concept, the present application further provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements the aforementioned information pushing method, where the information pushing method is any one of the information pushing methods mentioned in the embodiments of the present application, and related embodiments may be referred to above.

Based on the same inventive concept, the present application also provides a computer program product, which comprises a computer program, wherein the computer program, when executed by a processor, implements the foregoing information pushing method, and the information pushing method is any one of the information pushing methods mentioned in the embodiments of the present application, and related embodiments can be seen from the foregoing.

It should be noted that, the data (including, but not limited to, data for analysis, stored data, displayed data, etc.) related to the present application are all information and data authorized by the user or fully authorized by each party, and the collection, use and processing of the related data need to comply with the related laws and regulations and standards of the related country and region.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, database, or other medium used in embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high density embedded nonvolatile Memory, resistive random access Memory (ReRAM), magnetic random access Memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric Memory (Ferroelectric Random Access Memory, FRAM), phase change Memory (Phase Change Memory, PCM), graphene Memory, and the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, and the like. By way of illustration, and not limitation, RAM can be in the form of a variety of forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), and the like. The databases referred to in the embodiments provided herein may include at least one of a relational database and a non-relational database. The non-relational database may include, but is not limited to, a blockchain-based distributed database, and the like. The processor referred to in the embodiments provided in the present application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, or the like, but is not limited thereto.

The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.

The foregoing examples illustrate only a few embodiments of the application and are described in detail herein without thereby limiting the scope of the application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of the application should be assessed as that of the appended claims.

Claims

1. An information pushing method, characterized in that the method comprises:

2. The information pushing method according to claim 1, wherein the obtaining of the second context information includes:

3. The information pushing method as claimed in claim 2, wherein,

respectively obtaining attribute information of a customer class node of the customer to be tested, attribute information of a merchant class node of each merchant having a bank card transaction with the customer to be tested, and attribute information of each connecting edge corresponding to the customer to be tested and the merchant in the two attribute diagrams, wherein the method comprises the following steps:

Adding corresponding category information to the initial vectors to obtain respective primary augmented vectors And

4. The information pushing method as claimed in claim 3, wherein,

the corresponding category information is added to the initial vectors respectively to obtain respective primary augmentation vectors And->Comprising the following steps:

5. The information pushing method as claimed in claim 3, wherein,

the respective pairs ofAdding the corresponding association relation information of the nodes and the connecting edges to obtain respective secondary augmentation directionsQuantity->Comprising the following steps:

Combining the first orthogonal node identification matrix with the primary augmentation vector of the customer under testSplicing to obtain the secondary augmentation vector ++ ++of the client to be tested>

6. The information pushing method according to claim 1, wherein the graph translation model includes an encoder and a decoder;

7. The information pushing method according to claim 1, wherein the graph translation model is trained by:

8. An information pushing apparatus, characterized in that the apparatus comprises:

9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 7 when the computer program is executed.

10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.

11. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.