Product recommendation method and device
By constructing a product attribute relationship diagram and using a pre-trained recommendation model, the problems of low efficiency and poor accuracy of manual recommendation in the existing technology are solved, efficient and accurate product recommendations are achieved, and professionalism is improved.
Patent Information
- Application Number
- CN202411171022.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-23
- Publication Date
- 2024-12-17
AI Technical Summary
The existing manual product recommendation methods rely on historical experience, resulting in low recommendation efficiency and error-proneness in the case of a wide variety of products to be recommended, and cannot meet actual needs.
By obtaining customer product contract information, business information and attribute information, building a product attribute relationship diagram, eliminating mutually exclusive products, and using pre-trained product recommendation models to generate an efficient and accurate product portfolio to ensure the accuracy and professionalism of recommended products.
It realizes efficient and accurate product recommendations, reduces personnel workload, avoids the existence of mutually exclusive products in recommendations, and improves the professionalism of product recommendations.
Smart Images
Figure CN119151415A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology, and in particular to a product recommendation method and device. Background Art
[0002] In the field of logistics services, in order to provide better and more convenient services to logistics supply chain customers, logistics companies have developed various logistics service products, such as pharmaceutical cold chain products that provide cold chain services for pharmaceutical goods. Given that different businesses and different categories of goods correspond to different logistics service products, and facing a wide variety of products, and some products are mutually exclusive, existing product recommendation methods usually rely on manual historical experience to determine the appropriate recommended products or recommended product combinations based on the products that the customer has signed, in order to achieve the purpose of product promotion and customer service.
[0003] In the process of implementing the present invention, the inventors found that the prior art has the following problems:
[0004] The existing manual product recommendation method is highly dependent on historical experience. When there are many types of products to be recommended, the efficiency of product recommendation is low and it is prone to errors due to the lack of regularity in product recommendations, which cannot meet actual needs well. Summary of the invention
[0005] In view of this, an embodiment of the present invention provides a product recommendation method and device, which realizes efficient and accurate product recommendations, not only reducing the workload of personnel, but also making product recommendations based on the combination features that affect the customer's willingness to sign a contract and the product combination set that eliminates mutually exclusive products, thereby ensuring the accuracy of the recommended products and avoiding the existence of mutually exclusive products in the recommended products, thereby improving the professionalism of the product recommendations.
[0006] To achieve the above object, according to one aspect of an embodiment of the present invention, a product recommendation method is provided, comprising:
[0007] Acquire the customer's product contract information, business information, and attribute information, and combine the product contract information, the business information, and the attribute information into a combined feature of the customer;
[0008] Obtain a product attribute relationship graph, wherein the product attribute relationship graph defines each product to be recommended and relationship information between each of the products to be recommended, and each of the products to be recommended has a mutually exclusive relationship; combine each of the products to be recommended, and according to the relationship information, remove mutually exclusive products from the combined combination of products to be recommended, to generate a product combination set without a mutually exclusive relationship;
[0009] The product combination set and the combination features are input into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model.
[0010] Optionally, the product attribute relationship graph exists in the form of a product knowledge graph, each node in the product knowledge graph represents a product, and the line between two nodes represents the relationship between the two products.
[0011] Optionally, each of the products to be recommended is combined, and according to the relationship information, mutually exclusive products are removed from the combined product combinations to be recommended to generate a product combination set without mutually exclusive relationships, including: obtaining each of the products to be recommended according to the information of the nodes in the product knowledge graph; combining each of the products to be recommended to obtain multiple product combinations, and combining each of the product combinations and each of the products to be recommended to form a product element set; for each component element in the product element set: searching the product knowledge graph for the node corresponding to each product in the component element, and judging whether the component element contains mutually exclusive products according to the connection relationship between the nodes; and if the component element contains mutually exclusive products, deleting the component element; and forming a product combination set with the remaining components in the product element set.
[0012] Optionally, each of the products to be recommended is combined, and according to the relationship information, mutually exclusive products are removed from the combined product combinations to be recommended to generate a product combination set without mutually exclusive relationships, including: obtaining each of the products to be recommended according to the information of the nodes in the product knowledge graph; combining each of the products to be recommended to obtain multiple product combinations, and combining each of the product combinations and each of the products to be recommended to form a product element set; for each product to be recommended: searching and identifying the exclusive product corresponding to the product to be recommended from the product knowledge graph; and combining the product to be recommended and the exclusive product into mutually exclusive products; deleting the constituent elements in the product element set that contain the mutually exclusive products to obtain a product combination set.
[0013] Optionally, the product element set exists in the form of a node graph, and the node graph is constructed by taking each of the product combinations and each of the products to be recommended in the product element set as graph nodes, thereby obtaining a constraint node graph corresponding to the product combination set; the product combination set and the combination features are input into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model, including: inputting the constraint node graph and the combination features into a pre-trained product recommendation model constructed by a graph convolutional neural network model; through each convolution layer in the product recommendation model, combining the combination features to screen each level of adjacent nodes in the constraint node graph to obtain recommended nodes; parsing the recommended nodes to obtain corresponding recommended product combinations.
[0014] Optionally, the product recommendation model is trained in the following manner: training customers are determined from the customer base, and business information and attribute information of the training customers are retrieved according to the time granularity to form training features of the training customers at each time granularity; a product combination training set corresponding to the training customers is generated according to the product attribute relationship diagram; product contract signing information of the training customers at each time granularity is used as a sample label; training samples are constructed according to the training features, the product combination training set and the sample labels; and the training samples are used to train the product recommendation model by means of a loss function and gradient descent to obtain a trained product recommendation model.
[0015] Optionally, the method also includes: configuring multiple business goals for recommending products for the customer; inputting each of the business goals together with the product combination set and the combination features into a pre-trained product recommendation model to obtain the recommended product combination for the customer under each of the business goals output by the product recommendation model; based on the received recommendation result query request with business goals, querying the recommended product combinations for the customer under each of the business goals for a product combination that is consistent with the business goal in the recommendation result query request, and returning the queried product combination to the requester of the recommendation result query request.
[0016] According to a second aspect of an embodiment of the present invention, a product recommendation method is provided, comprising:
[0017] According to the received product recommendation request, obtain the customer's product contract real-time information, business real-time information, and attribute information, and combine the product contract real-time information, the business real-time information, and the attribute information into a combined feature of the customer;
[0018] Acquire a product attribute relationship graph, wherein the product attribute relationship graph has each product to be recommended and relationship information between each product to be recommended, and each product to be recommended has a mutually exclusive relationship; combine each product to be recommended, and according to the relationship information, remove mutually exclusive products from the combined combination of products to be recommended, so as to generate a product combination set without a mutually exclusive relationship;
[0019] The product combination set and the combination features are input into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model.
[0020] According to a third aspect of an embodiment of the present invention, there is provided a product recommendation device, comprising:
[0021] A feature determination module, used to obtain the customer's product contract information, business information, and attribute information, and combine the product contract information, the business information, and the attribute information into a combined feature of the customer;
[0022] A combination determination module is used to obtain a product attribute relationship diagram, wherein the product attribute relationship diagram defines each product to be recommended and the relationship information between each product to be recommended, and each product to be recommended has a mutually exclusive relationship; each product to be recommended is combined, and according to the relationship information, mutually exclusive products are removed from the combined product combination to be recommended, so as to generate a product combination set without a mutually exclusive relationship;
[0023] The product recommendation module is used to input the product combination set and the combination features into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model.
[0024] According to a fourth aspect of an embodiment of the present invention, there is provided an electronic device for product recommendation, including:
[0025] one or more processors;
[0026] a storage device for storing one or more programs,
[0027] When the one or more programs are executed by the one or more processors, the one or more processors implement the method provided in the first aspect of the embodiment of the present invention.
[0028] According to a fifth aspect of an embodiment of the present invention, there is provided a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method provided in the first aspect of the embodiment of the present invention is implemented.
[0029] According to a sixth aspect of an embodiment of the present invention, a computer program is provided. When the computer program is executed by a processor, the method provided by the first aspect of the embodiment of the present invention is implemented.
[0030] An embodiment of the invention has the following advantages or beneficial effects: by obtaining the customer's product contract information, business information, and attribute information, and combining the product contract information, business information, and attribute information into the customer's combined features; obtaining a product attribute relationship diagram, which defines each product to be recommended and the relationship information between each product to be recommended, and there is a mutually exclusive relationship between each product to be recommended; combining each product to be recommended, and according to the relationship information, eliminating mutually exclusive products from the combined product combination to be recommended, and generating a product combination set without a mutually exclusive relationship; inputting the product combination set and the combined features into a pre-trained product recommendation model to obtain a technical solution for the recommended product combination output by the product recommendation model, thereby achieving efficient and accurate product recommendations, which not only reduces the workload of personnel, but also makes product recommendations based on the combined features that affect the customer's willingness to sign a contract and the product combination set that eliminates mutually exclusive products, which not only ensures the accuracy of the recommended products, but also avoids the existence of mutually exclusive products in the recommended products, thereby improving the professionalism of the product recommendations. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.
[0032] Figure 1 is a schematic diagram of the main process of the product recommendation method according to an embodiment of the present invention;
[0033] Figure 2 is a schematic diagram of a product knowledge graph according to an embodiment of the present invention;
[0034] Figure 3 is a schematic diagram of the process of constructing a product recommendation model according to an embodiment of the present invention;
[0035] Figure 4 is a schematic diagram of generation and application of a constraint node graph according to an embodiment of the present invention;
[0036] Figure 5 is a schematic diagram of the composition of a product recommendation model according to an embodiment of the present invention;
[0037] Figure 6 Schematic diagram of the principle of the product recommendation method according to an embodiment of the present invention;
[0038] Figure 7 is a schematic diagram of main modules of a product recommendation device according to an embodiment of the present invention;
[0039] Figure 8 is an exemplary system architecture diagram to which embodiments of the present invention may be applied;
[0040] Fig. 9It is a schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing an embodiment of the present invention. DETAILED DESCRIPTION
[0041] It should be noted that in the technical solution of the present invention, the collection / collection, update, analysis, use, transmission, storage and other aspects of user personal information involved are in compliance with the provisions of relevant laws and regulations, are used for legal and reasonable purposes, are not shared, leaked or sold outside of these legal uses, and are subject to supervision and management by national regulatory authorities. Necessary measures should be taken for user personal information to selectively block the use or access to personal information data to prevent illegal access to such personal information data, ensure that persons who have access to personal information data comply with the provisions of relevant laws and regulations, and ensure the security of user personal information. In addition, once these user personal information data are no longer needed, the risks should be minimized by limiting or even prohibiting data collection and / or deleting data.
[0042] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.
[0043] The existing manual product recommendation method is highly dependent on historical experience. When there are many types of products to be recommended, the efficiency of product recommendation is low and it is prone to errors due to the lack of regularity in product recommendations, which cannot meet actual needs well.
[0044] In order to solve the above problems existing in the prior art, the present invention proposes a product recommendation method, which combines the products to be recommended based on the obtained product attribute relationship diagram that defines the products to be recommended and the product relationships, and eliminates mutually exclusive products from the combined product combinations to be recommended to generate a product combination set without mutually exclusive relationships, and then uses the product recommendation model to construct combination features and product combination sets based on relevant information that affects the customer's willingness to sign a contract as model inputs to obtain recommended product combinations, thereby achieving efficient and accurate product recommendations, which not only reduces the workload of personnel, but also makes product recommendations based on the combination features that affect the customer's willingness to sign a contract and the product combination sets that eliminate mutually exclusive products, thereby ensuring the accuracy of the recommended products and avoiding the existence of mutually exclusive products in the recommended products, thereby improving the professionalism of the product recommendations.
[0045] Figure 1 is a schematic diagram of the main process of the product recommendation method according to an embodiment of the present invention. Figure 1As shown, the product recommendation method of the embodiment of the present invention includes the following steps S101 to S103.
[0046] Step S101: Acquire the customer's product contract information, business information, and attribute information, and combine the product contract information, the business information, and the attribute information into a combined feature of the customer.
[0047] Specifically, in the product promotion business, the mining of the needs of existing customers is a major direction of product promotion. In order to provide accurate and adapted new products to existing customers, it is necessary to first obtain the user's demand characteristics. In order to ensure the accuracy of the demand characteristics, the embodiment of the present invention obtains relevant information that affects the customer's willingness to sign a contract from the underlying data platform, such as the customer's current contracted product information (product name, contract signing time, etc.), the customer's business data information (order volume, consultation volume, pageview volume, main order line, order time, etc.) and the customer's attribute information (customer manager's visit record, customer portrait: customer location, level, main business category, etc.), and combines the obtained customer's product contract information, business information, and attribute information to construct the customer's combined characteristics.
[0048] In addition, when combining product contract information, business information and attribute information into a combined feature of a customer, the Occam's razor principle can also be used to filter out key information from the product contract information, business information and attribute information respectively, and combine each key information into a combined feature of the customer.
[0049] It can be understood that the product recommendation of the embodiment of the present invention can be applicable to the recommendation of physical products, such as the recommendation of new electronic equipment products, and can also be applied to the recommendation of financial management products, and can also be applied to the recommendation of logistics service products, such as cold chain services, warehousing services, etc. For the sake of ease of explanation, the embodiment of the present invention mainly takes the recommendation of logistics service products as an example.
[0050] Based on the above-mentioned combined features constructed from various relevant information that affects customers' willingness to sign contracts, comprehensive and reliable data support is provided for product recommendations, ensuring the accuracy of the recommendation results.
[0051] Step S102, obtaining a product attribute relationship diagram, wherein the product attribute relationship diagram defines each product to be recommended and the relationship information between each of the products to be recommended, and each of the products to be recommended has a mutually exclusive relationship; combining each of the products to be recommended, and according to the relationship information, removing mutually exclusive products from the combined combination of products to be recommended, to generate a product combination set without a mutually exclusive relationship.
[0052] Specifically, the embodiment of the present invention pulls all product information from the product library, extracts the attribute information (product type, product name, etc.) of each product and the relationship information (mutually exclusive, binding, etc.) between products according to keywords or identifiers, generates and stores a product attribute relationship graph that defines the relationship information between each product to be recommended and the products to be recommended. When the recommendation system makes a product recommendation, the product attribute relationship graph is obtained, and each product to be recommended is combined according to the product attribute relationship graph. The combination of the embodiment of the present invention is to traverse any and unordered combination of each product to be recommended, for example, the combination of the three products N, B and C is NB, NC, BC and NBC. Then, according to the relationship information between each product to be recommended defined in the product attribute relationship graph, the products with mutually exclusive relationships in the combined combination of products to be recommended are identified, and are removed from the combined combination of products to be recommended to obtain a set of product combinations that do not have mutually exclusive relationships.
[0053] According to one embodiment of the present invention, the product attribute relationship graph exists in the form of a product knowledge graph, each node in the product knowledge graph represents a product, and the line between two nodes represents the relationship between the two products.
[0054] Specifically, in order to improve the readability and standardization of the product attribute relationship diagram, the embodiment of the present invention converts the product attribute relationship diagram into the form of a product knowledge graph, represents each product with a node, and attaches the attribute information of each product around the corresponding node, and then uses different connection colors, virtual and real lines and other visual features between two nodes to represent the relationship between products.
[0055] Figure 2 It is a schematic diagram of the product knowledge graph of an embodiment of the present invention. The figure illustrates the attributes and relationship information of four logistics service products (first-come-first-served, signed return, pharmaceutical cold chain and pharmaceutical LTL). Each node represents a product, and the information around each node is product attribute information. For example, for the pharmaceutical LTL node, the product type of the pharmaceutical LTL product is the main product, and the charge is the standard price. It is suitable for pharmaceutical retail enterprises in B2B business, requiring the side length and volume of the consigned items to meet XX conditions, and the consigned items will be temperature controlled. The dotted connecting lines in the figure represent mutually exclusive relationships, the thick solid connecting lines represent binding relationships, and the regular solid connecting lines represent the choice of binding on demand.
[0056] The representation of product knowledge in the form of graphs improves readability and standardization, making it easier to parse the product knowledge graph through algorithm programs and obtain the required data information quickly, accurately and efficiently.
[0057] According to another embodiment of the present invention, each of the products to be recommended is combined, and according to the relationship information, mutually exclusive products are eliminated from the combined product combinations to be recommended to generate a product combination set without mutually exclusive relationships, including: obtaining each of the products to be recommended according to the information of the nodes in the product knowledge graph; combining each of the products to be recommended to obtain multiple product combinations, and combining each of the product combinations and each of the products to be recommended to form a product element set; for each component element in the product element set: searching the product knowledge graph for the node corresponding to each product in the component element, and judging whether the component element contains mutually exclusive products according to the connection relationship between the nodes; and deleting the component element if the component element contains mutually exclusive products; and forming a product combination set with the remaining components in the product element set.
[0058] Specifically, based on the product knowledge graph constructed above, the information of the nodes in the product knowledge graph is read, and the nodes with the same customer group are determined according to the customer's customer group, such as pharmaceutical retail, and then the various products to be recommended are integrated and summarized. Then, the various products to be recommended are combined to obtain multiple product combinations. Considering that the subsequent product recommendation model needs to learn the dynamic changes of the contracted products of customers at different time granularities, in order to obtain more training knowledge, it is also necessary to add each product to be recommended to the product element set to form a complete product element set, that is, the product element set contains both product combinations after various products are combined and individual products. The constituent elements in the product element set can be product combinations or individual products.
[0059] Furthermore, for each component element in the product element set, the node corresponding to each product contained in the component element is searched from the product knowledge graph, and according to the connection relationship between the nodes presented in the product knowledge graph, it is determined whether the products contained in the component element have a mutually exclusive relationship. If there is a mutually exclusive relationship, the component element is deleted. If there is no mutually exclusive relationship, the component element is retained and it is determined that the component element is a reasonable product combination. Finally, the remaining components in the product element set after traversing all the components are combined into a product combination set that does not have a mutually exclusive relationship.
[0060] According to another embodiment of the present invention, each of the products to be recommended is combined, and according to the relationship information, mutually exclusive products are removed from the combined product combinations to be recommended to generate a product combination set without mutually exclusive relationships, including: obtaining each of the products to be recommended according to the information of the nodes in the product knowledge graph; combining each of the products to be recommended to obtain multiple product combinations, and combining each of the product combinations and each of the products to be recommended to form a product element set; for each product to be recommended: searching and identifying the exclusive product corresponding to the product to be recommended from the product knowledge graph; and combining the product to be recommended and the exclusive product into mutually exclusive products; deleting the constituent elements in the product element set that contain the mutually exclusive products to obtain a product combination set.
[0061] Specifically, the embodiment of the present invention also provides another method for removing mutually exclusive products from the combined product combination to be recommended to generate a product combination set without a mutually exclusive relationship. The construction of the product element set is similar to the above, and each product to be recommended is obtained based on the information of the node in the product knowledge graph, and then each product to be recommended is combined to form a product element set with the obtained multiple product combinations and each product to be recommended.
[0062] Furthermore, for each product to be recommended as determined above, the exclusion node with a mutually exclusive relationship with it is searched and identified from the product knowledge graph, and then the exclusion product corresponding to the exclusion node is obtained, and the product to be recommended and the exclusion product are combined into a pair of mutually exclusive products. After traversing all the products to be recommended, a list of mutually exclusive products is obtained. By comparing the mutually exclusive product list, the component elements in the product element set containing mutually exclusive products are deleted to obtain a product combination set without a mutually exclusive relationship.
[0063] The embodiment of the present invention makes full use of the product knowledge graph to obtain products with mutually exclusive relationships, and deletes the constituent elements with mutually exclusive relationships in the product element set to ensure that there are no mutually exclusive product combinations in the finally generated product combination set.
[0064] Step S103: input the product combination set and the combination features into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model.
[0065] Specifically, based on the above-constructed product combination set that is not mutually exclusive and the combination features that characterize customer demand characteristics, the learning ability of the intelligent learning model is utilized to pre-train and construct a product recommendation model, and the product combination set and the combination features are input into the pre-trained product recommendation model. The product recommendation model recommends suitable product combinations for customers based on the acquired input knowledge and outputs the recommendation results.
[0066] According to one embodiment of the present invention, the product recommendation model is trained in the following manner: training customers are determined from a customer base, and business information and attribute information of the training customers are retrieved according to time granularity to form training features of the training customers at each time granularity; a product combination training set corresponding to the training customers is generated according to the product attribute relationship graph; product contract signing information of the training customers at each time granularity is used as a sample label; training samples are constructed according to the training features, the product combination training set and the sample labels; and the training samples are used to train the product recommendation model by means of a loss function and gradient descent to obtain a trained product recommendation model.
[0067] Specifically, the training of the product recommendation model is inseparable from the construction of the training sample set. First, it is necessary to select training customers who can represent different customer groups from the customer base. Then, according to the time granularity, the business information and attribute information of the training customers are retrieved and integrated to construct <integrated data, timestamp>, and the training features of the training customers at each time granularity are obtained. At the same time, according to the product attribute relationship diagram, a product combination training set corresponding to the training customers that does not contain mutually exclusive relationships is generated. Finally, the contracted product information of the training customers at the corresponding time granularity is used as the sample label, and the training samples are formed using the training features, product combination training set and sample labels. Using the training samples, the product recommendation model is trained by the loss function and gradient descent. When the training converges, the trained product recommendation model is obtained.
[0068] Figure 3 This is a schematic diagram of the process of building a product recommendation model in an embodiment of the present invention. First, determine the training features of the training customers and the training set of the training customers' product combinations, and build the training samples of the product recommendation model in combination with the sample labels; divide the training samples into training sets, validation sets, and test sets according to the configured 7:1:2, and use the training sets in the training samples to perform multi-step training on the product recommendation model by gradient descent, and update the parameters of the product recommendation model; then evaluate and tune the model in combination with the test set and validation set in the training samples, and finally obtain a product recommendation model that can be used to recommend products.
[0069] According to another embodiment of the present invention, the product element set exists in the form of a node graph, and the node graph is constructed by taking each of the product combinations and each of the products to be recommended in the product element set as graph nodes, thereby obtaining a constraint node graph corresponding to the product combination set.
[0070] Specifically, an embodiment of the present invention represents the product element set in the form of a diagram, assigns corresponding graph nodes to each product combination and product to be recommended in the product element set, deletes the constituent elements where mutually exclusive products are located, and obtains a constraint node graph composed of the remaining graph nodes, in which there are no longer mutually exclusive product combinations in the constraint node graph.
[0071] Figure 4 Schematic diagram of the generation and application of the constraint node graph of the embodiment of the present invention. Taking three products to be recommended, N, B, and C, as an example, N, B, and C are combined to obtain<N,B> ,<B,C> ,<N,C> ,<N,B,C> These four combinations will be recommended products <n> 、 、 <c>The product element set is also added. In addition, in order to facilitate the subsequent model training to learn as much training knowledge as possible, considering that the initial state of the customer is that he has not signed any product, <> is also added to the product element set to obtain a node graph representing the product element set. According to the product knowledge graph, the exclusion products of the three recommended products N, B, and C are searched. The search result shows that B and C are mutually exclusive products. Then the component nodes containing B and C in the node graph are deleted, such as Figure 4 Add the gray nodes in the constraint node graph to get the constraint node graph.
[0072] Furthermore, when the constraint node graph corresponds to the product combination training set in the training sample of the model, the training features and sample labels in the training sample are combined to obtain a specific display of the training process in the constraint node graph. From the arrows between the nodes in the graph, we can see the product contract features of different training customers at different time granularities, for example <>→<N,C> It means that a training customer initially did not sign a contract for the task product, but signed a contract for the product combination N and C at the next time granularity.
[0073] According to another embodiment of the present invention, the product combination set and the combination features are input into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model, including: inputting the constraint node graph and the combination features into a pre-trained product recommendation model constructed by a graph convolutional neural network model; using each convolution layer in the product recommendation model in combination with the combination features to screen each level of adjacent nodes in the constraint node graph to obtain recommended nodes; parsing the recommended nodes to obtain corresponding recommended product combinations.
[0074] Specifically, based on the above-mentioned use of constraint node graphs to characterize product combination sets, the embodiment of the present invention uses a graph convolutional neural network model to construct a product recommendation model. The constraint node graph and combination features are input into the product recommendation model constructed by the pre-trained graph convolutional neural network. The product recommendation model can be expressed by the following formula:
[0075]
[0076] Among them, H (l) Represents the feature vector of the node at layer l, W (l) represents the parameters of the lth layer of convolution, A represents the adjacency matrix of the feature matrix, and D represents the metric matrix of the adjacency matrix. Through each convolution layer in the product recommendation model, combined with the combined features, the features of the nodes in the constraint node graph are updated to the weighted sum of their adjacent nodes, and processed by linear transformation and nonlinear activation function, and through multiple iterations, the customer's recommended node is obtained. Finally, the recommended node is parsed and the product combination recommended to the customer is obtained from the constraint node graph.
[0077] Figure 5 It is a schematic diagram of the composition of the product recommendation model of an embodiment of the present invention. The product recommendation model mainly includes an input layer, a graph convolution layer and an output layer, wherein the graph convolution layer is composed of a plurality of convolution layers and corresponding activation functions superimposed, each convolution layer combines the corresponding activation function to process the first-order neighborhood information in the constraint node graph, and transmits multi-order neighborhood information through several convolution layers, and finally outputs the recommended nodes in the constraint node graph through the output layer.
[0078] According to another embodiment of the present invention, the method also includes: configuring multiple business goals for recommending products for the customer; inputting each of the business goals together with the product combination set and the combination features into a pre-trained product recommendation model to obtain the recommended product combination for the customer under each of the business goals output by the product recommendation model; based on the received recommendation result query request with business goals, querying the recommended product combinations for the customer under each of the business goals for a product combination that is consistent with the business goal in the recommendation result query request, and returning the queried product combination to the requester of the recommendation result query request.
[0079] Specifically, considering different business goals will correspond to different product recommendation strategies. For example, in order to motivate customers, a product combination with a relatively large discount will be recommended. In order to renew the contracts of customers who are about to expire, a product combination with a discount for renewal will be recommended. In order to achieve target-oriented product recommendations, multiple business goals for recommending products are first configured for customers. Each business goal is input into the product recommendation model together with the product combination set and combination characteristics. The product recommendation model outputs the recommended product combination that meets each business goal based on the received business goals. Based on this, the account manager sends a recommended product query request to the product recommendation system by selecting a specific business goal on the client. The product recommendation system receives the query request, parses it into the business goal contained therein, queries the product combination consistent with the business goal from the results output by the product recommendation model, and returns the recommended product combination that meets the business goal to the account manager.
[0080] The embodiment of the present invention is based on the above-mentioned goal-oriented product recommendation method, which supports customer managers to specify specific business goals to obtain corresponding recommended product combinations, is more in line with actual use, and is conducive to promotion and application.
[0081] Based on the above-mentioned offline product recommendation method, an embodiment of the present invention also provides an online real-time product recommendation method, including: according to the received product recommendation request, obtaining the customer's product contract real-time information, business real-time information, and attribute information, and combining the product contract real-time information and the business real-time information, as well as the attribute information, into a combination feature of the customer; obtaining a product attribute relationship graph, the product attribute relationship graph having each product to be recommended and the relationship information between each product to be recommended, and each product to be recommended has a mutually exclusive relationship; combining each product to be recommended, and according to the relationship information, eliminating mutually exclusive products from the combined combination of products to be recommended, to generate a product combination set that does not have a mutually exclusive relationship; inputting the product combination set and the combination feature into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model.
[0082] Specifically, when a product recommendation request is received from an account manager, the real-time product contract information and business information of the customer are obtained based on the time of the product recommendation request, and the customer's attribute information is pulled, and the real-time product contract information and business information, as well as the attribute information, are combined into a combination feature of the customer. Then, a product combination set is constructed in a similar manner to the above-mentioned offline product recommendation method, and the product combination set and the combination features containing real-time information are input into the trained product recommendation model to obtain a real-time recommended product combination output by the product recommendation model.
[0083] Figure 6 It is a schematic diagram of the principle of the product recommendation method of an embodiment of the present invention. The customer's online real-time product contract information, business real-time information, and the product knowledge graph, customer portrait and other attribute information in the offline data warehouse are obtained, and the product contract real-time information and business real-time information, as well as the attribute information, are combined into the customer's combination characteristics. Based on the obtained product knowledge graph, a constraint node graph representing a set of product combinations that do not have mutually exclusive relationships is generated. The constraint node graph and the combination characteristics are input into the product recommendation model obtained through the establishment, model training and model evaluation of the product recommendation model to obtain the recommended product combination output by the product recommendation model and send it to the account manager. After receiving the recommended product combination for the customer, the account manager formulates a personalized visit plan to visit the customer to meet the customer's needs and achieve the purpose of recommending the product combination for the customer.
[0084] The embodiments of the present invention provide specific implementation methods for offline product recommendations and real-time product recommendations, which can meet the needs of users to the greatest extent and adapt to the needs of more business scenarios.
[0085] In addition, the embodiment of the present invention also experimentally verifies the product recommendation method. After offline training, the precision, recall, and F1 values of the product recommendation model are 0.78, 0.93, and 0.84, respectively. Through the propensity score matching algorithm, the differences between customers are eliminated, and the customer data is diverted. The customers are divided into an experimental group and a control group. The experimental group recommends products based on the product recommendation method of the embodiment of the present invention, and the control group is manually recommended by the customer manager based on historical experience. Finally, among the 900 customers in the experimental group, 43 customers signed the recommended products, and the signing rate was 4.8%; the control group was 33, and the signing rate was 3.7%. The experimental group increased by 28.6% compared with the control group. See Table 1 for details.
[0086] Table 1
[0087] Figure 7 FIG. 1 is a schematic diagram of the main modules of the product recommendation device according to an embodiment of the present invention. Figure 7 As shown, the product recommendation device 700 mainly includes a feature determination module 701 , a combination determination module 702 and a product recommendation module 703 .
[0088] The feature determination module 701 is used to obtain the product contract information, service information, and attribute information of the customer, and combine the product contract information, the service information, and the attribute information into a combined feature of the customer;
[0089] The combination determination module 702 is used to obtain a product attribute relationship diagram, which defines the relationship information between each product to be recommended and each product to be recommended, and there is a mutually exclusive relationship between each product to be recommended; combine each product to be recommended, and according to the relationship information, remove mutually exclusive products from the combined product combination to be recommended, so as to generate a product combination set without a mutually exclusive relationship;
[0090] The product recommendation module 703 is used to input the product combination set and the combination features into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model.
[0091] According to one embodiment of the present invention, the product attribute relationship graph exists in the form of a product knowledge graph, each node in the product knowledge graph represents a product, and the line between two nodes represents the relationship between the two products.
[0092] According to another embodiment of the present invention, the combination determination module 702 is further used for: reading all node information in the product knowledge graph to obtain each of the products to be recommended; combining each of the products to be recommended to obtain multiple product combinations, and combining each of the product combinations and each of the products to be recommended to form a product element set; for each component element in the product element set: searching the product knowledge graph for a node corresponding to each product in the component element, and judging whether the component element contains mutually exclusive products based on the connection relationship between the nodes; and deleting the component element if the component element contains mutually exclusive products; and combining the remaining components in the product element set to form a product combination set.
[0093] According to another embodiment of the present invention, the combination determination module 702 is further used for: reading all node information in the product knowledge graph to obtain each of the products to be recommended; combining each of the products to be recommended to obtain multiple product combinations, and combining each of the product combinations and each of the products to be recommended to form a product element set; for each of the products to be recommended: searching and identifying the excluded products corresponding to the product to be recommended from the product knowledge graph; and combining the product to be recommended and the excluded products into mutually exclusive products; deleting the constituent elements in the product element set that contain the mutually exclusive products to obtain a product combination set.
[0094] According to another embodiment of the present invention, the product element set exists in the form of a node graph, and the node graph is constructed by taking each of the product combinations and each of the products to be recommended in the product element set as graph nodes, thereby obtaining a constraint node graph corresponding to the product combination set; the product recommendation module 703 is further used to: input the constraint node graph and the combination features into a pre-trained product recommendation model constructed by a graph convolutional neural network model; through each convolution layer in the product recommendation model, combined with the combination features, to screen each level of adjacent nodes in the constraint node graph to obtain recommended nodes; and parse the recommended nodes to obtain corresponding recommended product combinations.
[0095] According to another embodiment of the present invention, the product recommendation model is trained in the following manner: training customers are determined from the customer base, and business information and attribute information of the training customers are retrieved according to the time granularity to form training features of the training customers at each time granularity; a product combination training set corresponding to the training customers is generated according to the product attribute relationship graph; product contract signing information of the training customers at each time granularity is used as a sample label; training samples are constructed according to the training features, the product combination training set and the sample labels; and the training samples are used to train the product recommendation model by means of a loss function and gradient descent to obtain a trained product recommendation model.
[0096] According to another embodiment of the present invention, the product recommendation device 700 also includes a multi-objective recommendation module (not shown in the figure), which is used to: configure multiple business objectives for recommending products for the customer; input each of the business objectives together with the product combination set and the combination features into a pre-trained product recommendation model to obtain the recommended product combination for the customer under each of the business objectives output by the product recommendation model; based on the received recommendation result query request with business objectives, query the recommended product combinations for the customer under each of the business objectives for product combinations that are consistent with the business objectives in the recommendation result query request, and return the queried product combination to the requester of the recommendation result query request.
[0097] Figure 8 is an exemplary system architecture diagram to which embodiments of the present invention can be applied.
[0098] like Figure 8 As shown, system architecture 800 may include terminal devices 801, 802, 803, a network 804, and a server 805. Network 804 is used to provide a medium for communication links between terminal devices 801, 802, 803 and server 805. Network 804 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.
[0099] Users can use terminal devices 801, 802, 803 to interact with server 805 through network 804 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 801, 802, 803, such as product recommendation applications, etc. (only as an example).
[0100] The terminal devices 801 , 802 , and 803 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.
[0101] Server 805 may be a server that provides various services, such as a background management server that provides support for product recommendations made by users using terminal devices 801, 802, and 803 (for example only). The background management server may obtain the customer's product contract information, business information, and attribute information, and combine the product contract information, the business information, and the attribute information into a combination feature of the customer; obtain a product attribute relationship graph, which defines the relationship information between each product to be recommended and each product to be recommended, and there is a mutually exclusive relationship between each product to be recommended; combine each product to be recommended, and according to the relationship information, remove mutually exclusive products from the combined product combination to be recommended, and generate a product combination set that does not have a mutually exclusive relationship; input the product combination set and the combination feature into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model, and feed back the processing results (such as a recommended product combination, etc. - for example only) to the terminal device.
[0102] It should be noted that the product recommendation method provided in the embodiment of the present invention is generally executed by the server 805 , and accordingly, the product recommendation device is generally arranged in the server 805 .
[0103] It should be understood that Figure 8 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.
[0104] Reference below Fig. 9 , which shows a schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing an embodiment of the present invention. Fig. 9 The terminal device or server shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.
[0105] like Fig. 9 As shown, the computer system 900 includes a central processing unit (CPU) 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage part 908 into a random access memory (RAM) 903. In the RAM 903, various programs and data required for the operation of the system 900 are also stored. The CPU 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.
[0106] The following components are connected to the I / O interface 905: an input section 906 including a keyboard, a mouse, etc.; an output section 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, a modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the I / O interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 910 as needed, so that a computer program read therefrom is installed into the storage section 908 as needed.
[0107] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 909, and / or installed from the removable medium 911. When the computer program is executed by the central processing unit (CPU) 901, the above-mentioned functions defined in the system of the present invention are executed.
[0108] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination thereof.
[0109] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a sequence different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0110] The units involved in the embodiments of the present invention may be implemented by software or hardware. The units described may also be arranged in a processor, for example, it may be described as: a processor including: a feature determination module, a combination determination module and a product recommendation module.
[0111] Among them, the names of these modules do not constitute a limitation of the modules themselves in certain cases. For example, the product recommendation module can also be described as "a module used to input the product combination set and the combination features into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model."
[0112] On the other hand, the present invention also provides a computer-readable medium, which may be included in the device described in the embodiment; or it may exist independently and not be assembled into the device. The computer-readable medium carries one or more programs. When the one or more programs are executed by a device, the device includes: obtaining the customer's product contract information, business information, and attribute information, and combining the product contract information, the business information, and the attribute information into a combination feature of the customer; obtaining a product attribute relationship graph, the product attribute relationship graph defines each product to be recommended and the relationship information between each product to be recommended, and each product to be recommended has a mutually exclusive relationship; combining each product to be recommended, and according to the relationship information, removing mutually exclusive products from the combined product combination to be recommended, generating a product combination set that does not have a mutually exclusive relationship; inputting the product combination set and the combination feature into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model.
[0113] The technical solution according to the embodiment of the present invention has the following advantages or beneficial effects: by obtaining the customer's product contract information, business information, and attribute information, and combining the product contract information, business information, and attribute information into the customer's combined features; obtaining a product attribute relationship diagram, the product attribute relationship diagram defines each product to be recommended and the relationship information between each product to be recommended, and there is a mutually exclusive relationship between each product to be recommended; combining each product to be recommended, and according to the relationship information, eliminating mutually exclusive products from the combined product combination to be recommended, and generating a product combination set that does not have a mutually exclusive relationship; inputting the product combination set and the combination features into a pre-trained product recommendation model to obtain a technical solution for the recommended product combination output by the product recommendation model, thereby achieving efficient and accurate product recommendations, which not only reduces the workload of personnel, but also makes product recommendations based on the combination features that affect the customer's willingness to sign a contract and the product combination set that eliminates mutually exclusive products, which not only ensures the accuracy of the recommended products, but also avoids the existence of mutually exclusive products in the recommended products, thereby improving the professionalism of the product recommendations.
[0114] The specific implementation methods described herein do not constitute limitations on the scope of protection of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.< / c> < / n>
Claims
1. A product recommendation method, It is characterized in that include: Acquire the customer's product contract information, business information, and attribute information, and combine the product contract information, the business information, and the attribute information into a combined feature of the customer; Obtain a product attribute relationship graph, wherein the product attribute relationship graph defines each product to be recommended and relationship information between each of the products to be recommended, and each of the products to be recommended has a mutually exclusive relationship; combine each of the products to be recommended, and according to the relationship information, remove mutually exclusive products from the combined combination of products to be recommended, to generate a product combination set without a mutually exclusive relationship; The product combination set and the combination features are input into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model.
2. The method according to claim 1, It is characterized in that The product attribute relationship graph exists in the form of a product knowledge graph. Each node in the product knowledge graph represents a product, and the line between two nodes represents the relationship between the two products.
3. The method according to claim 2, It is characterized in that Combining the products to be recommended, and removing mutually exclusive products from the combined product combinations to be recommended according to the relationship information, to generate a product combination set without mutually exclusive relationships, including: Obtaining each of the products to be recommended according to the information of the nodes in the product knowledge graph; Combining the products to be recommended to obtain a plurality of product combinations, and combining the product combinations and the products to be recommended to form a product element set; For each component element in the product element set: searching the product knowledge graph for a node corresponding to each product in the component element, and judging whether the component element contains mutually exclusive products according to the connection relationship between the nodes; and deleting the component element if the component element contains mutually exclusive products; The remaining elements in the product element set are combined into a product combination set.
4. The method according to claim 2, It is characterized in that Combining the products to be recommended, and removing mutually exclusive products from the combined product combinations to be recommended according to the relationship information, to generate a product combination set without mutually exclusive relationships, including: Obtaining each of the products to be recommended according to the information of the nodes in the product knowledge graph; Combining the products to be recommended to obtain a plurality of product combinations, and combining the product combinations and the products to be recommended to form a product element set; For each of the products to be recommended: searching and identifying the excluded products corresponding to the product to be recommended from the product knowledge graph; and forming the product to be recommended and the excluded products into mutually exclusive products; The constituent elements including the mutually exclusive products in the product element set are deleted to obtain a product combination set.
5. The method according to claim 3 or 4, It is characterized in that The product element set exists in the form of a node graph, and the node graph is constructed by taking each of the product combinations and each of the products to be recommended in the product element set as graph nodes, thereby obtaining a constraint node graph corresponding to the product combination set; Inputting the product combination set and the combination features into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model, including: Inputting the constraint node graph and the combined features into a pre-trained product recommendation model constructed by a graph convolutional neural network model; By combining the combined features with each convolutional layer in the product recommendation model, each level of adjacent nodes in the constraint node graph is screened to obtain recommended nodes; The recommendation node is parsed to obtain a corresponding recommended product combination.
6. The method according to claim 1, It is characterized in that The product recommendation model is trained in the following way: Determine a training customer from the customer group, retrieve the business information and attribute information of the training customer according to the time granularity, and form the training features of the training customer at each time granularity; Generating a product combination training set corresponding to the training customer according to the product attribute relationship diagram; Using the product contract signing information of the training customers at each of the time granularities as sample labels; Constructing a training sample according to the training features, the product combination training set and the sample label; The product recommendation model is trained using the training samples through a loss function and gradient descent to obtain a trained product recommendation model.
7. The method according to claim 1, It is characterized in that The method further comprises: configuring a plurality of business objectives of recommending products for said customer; Inputting each of the business objectives together with the product combination set and the combination features into a pre-trained product recommendation model, and obtaining a recommended product combination for the customer under each of the business objectives output by the product recommendation model; According to the received recommendation result query request with business objectives, the product combination consistent with the business objectives in the recommendation result query request is queried from the recommended product combinations under each of the business objectives of the customer, and the queried product combination is returned to the requester of the recommendation result query request.
8. A product recommendation method, It is characterized in that include: According to the received product recommendation request, obtain the customer's product contract real-time information, business real-time information, and attribute information, and combine the product contract real-time information, the business real-time information, and the attribute information into a combined feature of the customer; Acquire a product attribute relationship graph, wherein the product attribute relationship graph has each product to be recommended and relationship information between each product to be recommended, and each product to be recommended has a mutually exclusive relationship; combine each product to be recommended, and according to the relationship information, remove mutually exclusive products from the combined combination of products to be recommended, so as to generate a product combination set without a mutually exclusive relationship; The product combination set and the combination features are input into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model.
9. A product recommendation device, It is characterized in that include: A feature determination module, used to obtain the customer's product contract information, business information, and attribute information, and combine the product contract information, the business information, and the attribute information into a combined feature of the customer; A combination determination module is used to obtain a product attribute relationship diagram, wherein the product attribute relationship diagram defines each product to be recommended and the relationship information between each product to be recommended, and each product to be recommended has a mutually exclusive relationship; each product to be recommended is combined, and according to the relationship information, mutually exclusive products are removed from the combined product combination to be recommended, so as to generate a product combination set without a mutually exclusive relationship; The product recommendation module is used to input the product combination set and the combination features into a pre-trained product recommendation model to obtain a recommended product combination output by the product recommendation model.
10. A mobile electronic device terminal, It is characterized in that include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.
11. A computer readable medium having a computer program stored thereon, It is characterized in that When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.
12. A computer program product comprising a computer program, It is characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.