CN112581281A

CN112581281A - Product recommendation method and device, storage medium and electronic equipment

Info

Publication number: CN112581281A
Application number: CN202011576242.7A
Authority: CN
Inventors: 周继顺; 张航; 刘宏吉; 董莹; 薛平
Original assignee: China Construction Bank Corp
Current assignee: China Construction Bank Corp
Priority date: 2020-12-28
Filing date: 2020-12-28
Publication date: 2021-03-30

Abstract

The invention provides a product recommendation method and device, a storage medium and electronic equipment, which are characterized in that nodes corresponding to target clients and nodes corresponding to target products which are not purchased by the target clients are determined through a knowledge graph network trained based on a webpage ranking algorithm, and ranking indexes of the nodes corresponding to the target products are determined, wherein the ranking indexes represent the probability of the nodes corresponding to the target clients migrating to the nodes corresponding to the target products. And determining the recommendation index of each target product based on the ranking index of the node corresponding to each target product, and selecting a plurality of recommended products from all the target products according to the recommendation index of each target product so as to recommend the products to target customers. By applying the method provided by the invention, the recommended products are determined based on the random walk of the nodes in the network, so that the diversity of the products recommended to the customers is improved, the accuracy of personalized recommendation is improved, the product recommendation effect is improved, and the service experience is improved.

Description

Product recommendation method and device, storage medium and electronic equipment

Technical Field

The present invention relates to the field of computer technologies, and in particular, to a method and an apparatus for recommending a product, a storage medium, and an electronic device.

Background

In the normal operation process of a financial institution, product recommendation is made to a customer to guide the customer to purchase a product is a common way to widen a business channel. Wherein, the product recommended to the client is one of the key elements influencing the recommendation effect.

At present, products recommended to customers are generally determined individually based on recommendation algorithms such as binary classification or collaborative filtering. The method mainly comprises the steps of taking products signed by a client as reference objects, finding out products with the height similar to that of the reference objects through the similarity among the products, and recommending the products to the client, or recommending the signed products of similar clients to the client by using the behavior habits of the client.

In an actual recommendation scenario, the amount of information grasped about a recommended object of a product is sometimes very rare, and a recommended object facing a product such as a settlement product is often a new customer. Under the condition, products recommended to the customers are determined based on the existing mode, accurate recommendation judgment is difficult to be made by utilizing customer information, namely, the accuracy of personalized recommendation is low, the product recommendation effect is influenced, and poor service experience is caused for the customers.

Disclosure of Invention

In view of this, the embodiment of the present invention provides a product recommendation method to solve the problems that the accuracy of personalized recommendation is low, the product recommendation effect is affected, and the service experience of a customer is poor.

The embodiment of the invention also provides a product recommendation device which is used for ensuring the actual realization and application of the method.

In order to achieve the above purpose, the embodiments of the present invention provide the following technical solutions:

a method of product recommendation, comprising:

when a product recommendation request sent by a user is received, determining a target client corresponding to the product recommendation request;

determining each product not purchased by the target customer as a target product among all products;

determining nodes corresponding to the target customers and nodes corresponding to each target product in a knowledge graph network trained based on a webpage ranking algorithm, wherein the knowledge graph network is constructed according to all customers, all products and the incidence relation between each customer and each product;

determining a ranking index of a node corresponding to each target product, wherein the ranking index of the node corresponding to each target product represents the probability of the node corresponding to the target client migrating to the node corresponding to the target product;

determining a recommendation index of each target product based on the ranking index of the node corresponding to the target product;

and selecting a plurality of recommended products from all the target products according to the recommendation indexes of the target products, and recommending the target customers with the recommended products.

The method described above, optionally, the process of constructing the knowledge-graph network includes:

determining all entities required for constructing the knowledge-graph network, wherein all entities comprise each customer, each product, each preset customer transaction characteristic and each preset transaction behavior characteristic;

acquiring customer information of each customer, transaction information of each customer and product information of each product;

determining the corresponding relation between the entities according to the customer information of each customer, the transaction information of each customer and the product information of each product;

constructing nodes corresponding to each entity, and constructing relationship edges between the nodes corresponding to each entity according to the corresponding relationship between each entity;

setting the weight corresponding to each relation edge according to a preset weight assignment rule;

and generating the knowledge graph network based on the nodes corresponding to the entities, the relationship edges among the nodes corresponding to the entities and the weight corresponding to each relationship edge.

Optionally, the determining, according to the customer information of each customer, the transaction information of each customer, and the product information of each product, a corresponding relationship between each entity includes:

determining the upstream and downstream relation among the clients according to the client information of the clients;

according to the transaction information of each customer, determining the customer transaction characteristics corresponding to each customer in each preset customer transaction characteristics, and establishing an association relationship between each customer and the corresponding customer transaction characteristics;

determining each transaction behavior characteristic corresponding to each customer in each preset transaction behavior characteristic, and establishing an association relationship between each customer and each corresponding transaction behavior characteristic;

determining products corresponding to each transaction behavior characteristic corresponding to each customer in each product, and establishing the use relationship between each transaction behavior characteristic corresponding to each customer and the corresponding product;

determining each product purchased by each customer in each product, and establishing a purchasing relationship between each customer and each product purchased by the customer;

and determining the association relation among the products according to the product information of the products.

The method described above, optionally, the training process of the knowledge-graph network includes:

according to a preset webpage ranking algorithm, iteratively calculating a webpage ranking value corresponding to each node in the constructed knowledge graph network until the webpage ranking value of each node in the knowledge graph network reaches a stable state;

when the webpage ranking value of each node in the knowledge graph network reaches a stable state, determining an individualized recommendation index corresponding to the knowledge graph network;

and judging whether the personalized recommendation index meets a preset effective index condition, and finishing the training of the knowledge graph network if the personalized recommendation index meets the preset effective index condition.

Optionally, the determining the ranking index of the node corresponding to each target product includes:

for each node corresponding to the target product, determining each path of the node corresponding to the target customer to walk to the node corresponding to the target product;

calculating a webpage ranking value corresponding to each path according to the webpage ranking value of each node where each path is routed;

and performing summation operation on the webpage ranking values corresponding to all the paths, and determining an operation result as a ranking index of the node corresponding to the target product.

Optionally, in the method, the selecting, according to the recommendation index of each target product, each recommended product from each target product includes:

and comparing the recommendation index of each target product with a preset threshold, and selecting the target product with the recommendation index larger than the preset threshold as a recommended product.

In the foregoing method, optionally, the web page ranking algorithm is a personalized web page ranking algorithm.

A product recommendation device comprising:

the system comprises a first determining unit, a second determining unit and a third determining unit, wherein the first determining unit is used for determining a target client corresponding to a product recommendation request when the product recommendation request sent by a user is received;

a second determining unit configured to determine, as a target product, each product that is not purchased by the target customer, among all products;

a third determining unit, configured to determine, in a knowledge-graph network trained based on a web-based ranking algorithm, a node corresponding to the target customer and a node corresponding to each target product, where the knowledge-graph network is constructed according to all customers, all products, and an association relationship between each customer and each product;

a fourth determining unit, configured to determine a ranking index of a node corresponding to each target product, where the ranking index of the node corresponding to each target product represents a probability that a node corresponding to the target client walks to the node corresponding to the target product;

a fifth determining unit, configured to determine a recommendation index of each target product based on the ranking index of the node corresponding to the target product;

and the selecting unit is used for selecting a plurality of recommended products from all the target products according to the recommendation indexes of the target products, and recommending the target products to the target customers according to the recommended products.

The above apparatus, optionally, further comprises:

a sixth determining unit, configured to determine all entities required to construct the knowledgegraph network, where the all entities include each customer, each product, each preset customer transaction characteristic, and each preset transaction behavior characteristic;

an acquisition unit configured to acquire customer information of each of the customers, transaction information of each of the customers, and product information of each of the products;

a seventh determining unit, configured to determine a correspondence between the entities according to customer information of each customer, transaction information of each customer, and product information of each product;

the construction unit is used for constructing nodes corresponding to each entity and constructing relationship edges between the nodes corresponding to each entity according to the corresponding relationship between each entity;

the setting unit is used for setting the weight corresponding to each relation edge according to a preset weight assignment rule;

and the generating unit is used for generating the knowledge graph network based on the nodes corresponding to the entities, the relationship edges among the nodes corresponding to the entities and the weight corresponding to each relationship edge.

The above apparatus, optionally, the seventh determining unit includes:

the first determining subunit is used for determining the upstream-downstream relationship among the clients according to the client information of the clients;

the second determining subunit is configured to determine, according to the transaction information of each customer, a customer transaction characteristic corresponding to each customer in each preset customer transaction characteristic, and establish an association relationship between each customer and the customer transaction characteristic corresponding to the customer;

the third determining subunit is configured to determine, in each preset transaction behavior feature, each transaction behavior feature corresponding to each customer, and establish an association relationship between each customer and each corresponding transaction behavior feature;

the fourth determining subunit is configured to determine, in each product, a product corresponding to each transaction behavior feature corresponding to each customer, and establish a usage relationship between each transaction behavior feature corresponding to each customer and the product corresponding to the customer;

a fifth determining subunit, configured to determine, among the respective products, each product that has been purchased by each customer, and establish a purchasing relationship between each customer and each product that has been purchased by the customer;

and the sixth determining subunit is used for determining the association relationship among the products according to the product information of the products.

The above apparatus, optionally, further comprises:

the calculation unit is used for iteratively calculating the webpage ranking value corresponding to each node in the constructed knowledge graph network according to a preset webpage ranking algorithm until the webpage ranking value of each node in the knowledge graph network reaches a stable state;

the eighth determining unit is used for determining the personalized recommendation index corresponding to the knowledge graph network after the webpage ranking value of each node in the knowledge graph network reaches a stable state;

and the judging unit is used for judging whether the personalized recommendation index meets a preset effective index condition or not, and if the personalized recommendation index meets the preset effective index condition, finishing the training of the knowledge graph network.

The above apparatus, optionally, the fourth determining unit includes:

a seventh determining subunit, configured to determine, for each node corresponding to the target product, each path through which the node corresponding to the target customer travels to the node corresponding to the target product;

the calculation subunit is configured to calculate, according to the webpage ranking value of each node where each path is routed, a webpage ranking value corresponding to each path;

and the eighth determining subunit is configured to perform summation operation on the webpage ranking values corresponding to all the paths, and determine an operation result as the ranking index of the node corresponding to the target product.

The above apparatus, optionally, the selecting unit includes:

and the comparison subunit is used for comparing the recommendation index of each target product with a preset threshold value, and selecting the target product with the recommendation index larger than the preset threshold value as the recommended product.

Optionally, in the apparatus described above, the web page ranking algorithm is a personalized web page ranking algorithm.

A storage medium comprising stored instructions, wherein the instructions, when executed, control a device on which the storage medium is located to perform the above-mentioned product recommendation method.

An electronic device comprising a memory, and one or more instructions, wherein the one or more instructions are stored in the memory and configured to be executed by the one or more processors to perform the product recommendation method.

Based on the above product recommendation method provided by the embodiment of the present invention, the method includes: when a product recommendation request sent by a user is received, a target client corresponding to the request is determined. Among all the products, each product not purchased by the target customer is determined as the target product. And determining nodes corresponding to the target clients and nodes corresponding to each target product in the knowledge-graph network trained based on the webpage ranking algorithm. The knowledge graph network is constructed according to all customers, all products and the incidence relation between each customer and each product. And determining the ranking index of the node corresponding to each target product, wherein the ranking index of the node corresponding to each target product represents the probability of the node corresponding to the target client migrating to the node corresponding to the target product. And determining the recommendation index of each target product based on the ranking index of the node corresponding to the target product. And selecting a plurality of recommended products from all the target products according to the recommendation index of each target product, and recommending the target customers with each recommended product. By applying the method provided by the embodiment of the invention, the probability that the node corresponding to the target customer walks to the node corresponding to each target product can be determined through the knowledge graph network covering all customers, all products and the incidence relation between the customers and the products, and then the product recommended to the target customer is determined. The knowledge graph network is a global information network, each node can be spread to a plurality of other nodes, the nodes corresponding to customers are easy to spread to other nodes in the migration process, and the possibility that the nodes corresponding to products can be migrated is high. Under the condition of less customer information, the diversity of products recommended to customers can be improved based on the global information, so that the accuracy of personalized recommendation is improved, the product recommendation effect is improved, and the service experience is improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.

FIG. 1 is a flowchart of a method for recommending products according to an embodiment of the present invention;

FIG. 2 is a flowchart of another method of a product recommendation method according to an embodiment of the present invention;

FIG. 3 is a flowchart of another method of a product recommendation method according to an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a product recommendation device according to an embodiment of the present invention;

fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In this application, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

As known in the background art, the product recommendation is usually performed by the financial institution based on a model established by a recommendation algorithm such as binary classification or collaborative filtering. Such recommendation algorithms typically recommend based on the product that the customer has selected. When a new client enters the network, the model does not know the selection of the new client, generally, only products which generally reflect better can be recommended to the client, in addition, transaction data of many common clients are quite sparse, and the model is difficult to perform accurate personalized recommendation.

Therefore, the embodiment of the invention provides a product recommendation method, which can determine products recommended to customers through a knowledge graph network containing global information, can determine the products recommended to the customers by utilizing the incidence relation in the global information, can improve the diversity of recommended products, further improve the accuracy of personalized recommendation, and is beneficial to improving the product recommendation effect and improving the service experience.

An embodiment of the present invention provides a product recommendation method, where the method is applicable to a product recommendation system, an execution subject of the method may be a server or a processor of the product recommendation system, and a flowchart of the method is shown in fig. 1, and includes:

s101: when a product recommendation request sent by a user is received, determining a target client corresponding to the product recommendation request;

in the method provided by the embodiment of the invention, the user can be a staff of a financial institution, when a product recommended to a customer needs to be determined, the user can input or select the customer ID of a target customer of the product to be recommended through the front end of the product recommendation system, and the front end can generate a product recommendation request based on the customer ID in response to the operation of the user and send the product recommendation request to the product recommendation system.

In an actual service scenario, a user usually needs to determine recommended products of clients in batches, a client group can be directly selected, each client in the client group is used as a target client to generate a corresponding product recommendation request, and the product recommendation requests corresponding to each target client are processed according to the method provided by the embodiment of the invention.

After receiving a product recommendation request sent by a user, a product recommendation system can analyze the product recommendation request to obtain a target client needing to determine a recommended product, and usually can directly represent the client by a client ID.

S102: determining each product not purchased by the target customer as a target product among all products;

in the method provided by the embodiment of the invention, the product information of all products and the signed product information of all customers are stored in the system in advance, the signed products of the customers are the products purchased by the customers, and optionally, the products can be represented by the product ID corresponding to each product.

After the target customer of the current product to be recommended is determined, the signed product information of the target customer can be obtained, and each product purchased by the target customer is determined according to the signed product information. Optionally, in all the products, each product may be compared with each product already purchased by the target customer, and if the product is different from each product already purchased by the target customer, the product is determined to be a product not purchased by the target customer, and is determined to be the target product. Alternatively, the respective products that the target customer has purchased may be determined among all the products, and thus the products that the target customer has not purchased will be determined.

It should be noted that, in a specific service scenario, the contracted product information of the client may include one or more contracted products, or may not include the contracted product, that is, the client does not purchase any product currently. When the customer does not purchase any product currently, each of the above-mentioned all products is a target product.

On the other hand, it should be noted that all products mentioned in the method provided by the embodiment of the present invention may be a certain class of products, such as settlement products. Optionally, the user may also pre-define some specific products according to the business requirements, for example, according to the current operation requirements, the promotion and recommendation strength of some products needs to be increased, and recommendation can be performed on these products. Alternatively, all products currently online may be used.

S103: determining nodes corresponding to the target customers and nodes corresponding to each target product in a knowledge graph network trained based on a webpage ranking algorithm, wherein the knowledge graph network is constructed according to all customers, all products and the incidence relation between each customer and each product;

in the method provided by the embodiment of the invention, the knowledge graph network can be constructed in advance according to the clients, the products and the incidence relation among the clients and the products, and the constructed knowledge graph network is trained through the webpage ranking algorithm to obtain the knowledge graph network trained based on the webpage ranking algorithm. The construction principle of the knowledge graph network is the prior art, and is not specifically described herein.

The construction of the knowledge graph network takes each customer and each product as entities, can set intermediate elements based on the incidence relation between the customer and the product, the customer information of all the customers and the product information of all the products, and takes each intermediate element as an entity, if the intermediate element can be based on the use relation between the customer and the product, the use purpose or the use path is taken as the intermediate element, and the intermediate elements reflecting the characteristic attributes of the customer, such as the product use habit of the customer and the like, can be preset. And setting the relationship edges of each entity according to the actual business requirements, namely determining the recommended products through which associated elements are expected to pass.

The web page ranking algorithm for training the knowledge graph network is a google PageRank algorithm, is a technology for calculating according to mutual hyperlinks between web pages, and gives a PageRank to each web page to represent the importance of the web page, wherein the web page ranking of one web page is obtained by recursion algorithm from the importance of all links to the web page. The method is applied to the training of the knowledge-graph network, and the knowledge-graph network with each node having a corresponding webpage ranking value (PageRank, PR value) can be obtained. It should be noted that the web page ranking algorithm mentioned in the embodiment of the present invention is a generic term of a class of algorithms, including a traditional web page ranking algorithm, and also including an algorithm optimized based on the traditional web page ranking algorithm, such as a personalized web page ranking algorithm, which are all existing algorithms, and will not be described herein too much.

In the method provided by the embodiment of the invention, the nodes corresponding to the target customers and the nodes corresponding to each target product can be determined in the trained knowledge graph network through the customer IDs of the target customers and the product IDs of each target product.

S104: determining a ranking index of a node corresponding to each target product, wherein the ranking index of the node corresponding to each target product represents the probability of the node corresponding to the target client migrating to the node corresponding to the target product;

in the method provided by the embodiment of the invention, in the trained knowledge-graph network, the nodes corresponding to the target clients start to walk, the nodes corresponding to each target product are used as the terminal, and the probability of the nodes corresponding to the target clients to walk to each target product is determined based on the PR value of each node, so that the ranking index of the nodes corresponding to each target product, namely the importance of the nodes corresponding to each target product to the nodes corresponding to the target client products is determined.

S105: determining a recommendation index of each target product based on the ranking index of the node corresponding to the target product;

in the method provided by the embodiment of the invention, the recommendation index of the target product corresponding to each node can be determined through the ranking index of the node corresponding to each target product, the ranking index of the node can be directly used as the recommendation index of the target product corresponding to the node, and conversion can also be performed according to a certain corresponding relation, wherein generally, the higher the ranking index of the node is, the higher the recommendation index of the corresponding target product is. The recommendation index for a target product characterizes the extent to which the product is recommended for a target customer.

It should be noted that, in a specific implementation process, the ranking index of the corresponding node may be determined in the knowledge graph network only for the target product, so as to obtain the recommendation index of the target product. And determining the ranking indexes of the corresponding nodes of all the products to obtain the recommendation indexes of all the products, and removing the products purchased by the target customer to obtain the information corresponding to the target products. The specific implementation manner does not influence the implementation function of the method provided by the embodiment of the invention.

S106: and selecting a plurality of recommended products from all the target products according to the recommendation indexes of the target products, and recommending the target customers with the recommended products.

In the method provided by the embodiment of the invention, a plurality of recommended products can be selected from the target products according to the sequence from high to low of the recommendation indexes of the target products, and are subsequently used for recommending products to the target customers, for example, a recommended product list of the target customers is generated and provided for a customer manager to recommend, and the method can also be used for recommending various target customer-oriented terminals on pages.

Regarding the selection of recommended products, specifically, the number of recommended products may be preset, the target products are sorted according to the recommendation index from high to low, and the products with the preset number, which are sorted at the front, are selected as the recommended products according to the sorting order. The percentage can also be preset, the number of recommended products is determined according to the number of the current target products, and the recommended products with the corresponding number are selected according to the recommendation index from high to low. Index thresholds can also be preset, and products with each recommended index being greater than the preset threshold are taken as recommended products, and the like. It should be noted that, in the specific implementation process, the specific manner of selecting the recommended product may be set according to actual requirements, and the implementation function of the method provided by the embodiment of the present invention is not affected.

Based on the method provided by the embodiment of the invention, the nodes corresponding to each target product which is not purchased by the target customer can be determined based on the knowledge graph network trained based on the webpage ranking algorithm, the importance degree of the nodes corresponding to the target customer is relatively determined, the recommendation index of each target product is determined, and then the product recommended to the target customer is determined. The knowledge graph network covers all customers, all products and global information of incidence relations between the customers and the products, each node can be spread to a plurality of other nodes, the nodes corresponding to the customers are easy to spread to other nodes in the migration process, the probability that the nodes corresponding to the products can be migrated is high, and the knowledge graph network is beneficial to bringing a plurality of products into a recommendation range. Under the condition of less customer information, direct or indirect association between the nodes and various types of nodes in the knowledge graph network can be realized based on the nodes corresponding to the customers, so that the diversity of products recommended to the customers is improved, the accuracy of personalized recommendation is improved, the product recommendation effect is improved, and the service experience is improved.

In order to better illustrate the knowledge-graph network in the method provided by the embodiment of the present invention, in combination with the flowchart shown in fig. 2, on the basis of the method shown in fig. 1, the embodiment of the present invention provides another product recommendation method, wherein a construction process of the knowledge-graph network includes:

s201: determining all entities required for constructing the knowledge-graph network, wherein all entities comprise each customer, each product, each preset customer transaction characteristic and each preset transaction behavior characteristic;

in the method provided by the embodiment of the invention, the product is a settlement product of the financial institution, the settlement product refers to a product which is used for assisting a large-scale enterprise customer to scientifically analyze cash flow and determine a reasonable cash balance in order to meet the cash management needs of the enterprise by utilizing the advantages of talents, information, technology and equipment of the financial institution, and the surplus cash (including stock cash, current deposit, bank draft and other monetary assets) is used for short-term investment to increase the income of the enterprise. Common settlement products are settlement cards, timed cash pools, real-time fund pools, bill pools, payment and collection and bank to public networks. Correspondingly, the client in the method provided by the embodiment of the invention is also an enterprise client generally, which is also called a public client.

In the method provided by the embodiment of the invention, each customer, each product, each preset customer transaction characteristic and each preset transaction behavior characteristic are used as entities in the knowledge graph network. The client and the product can be obtained from the client directory and the product directory, and the transaction characteristics and the transaction behavior characteristics of the client can be set based on actual business requirements.

The customer transaction characteristics represent transaction characteristics reflected by a settlement transaction exchange completed by a customer, specifically, the transaction characteristics presented by the settlement transaction exchange can be determined based on the fund amount, the transaction frequency and the transaction time point of the customer transaction, and corresponding tags are set for each type of transaction characteristics to be respectively used as the customer transaction characteristics.

The transaction behavior characteristics represent behavior characteristics reflected by a settlement exchange using settlement products by a client, and specifically, the transaction behavior characteristics can be set based on transaction channels and transaction purposes of the settlement products used by the client, and a transaction behavior characteristic is represented by cross terms of each transaction channel and various transaction purposes, for example, if the transaction channel of a certain settlement product comprises A1, A2 and A3, and the transaction purposes correspond to B1 and B2, A1-B1, A1-B2, A2-B1, A2-B2, A3-B1 and A3-B2, which are transaction behavior characteristics respectively.

S202: acquiring customer information of each customer, transaction information of each customer and product information of each product;

in the method provided by the embodiment of the invention, the client information of all clients, including enterprise attributes of the clients and the like, can be acquired from the database. The method comprises the steps of obtaining transaction information of all clients, wherein the transaction information comprises settlement transaction records, transaction behavior characteristics of the clients, transaction characteristics of the clients and the like, and the transaction behavior characteristics of the clients and the transaction characteristics of the clients are correspondingly set labels after the settlement transaction records of the clients are analyzed in advance. Meanwhile, product information of all products, including attributes, purposes, and the like of the products, can be acquired.

S203: determining the corresponding relation between the entities according to the customer information of each customer, the transaction information of each customer and the product information of each product;

in the method provided by the embodiment of the invention, the association elements recommended by the product can be set according to actual requirements, namely, the nodes in the knowledge graph network are expected to be connected through the association relations, so that a certain node can be diffused to other nodes, and the association elements can also be understood as factors considered when the product is determined to be recommended.

Determining the corresponding relationship between the entities determined in step S201, that is, the entities are connected in a certain relationship, based on the actual demand, according to the customer information of all customers, the transaction information of the customers, and the product information of the products; for example, if the transaction characteristics are used as the association elements and the transaction information of the client a is used to determine that the client a corresponds to the transaction characteristics 1, it is determined that the entity client a and the entity transaction characteristics 1 have a corresponding relationship in a predetermined entity.

S204: constructing nodes corresponding to each entity, and constructing relationship edges between the nodes corresponding to each entity according to the corresponding relationship between each entity;

in the method provided by the embodiment of the invention, the relationship edges between the nodes corresponding to the entities can be established based on the corresponding relationship between the entities, and the entities without the corresponding relationship have no relationship edges. For example, customer a corresponds to transaction characteristic 1, and customer B also corresponds to transaction characteristic 1, a relationship edge is established between the node of customer a and the node of transaction characteristic 1, and a relationship edge is also established between the node of customer B and the node of transaction characteristic 1.

S205: setting the weight corresponding to each relation edge according to a preset weight assignment rule;

in the method provided by the embodiment of the invention, the weight assignment rule for setting the weight for each type of relation edge can be determined in advance according to the business requirements, such as what weight is given to the relation edge between products, what weight is given to the relation edge between the customer and the customer transaction characteristics, and the like.

After all entities and all relation edges to be established are established, an initial weight value can be set for each relation edge according to a preset weight assignment rule, and then weight normalization is carried out on various relation edges so as to set the weight corresponding to each relation edge.

S206: and generating the knowledge graph network based on the nodes corresponding to the entities, the relationship edges among the nodes corresponding to the entities and the weight corresponding to each relationship edge.

In the method provided by the embodiment of the invention, after the nodes corresponding to each entity are constructed, all the relation edges are established and the weight is set for each relation edge, a knowledge graph network can be generated based on the above contents, and the knowledge graph network can be regarded as a settlement transaction graph and covers the information of each client about settlement transactions.

Based on the method provided by the embodiment of the invention, the knowledge network map for settling the product recommendation can be constructed. On the other hand, the weights corresponding to various relation edges are determined according to actual service requirements, so that the importance of key elements in a network can be improved, and the service operation requirements can be favorably met.

Further, to better illustrate the process of determining the corresponding relationship between the entities in step S203, on the basis of the method shown in fig. 2, an embodiment of the present invention provides another product recommendation method, where in step S203, the determining the corresponding relationship between the entities according to the customer information of each customer, the transaction information of each customer, and the product information of each product includes:

in the method provided by the embodiment of the invention, the corresponding relations of upstream and downstream enterprises in a common supply chain can be preset, and the corresponding relations are represented according to the enterprise attributes. The upstream and downstream relationship between the respective customers of the financial institution can be determined based on the business attributes of the customers included in the customer information of the respective customers. It should be noted that the upstream and downstream relationship of the client is determined according to the actual enterprise attribute, and the enterprise client usually in the same supply chain may have the upstream and downstream relationship, not the relationship between every two clients.

in the method provided by the embodiment of the invention, the client transaction characteristic label included in the transaction information of each client can be acquired, and the client transaction characteristic label is a label preset according to the information such as the fund amount, the transaction frequency and the transaction time of the settlement transaction completed by the client. According to the label. And determining the customer transaction characteristics corresponding to the customer in each preset customer transaction characteristic to establish an association relationship.

in the method provided by the embodiment of the invention, the transaction behavior feature tag contained in the transaction information of each client can be acquired, the transaction behavior feature tag is set in advance according to the settlement transaction record of the client, and each transaction behavior feature tag corresponds to the settlement transaction made by the client under a certain transaction channel and a certain transaction purpose. According to the label, the association relationship between the customer and the corresponding transaction behavior characteristics can be determined in each preset transaction behavior characteristic.

in the method provided by the embodiment of the invention, each transaction content corresponds to a transaction channel, a transaction purpose and a used settlement product according to the settlement transaction record contained in the transaction information of each client, so that the settlement product used by the client under a certain transaction channel and a certain transaction purpose can be obtained, the product corresponding to each transaction behavior characteristic corresponding to the client is determined, and the use relationship between the transaction behavior characteristic corresponding to the client and the product can be established. In the method provided by the embodiment of the invention, the transaction channel can be an internet banking channel, a bill pool and the like, the transaction purpose can be property management fee payment, collection, cross-bank transfer and the like, and each settlement product has various corresponding transaction channels and various transaction purposes in a specific service scene, which is not described herein.

in the method provided by the embodiment of the invention, the purchased products of each customer can be determined according to the purchased product information of the customer contained in the transaction information of the customer, and a purchasing relationship is established between the customer and the purchased products.

In the method provided by the embodiment of the invention, under an actual service scene, a potential relation in service may exist between certain products, for example, a certain two products are conventional collocation in service. The association relationship among the products can be determined according to the product information and the actual business requirements of the products.

Further, based on the six types of relationships determined in the above process, the weight setting mentioned in step S205, in the method provided by the embodiment of the present invention, the relationship side corresponding to the upstream and downstream relationship between the customer and the client, the relationship side corresponding to the relationship between the customer and the corresponding customer transaction feature thereof, and the relationship side corresponding to the purchase relationship between the customer and the product are assigned with default initial weight values, that is, the initial value is one, for the relationship side corresponding to the relationship between the product and the product, the initial weight value is determined by the actual business requirement, and for the relationship side corresponding to the relationship between the customer and the transaction behavior feature, the initial weight value is determined by normalizing according to the number of transactions and the transaction amount corresponding to the customer under the transaction behavior feature, and the relationship side corresponding to the usage relationship between the transaction behavior feature and the product, the initial weight value is also determined according to the actual service requirement.

In order to better illustrate the method provided by the embodiment of the present invention, the embodiment of the present invention provides another product recommendation method, and on the basis of the method shown in fig. 1 or fig. 2, in the method provided by the embodiment of the present invention, as shown in fig. 3, the training process of the knowledge-graph network includes:

s301: according to a preset webpage ranking algorithm, iteratively calculating a webpage ranking value corresponding to each node in the constructed knowledge graph network until the webpage ranking value of each node in the knowledge graph network reaches a stable state;

in the method provided by the embodiment of the invention, each node in the knowledge graph network can be endowed with a webpage ranking value based on a webpage ranking algorithm, the webpage ranking (PageRank) value is a PR value in the webpage ranking algorithm, iterative calculation is carried out on the PR value on each node through the relationship edge among the nodes in the knowledge graph network and the weight of the relationship edge, and a stable PR value graph network can be formed through multiple iterations.

S302: when the webpage ranking value of each node in the knowledge graph network reaches a stable state, determining an individualized recommendation index corresponding to the knowledge graph network;

in the method provided by the embodiment of the invention, after the PR value of each node in the knowledge graph network is stable, the individual recommendation index corresponding to the knowledge graph network can be determined based on the current knowledge graph network, and the individual recommendation index represents the accuracy of determining recommended products by the current knowledge graph network. Specifically, the method includes the steps that a sample client is selected to be used for testing the knowledge graph network, a node corresponding to the sample client is used as a testing node, the ranking index of the node corresponding to each product is determined, namely the probability that the testing node walks to the node corresponding to each product is determined, the recommendation index of each product for the sample client is determined, all products are ranked according to the recommendation index from high to low, product information purchased by the sample client is obtained, and the personalized recommendation index corresponding to the knowledge graph network is determined according to the matching condition of the products purchased by the sample client and the products ranked in the front. Specifically, the personalized recommendation index can be determined by adopting a personalized recommendation index calculation mode of the existing recommendation model.

S303: judging whether the personalized recommendation index meets a preset effective index condition or not;

in the method provided by the embodiment of the invention, the effective index conditions of the personalized recommendation index can be set in advance according to the service requirements, namely, when the conditions that the personalized recommendation index meets are met, the recommendation effect representing the network is effective. For example, an effective index value range may be set, for example, in a calculation manner based on the personalized recommendation index, when the product purchased by the sample customer is increasingly matched with the product ranked in the front, the smaller the value of the personalized index is, the personalized recommendation index is a percentage value, and thus 0 to 10% of the value range is set as the effective index value range. When the value of the personalized recommendation index falls into the range (namely the personalized recommendation index is less than or equal to 10%), representing that the personalized recommendation index meets the preset effective index condition. It should be noted that the effective index condition provided by the embodiment of the present invention is only one embodiment, and is not limited to the determination standard of the actual application.

If the personalized recommended index does not meet the preset effective index condition, the step S304 is executed. If the personalized recommendation index meets the preset effective index condition, the method goes to step S305, and the training of the knowledge graph network is finished, and the network can be used for product recommendation.

S304: adjusting the weight corresponding to each relation edge in the knowledge graph network;

if the personalized recommendation index does not meet the preset effective index condition, resetting the weight corresponding to each relationship edge in the knowledge graph network, adjusting the weight value of each edge, then returning to the step S301, and continuing to use the webpage ranking algorithm to perform iterative calculation on the PR value of each node until the personalized recommendation index of the stabilized knowledge graph network meets the preset effective index condition.

S305: and finishing the training of the knowledge-graph network.

Based on the method provided by the embodiment of the invention, in the process of training the knowledge graph network by applying the webpage ranking algorithm, whether the knowledge graph network can meet the recommendation requirement can be judged through the personalized recommendation index, and the recommendation accuracy is favorably improved.

In a specific application process, the trained knowledge graph network can be applied, part of customers are extracted to determine recommended products, actual product recommendation is carried out, and the weight of the knowledge graph network is continuously adjusted for training according to the actual marketing feedback result.

Further, on the basis of any one of the foregoing embodiments, in the method provided by the embodiment of the present invention, the determining process of the ranking index of the node corresponding to each target product in step S104 includes:

In the method provided by the embodiment of the invention, all paths from the node corresponding to the target customer to the node corresponding to the target product can be determined through the relation edges among all nodes in the knowledge graph network. According to the PR value of each node of each path, the PR value corresponding to the path, namely the probability of wandering along the path in the network can be calculated. And adding the PR values of all the paths corresponding to the nodes of the target product, and taking the sum of the PR values of all the paths as the ranking index of the nodes of the target product. I.e., the ranking index of the node corresponding to each target product, is the sum of the PR values of all paths that travel from the node of the target client to the node of the product.

Based on the method provided by the embodiment of the invention, all paths of the nodes of the target customer to the nodes of the target product are considered, the target product is linked by various associated factors, and the diversity of the recommended product can be improved.

Further, on the basis of the method provided in any of the above embodiments, in the method provided in the embodiment of the present invention, the step S106 of selecting each recommended product from the target products includes:

In the method provided by the embodiment of the invention, the threshold value can be preset, and the target product with the recommendation index larger than the threshold value is recommended as the recommended product. In a specific implementation process, the recommendation indexes of the target products are presented in a list mode, the target products can be sorted in the list according to the sequence from high to low of the recommendation indexes of the target products, and the list of the recommended products is determined according to the comparison between the recommendation indexes of the target products and a threshold value.

According to the method provided by the embodiment of the invention, the recommended products are defined by the threshold value, so that the method is convenient and quick, and meanwhile, the information is presented in a table mode, so that the visualization effect is good.

Further, in the method provided by the embodiment of the present invention, the web page ranking algorithm for training the knowledge graph network adopts a Personalized web page ranking algorithm, and the Personalized web page ranking algorithm refers to a Personalized Pagerank algorithm, which is an existing algorithm improved based on a traditional web page ranking algorithm and is not described herein. Based on the method provided by the embodiment of the invention, in the process of training the knowledge graph network, each jump in random walk can not be randomly selected to any node, and certain nodes, namely nodes with higher weight values, need to jump, which is beneficial to adjusting the influence factors of key elements in product recommendation according to business requirements.

Corresponding to the product recommendation method shown in fig. 1, an embodiment of the present invention further provides a product recommendation device, which is used for implementing the method shown in fig. 1 specifically, and a schematic structural diagram of the product recommendation device is shown in fig. 4, where the product recommendation device includes:

a first determining unit 401, configured to determine, when a product recommendation request sent by a user is received, a target client corresponding to the product recommendation request;

a second determining unit 402 for determining each product not purchased by the target customer among all products as a target product;

a third determining unit 403, configured to determine nodes corresponding to the target customers and nodes corresponding to each target product in a knowledge graph network trained based on a web page ranking algorithm, where the knowledge graph network is constructed according to all customers, all products, and association relationships between each customer and each product;

a fourth determining unit 404, configured to determine a ranking index of a node corresponding to each target product, where the ranking index of a node corresponding to each target product represents a probability that a node corresponding to the target client walks to the node corresponding to the target product;

a fifth determining unit 405, configured to determine a recommendation index of each target product based on the ranking index of the node corresponding to the target product;

the selecting unit 406 is configured to select a plurality of recommended products from all the target products according to the recommendation index of each target product, and recommend the target customer with each recommended product.

Based on the device provided by the embodiment of the invention, the nodes corresponding to each target product which is not purchased by the target customer can be determined based on the knowledge graph network trained based on the webpage ranking algorithm, the importance degree of the nodes corresponding to the target customer is relatively determined, the recommendation index of each target product is determined, and then the product recommended to the target customer is determined. The knowledge graph network covers all customers, all products and global information of incidence relations between the customers and the products, each node can be spread to a plurality of other nodes, the nodes corresponding to the customers are easy to spread to other nodes in the migration process, the probability that the nodes corresponding to the products can be migrated is high, and the knowledge graph network is beneficial to bringing a plurality of products into a recommendation range. Under the condition of less customer information, direct or indirect association between the nodes and various types of nodes in the knowledge graph network can be realized based on the nodes corresponding to the customers, so that the diversity of products recommended to the customers is improved, the accuracy of personalized recommendation is improved, the product recommendation effect is improved, and the service experience is improved.

Further, on the basis of the above apparatus, the apparatus provided in the embodiment of the present invention further includes:

Further, on the basis of the above apparatus, in an apparatus provided in an embodiment of the present invention, the seventh determining unit includes:

Further, on the basis of the foregoing apparatus, in the apparatus provided in the embodiment of the present invention, the fourth determining unit 404 includes:

Further, on the basis of the above apparatus, in the apparatus provided in the embodiment of the present invention, the selecting unit 406 includes:

Further, on the basis of the device, the webpage ranking algorithm in the embodiment of the invention is a personalized webpage ranking algorithm.

The embodiment of the invention also provides a storage medium, which comprises stored instructions, wherein when the instructions are executed, the equipment where the storage medium is located is controlled to execute the product recommendation method.

An electronic device is provided in an embodiment of the present invention, and the structural diagram of the electronic device is shown in fig. 5, which specifically includes a memory 501 and one or more instructions 502, where the one or more instructions 502 are stored in the memory 501, and are configured to be executed by one or more processors 503 to perform the following operations according to the one or more instructions 502:

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the system or system embodiments are substantially similar to the method embodiments and therefore are described in a relatively simple manner, and reference may be made to some of the descriptions of the method embodiments for related points. The above-described system and system embodiments are only illustrative, wherein the units described as separate parts may or may not be physically separate, and the parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.

Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for recommending products, comprising:

2. The method of claim 1, wherein the construction of the knowledge-graph network comprises:

3. The method of claim 2, wherein determining the correspondence between each of the entities based on customer information for each of the customers, transaction information for each of the customers, and product information for each of the products comprises:

4. The method according to claim 1 or 2, wherein the training process of the knowledge-graph network comprises:

5. The method of claim 1, wherein determining the ranking index of the node corresponding to each target product comprises:

6. The method of claim 1, wherein the selecting each recommended product among each of the target products according to the recommendation index of each of the target products comprises:

7. The method of claim 1, wherein the web page ranking algorithm is a personalized web page ranking algorithm.

8. A product recommendation device, comprising:

9. A storage medium comprising stored instructions, wherein the instructions, when executed, control a device on which the storage medium is located to perform a product recommendation method according to any one of claims 1 to 7.

10. An electronic device comprising a memory and one or more instructions, wherein the one or more instructions are stored in the memory and configured to be executed by the one or more processors to perform the method of any one of claims 1-7.