Product recommendation method, product recommendation model training method, device and equipment

Abstract

The present disclosure provides a product recommendation method, a product recommendation model training method, an apparatus, and a device, which can be applied to the fields of artificial intelligence technology and financial technology. The method comprises: with the permission of the target object, converting the object information of the target object into a graph vector, wherein the object information includes a plurality of sub-object information, and the graph vector includes a plurality of node vectors, and the node vector is used to represent the object sub-information; when the target object is determined to be a potential object based on the graph vector, a first sub-graph vector that matches the preset product attribute information is determined from the graph vector based on the node identifiers of the plurality of node vectors; the first sub-graph vector is input into the product recommendation model to obtain the object category of the target object and a product recommendation list that matches both the object category and the target object; and the product to be recommended determined from the product recommendation list is recommended to the target object.

Description

Technical Field

[0001] The present disclosure relates to the fields of artificial intelligence technology and financial technology, and specifically to a product recommendation method, a product recommendation model training method, an apparatus, a device, a medium, and a program product. Background Art

[0002] With the rapid development of financial technology, customers are faced with more and more product choices. Using product recommendation methods to recommend products to customers is one of the current industry service trends. Currently, common product recommendation methods usually recommend products based on users' historical purchase history, and when making recommendations, the recommended products may be determined for all customers.

[0003] In the process of realizing the concept of the present disclosure, the inventors found that there are at least the following problems in the related art: when making product recommendations, the recommended objects lack specificity, and the method of using users' historical purchase history to make recommendations may be affected by data sparsity and "cold start" problems. For new users or new products, due to the lack of sufficient data, there may be problems with low recommendation accuracy and failure to meet customer needs. Summary of the Invention

[0004] In view of the above problems, the present disclosure provides a product recommendation method, a product recommendation model training method, an apparatus, a device, a medium and a program product.

[0005] One aspect of the present disclosure provides a product recommendation method, comprising: with permission from a target object, converting object information of the target object into a graph vector, wherein the object information includes multiple sub-object information, the graph vector includes multiple node vectors, and the node vector is used to represent the object sub-information; when the target object is determined to be a potential object based on the graph vector, determining a first sub-graph vector that matches preset product attribute information from the graph vector based on node identifiers of multiple node vectors; inputting the first sub-graph vector into a product recommendation model to obtain an object category of the target object and a product recommendation list that matches both the object category and the target object; and recommending the to-be-recommended product determined from the product recommendation list to the target object.

[0006] According to an embodiment of the present disclosure, a product recommendation model includes a category determination sub-model and a product recommendation sub-model; inputting a first sub-graph vector into the product recommendation model to obtain an object category of a target object and a product recommendation list that matches both the object category and the target object, including: inputting the first sub-graph vector into the category determination sub-model to obtain an object category of the target object; inputting preferred products of similar objects with the object category and the first sub-graph vector into the product recommendation sub-model to obtain a product recommendation list that matches both the object category and the target object.

[0007] According to an embodiment of the present disclosure, the preferred products and the first sub-graph vector of similar objects with an object category are input into a product recommendation sub-model to obtain a product recommendation list that matches both the object category and the target object, including: inputting the preferred products and the first sub-graph vector of the similar objects into the product recommendation sub-model to obtain the first preferred product feature of the similar objects and the second preferred product feature of the target object; performing similarity calculation on the first preferred product feature of each of a plurality of similar objects and the product feature of each of a plurality of preset products to obtain first similarities corresponding to the plurality of preset products respectively; performing similarity calculation on the second preferred product feature and the product feature of each of a plurality of preset products to obtain second similarities corresponding to the plurality of preset products respectively; determining the recommendation value of each of the plurality of preset products based on the first weight value corresponding to the first preference product feature, the first similarities corresponding to the plurality of preset products respectively, the second weight value corresponding to the second preference product feature and the second similarities corresponding to the plurality of preset products respectively; and obtaining a product recommendation list based on the recommendation value of each of the plurality of preset products.

[0008] According to an embodiment of the present disclosure, the product recommendation method also includes: determining a second sub-graph vector that matches preset object attribute information from the graph vector based on node identifiers of multiple node vectors; inputting the second sub-graph vector into a potential object prediction model to obtain a target probability that the target object is a potential object; and determining that the target object is a potential object when it is determined that the target probability is greater than or equal to a preset probability threshold.

[0009] According to an embodiment of the present disclosure, object information also includes: the association relationship between multiple sub-object information; converting the object information of the target object into a graph vector, including: performing information correction processing on the object information to obtain corrected object information; extracting multiple sub-object information and mapping the association relationship between multiple sub-object information on the corrected object information to obtain a knowledge graph corresponding to the target object, wherein the knowledge graph includes nodes and edges, the nodes represent the sub-object information, and the edges represent the association relationship between multiple sub-object information; mapping the knowledge graph into a vector space to obtain a graph vector.

[0010] According to an embodiment of the present disclosure, object information is corrected to obtain corrected object information, including: determining target sub-object information of a numerical type from multiple sub-object information; when it is determined that there is an abnormality in the information format of the target object information, correcting the target object information based on adjacent object information adjacent to the target object information to obtain first target sub-object information; when it is determined that there is an abnormality in the information value of the target object information, determining homogeneous object information having the same information identifier as the target object information; determining a replacement interval based on the information value of the homogeneous object information; correcting the information value of the target object information to a correction value in the replacement interval to obtain second target sub-object information; and determining the corrected object information based on the first target sub-object information and the second target sub-object information.

[0011] Another aspect of the present disclosure provides a product recommendation model training method, which includes a category determination submodel and a product recommendation submodel; including: obtaining a first subgraph vector sample of each of multiple potential object samples; for each potential object sample, determining multiple category feature values ​​of the potential object sample based on the node identifiers and node values ​​of multiple node vector samples included in the first subgraph vector sample; determining the object category label of the potential object sample based on the multiple category feature values; training an initial category determination submodel based on the first subgraph vector samples of each of the multiple potential object samples and the object category labels of the multiple potential object samples to obtain a category determination submodel; training an initial product recommendation submodel based on the first subgraph vector samples of each of the multiple potential object samples and the preferred product samples of the same type of object samples of the multiple potential object samples to obtain a product recommendation submodel, wherein the object category of the same type of object sample is the same as the object category label of the potential object sample; obtaining a product recommendation model based on the category determination submodel and the product recommendation submodel.

[0012] Another aspect of the present disclosure provides a product recommendation device, including: a conversion module, for converting the object information of the target object into a graph vector with the permission of the target object, wherein the object information includes multiple sub-object information, the graph vector includes multiple node vectors, and the node vector is used to represent the object sub-information; a first determination module, for determining a first sub-graph vector that matches preset product attribute information from the graph vector based on the node identifiers of the multiple node vectors when the target object is determined to be a potential object based on the graph vector; a second determination module, for inputting the first sub-graph vector into a product recommendation model to obtain the object category of the target object and a product recommendation list that matches both the object category and the target object; a recommendation module, for recommending the to-be-recommended product determined from the product recommendation list to the target object.

[0013] Another aspect of the present disclosure provides a product recommendation model training device, wherein the product recommendation model includes a category determination submodel and a product recommendation submodel; the device includes: an acquisition module for acquiring a first subgraph vector sample of each of a plurality of potential object samples; a third determination module for determining, for each potential object sample, a plurality of category feature values ​​of the potential object sample based on the node identifiers and node values ​​of a plurality of node vector samples included in the first subgraph vector sample; a fourth determination module for determining an object category label of the potential object sample based on the plurality of category feature values; a first training module for training an initial category determination submodel based on the first subgraph vector samples of each of the plurality of potential object samples and the object category labels of the plurality of potential object samples to obtain a category determination submodel; a second training module for training an initial product recommendation submodel based on the first subgraph vector samples of each of the plurality of potential object samples and the preferred product samples of the same type of object samples of the plurality of potential object samples to obtain a product recommendation submodel, wherein the object category of the same type of object sample is the same as the object category label of the potential object sample; and a model determination module for obtaining a product recommendation model based on the category determination submodel and the product recommendation submodel.

[0014] Another aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0015] Another aspect of the present disclosure further provides a computer-readable storage medium having a computer program or instructions stored thereon, which implements the steps of the above method when the computer program or instructions are executed by a processor.

[0016] The fifth aspect of the present disclosure further provides a computer program product, comprising a computer program or instructions, which implement the steps of the above method when executed by a processor.

[0017] According to the product recommendation method disclosed herein, by converting object information into a graph vector and determining whether the target object is a potential object through the graph vector, when the target object is determined to be a potential object, the first sub-graph vector for product recommendation is filtered from the graph vector, thereby achieving effective screening of input data and recommended objects. The first sub-graph vector is then input into the product recommendation model to obtain the object category to which the target object belongs and a list of product recommendations that match both the object category and the target object. This method determines products that may be of interest to the target object based on both the preferences of similar objects and the preferences of the target object itself, thereby at least partially resolving the technical problems of lack of pertinence and low accuracy of product recommendations in related technologies, and achieving technical effects such as improving the accuracy of subsequent product recommendations and reducing waste of recommendation resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0019] Figure 1 Schematically illustrates an application scenario diagram of a product recommendation method, a product recommendation model training method, an apparatus, a device, a medium, and a program product according to an embodiment of the present disclosure;

[0020] Figure 2 The following schematically shows a flow chart of a product recommendation method according to an embodiment of the present disclosure;

[0021] Figure 3 A schematic diagram of a knowledge graph according to an embodiment of the present disclosure is schematically shown;

[0022] Figure 4 The flowchart of the product recommendation model training method according to an embodiment of the present disclosure is schematically shown;

[0023] Figure 5 A schematic diagram of category feature values ​​for different potential object samples according to an embodiment of the present disclosure is schematically shown;

[0024] Figure 6 Schematically shows a data flow diagram of a product recommendation model training method according to an embodiment of the present disclosure;

[0025] Figure 7 Schematically shows a structural block diagram of a product recommendation device according to an embodiment of the present disclosure;

[0026] Figure 8 A structural block diagram of a product recommendation model training device according to an embodiment of the present disclosure is schematically shown; and

[0027] Figure 9A block diagram of an electronic device suitable for implementing a product recommendation method and a product recommendation model training method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0028] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0029] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of the features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0030] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0031] When expressions such as "at least one of A, B, and C, etc." are used, they should generally be interpreted in accordance with the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0032] It should be noted that the product recommendation method and product recommendation model training method disclosed herein can be used in the fields of artificial intelligence technology and financial technology technology, and can also be used in any field other than the fields of artificial intelligence technology and financial technology technology, such as the field of computer technology. The application fields of the product recommendation method and product recommendation model training method disclosed herein are not limited.

[0033] In the technical solutions disclosed herein, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0034] In scenarios where personal information is used for automated decision-making, the methods, devices, and systems provided by the embodiments of the present disclosure all provide users with corresponding operation portals for them to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered. The expression "automated decision-making" here refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests and hobbies, or economic, health, credit status, etc. through computer programs and making decisions. The expression "expert decision-making" here refers to the activity of making decisions by people who specialize in a certain field, have specialized experience, knowledge, and skills, and have reached a certain level of professionalism.

[0035] During their research, the inventors discovered that product recommendation methods in related technologies may be affected by data sparsity and the "cold start" problem. For new users or new products, recognition results may be unstable or inaccurate due to a lack of sufficient data. Furthermore, these methods may struggle to address complex issues, such as balancing users' long-term and short-term interests and handling inconsistencies between implicit and explicit feedback.

[0036] Furthermore, the customer data collected by these technologies is mostly static, failing to deeply analyze customers' dynamic financial needs and behavioral patterns. Furthermore, a lack of in-depth research into key information such as potential customer value predictions and risk preferences hinders the accurate identification of potential high-end customers.

[0037] An embodiment of the present disclosure provides a product recommendation method, comprising: with permission from a target object, converting object information of a target object into a graph vector, wherein the object information includes multiple sub-object information, the graph vector includes multiple node vectors, and the node vector is used to represent the object sub-information; in the case where the target object is determined to be a potential object based on the graph vector, determining a first sub-graph vector that matches preset product attribute information from the graph vector based on node identifiers of multiple node vectors; inputting the first sub-graph vector into a product recommendation model to obtain an object category of the target object and a product recommendation list that matches both the object category and the target object; and recommending the products to be recommended determined from the product recommendation list to the target object.

[0038] Figure 1 The application scenario diagram of the product recommendation method, product recommendation model training method, device, equipment, medium and program product according to the embodiments of the present disclosure is schematically shown.

[0039] like Figure 1 As shown, the application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is used as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or optical fiber cables.

[0040] A user may use a first terminal device 101, a second terminal device 102, or a third terminal device 103 to interact with a server 105 via a network 104 to receive or send messages, etc. Various communication client applications may be installed on the first terminal device 101, the second terminal device 102, or the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (for example only).

[0041] The first terminal device 101 , the second terminal device 102 , and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0042] The server 105 may be a server that provides various services, such as a background management server (for example only) that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process received data such as user requests, and feed back processing results (e.g., web pages, information, or data obtained or generated based on user requests) to the terminal devices.

[0043] It should be noted that the product recommendation method and product recommendation model training method provided in the embodiments of the present disclosure can generally be executed by the server 105. Accordingly, the product recommendation device and product recommendation model training device provided in the embodiments of the present disclosure can generally be set in the server 105. The product recommendation method and product recommendation model training method provided in the embodiments of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the product recommendation device and product recommendation model training device provided in the embodiments of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0044] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0045] The following will be based on Figure 1 The scene described by Figures 2 to 6 The product recommendation method and product recommendation model training method of the disclosed embodiment are described in detail.

[0046] Figure 2 The flowchart of the product recommendation method according to an embodiment of the present disclosure is schematically shown.

[0047] like Figure 2 As shown, the method includes operations S210 to S240.

[0048] In operation S210, with permission from the target object, object information of the target object is converted into a graph vector, wherein the object information includes a plurality of sub-object information, and the graph vector includes a plurality of node vectors, and the node vector is used to represent the object sub-information.

[0049] In operation S220, when the target object is determined to be a potential object based on the graph vector, a first sub-graph vector matching the preset product attribute information is determined from the graph vector based on the node identifiers of the plurality of node vectors.

[0050] In operation S230 , the first sub-graph vector is input into a product recommendation model to obtain an object category of the target object and a product recommendation list that matches both the object category and the target object.

[0051] In operation S240 , the to-be-recommended product determined from the product recommendation list is recommended to the target object.

[0052] According to an embodiment of the present disclosure, the target object may be a customer who is likely to purchase the product.

[0053] According to the embodiments of the present disclosure, the object information of the target object is not limited and can be any information of the target object. For example, the object information can include basic object information, object transaction data, object interaction information, object social relationship information, etc. Sub-object information can be specific information among the above information, such as the products traded in the object transaction data, where each product can be a sub-object information.

[0054] According to an embodiment of the present disclosure, a node vector may correspond to sub-object information in a one-to-one manner; a node identifier may correspond to an information identifier of the sub-object information.

[0055] According to an embodiment of the present disclosure, the basic information of the object may include: the name, age, occupation, income, products held and other information of the target object; the object transaction data may include: the target object's deposits, loans, investments, transfers and other information; the object interaction information may include: the target object's online consultation, complaints, feedback and other information; the object social relationship information may include: the target object's unit information, unit equity information, unit executive information, information on guarantees or guarantees provided for the unit or individual, unit salary payment information, corporate WeChat clicks, browsing, purchases, recommendations, chat information, etc.

[0056] According to an embodiment of the present disclosure, before obtaining the object information of the target object, the consent or authorization of the target object may be obtained. For example, before obtaining the object information, a request to obtain the object information may be issued to the target object. If the target object agrees or authorizes the object information to be obtained, the object information of the target object is obtained.

[0057] According to embodiments of the present disclosure, a corresponding operation entry can be provided for the target subject, allowing the target subject to choose whether to agree or reject the automated decision result. That is, before processing / decision-making on the subject's information, an instruction to agree or reject the processing / decision-making process can be obtained from the target subject through the corresponding operation entry. If the user agrees to the processing / decision-making process, the subject information is processed / decided, i.e., step S210 is executed. If the user rejects the processing / decision-making process, the expert decision-making process is entered.

[0058] According to an embodiment of the present disclosure, a graph vector may be obtained by converting a knowledge graph obtained from object information. The graph vector may include multiple node vectors and a vector representing the association relationship between the multiple node vectors.

[0059] According to an embodiment of the present disclosure, the preset product attribute information may be information that affects the products purchased by the target object. Specifically, the preset product attribute information may include: natural attribute information, financial status information, credit information, product information, customer relationship information, transaction information, etc.

[0060] According to embodiments of the present disclosure, financial status information may include information such as income, expenditure, deposits, liabilities, and financial investments. Credit information may include information such as credit rating, overdue payments, and on-time payment rates. Product information may include information such as purchase time, amount, frequency, product type, and payment method. Customer relationship information may include information such as the frequency of mobile banking, telephone consultations, and branch usage, complaints, and customer satisfaction surveys. Transaction information may include information about payment applications.

[0061] According to an embodiment of the present disclosure, potential targets may be customer groups that have outstanding performance in terms of asset size, transaction activity, and contribution, or have potential in the above aspects.

[0062] According to the embodiments of the present disclosure, there is no limitation on the object category, and it can be determined specifically according to the actual classification situation. It can be a more fine-grained classification of the target object through the first sub-graph vector of the target object.

[0063] According to an embodiment of the present disclosure, multiple node identifiers of a graph vector may be matched with preset product attribute information, thereby determining a first sub-graph vector that matches the preset product attribute information.

[0064] According to the embodiments of the present disclosure, there is no limitation on the implementation method of the product recommendation model, which can be implemented through a multi-level neural network or a multi-level deep learning algorithm.

[0065] According to an embodiment of the present disclosure, the product recommendation list may be a product list obtained by considering two factors: the object category and the first sub-graph vector of the target object.

[0066] According to an embodiment of the present disclosure, the product recommendation list may include recommendation values ​​for each of multiple preset products. The multiple preset products can be sorted by the recommendation values, so that the product to be recommended can be determined from the multiple preset products and recommended to the target object.

[0067] According to the product recommendation method disclosed herein, by converting object information into a graph vector and determining whether the target object is a potential object through the graph vector, when the target object is determined to be a potential object, the first sub-graph vector for product recommendation is filtered from the graph vector, thereby achieving effective screening of input data and recommended objects. The first sub-graph vector is then input into the product recommendation model to obtain the object category to which the target object belongs and a list of product recommendations that match both the object category and the target object. This method determines products that may be of interest to the target object based on both the preferences of similar objects and the preferences of the target object itself, thereby at least partially resolving the technical problems of lack of pertinence and low accuracy of product recommendations in related technologies, and achieving technical effects such as improving the accuracy of subsequent product recommendations and reducing waste of recommendation resources.

[0068] According to an embodiment of the present disclosure, the product recommendation model includes a category determination sub-model and a product recommendation sub-model; inputting the first sub-graph vector into the product recommendation model to obtain the object category of the target object and a product recommendation list that matches both the object category and the target object may include the following operations.

[0069] The first sub-graph vector is input into the category determination sub-model to obtain the object category of the target object; the preferred products of similar objects with the object category and the first sub-graph vector are input into the product recommendation sub-model to obtain a product recommendation list that matches both the object category and the target object.

[0070] According to the embodiments of the present disclosure, there is no limitation on the implementation methods of the category determination sub-model and the product recommendation sub-model, and they can be implemented through neural network models or deep learning algorithms respectively, for example, they can be implemented through restricted Boltzmann machines (RBM) or recurrent neural networks, etc.

[0071] According to the embodiments of the present disclosure, RBM can effectively learn and simulate the intrinsic structure of data, thereby predicting future customer behavior trends and helping to formulate strategies in advance to better meet the needs of potential targets.

[0072] According to an embodiment of the present disclosure, the category determination submodel can be trained in the following manner: obtaining a first sub-graph vector sample of each of multiple potential object samples; for each potential object sample, determining multiple category feature values ​​of the potential object sample based on the node identifiers and node values ​​of multiple node vector samples included in the first sub-graph vector sample; determining the object category label of the potential object sample based on the multiple category feature values; training the initial category determination submodel based on the first sub-graph vector samples of each of the multiple potential object samples and the object category labels of the multiple potential object samples to obtain the category determination submodel.

[0073] According to an embodiment of the present disclosure, similar objects may be objects having the same object category as the target object, and preferred products of similar objects may be pre-stored in a database.

[0074] According to an embodiment of the present disclosure, the preferred products of similar objects with an object category and the first sub-graph vector are input into a product recommendation sub-model to obtain a product recommendation list that matches both the object category and the target object. In some embodiments, the preferred products of the target object can be obtained by inputting the first sub-graph vector into the product recommendation sub-model, and the preferred products of the target object and the preferred products of similar objects are intersected, and the preferred products preferred by both the target object and the similar objects are given a higher score value, and other preferred products are given a lower score value, thereby obtaining a product recommendation list.

[0075] According to an embodiment of the present disclosure, the object category of the target object is determined by a category determination sub-model, and a product recommendation list that matches both the object category and the target object is jointly determined by similar objects with the same object category and the first sub-graph vector of the target object itself, thereby determining products that may be of interest to the target object from multiple angles based on the product preferences of objects of the same category and the product preferences of the target object.

[0076] According to an embodiment of the present disclosure, inputting preferred products of similar objects with an object category and a first sub-graph vector into a product recommendation sub-model to obtain a product recommendation list that matches both the object category and the target object may include the following operations.

[0077] The preferred products and the first sub-graph vector of the same type of objects are input into the product recommendation sub-model to obtain the first preferred product features of the same type of objects and the second preferred product features of the target object; the similarity between the first preferred product features of multiple similar objects and the product features of multiple preset products are calculated to obtain the first similarities corresponding to the multiple preset products respectively; the similarity between the second preferred product features and the product features of multiple preset products are calculated to obtain the second similarities corresponding to the multiple preset products respectively; based on the first weight value corresponding to the first preferred product feature, the first similarities corresponding to the multiple preset products respectively, the second weight value corresponding to the second preferred product feature and the second similarities corresponding to the multiple preset products respectively, the recommendation value of each of the multiple preset products is determined; based on the recommendation values ​​of each of the multiple preset products, a product recommendation list is obtained.

[0078] According to an embodiment of the present disclosure, in one implementation, the product recommendation sub-model can be used to perform feature extraction on the preferred products and the first sub-graph vector of the same type of objects, thereby obtaining the first preferred product feature and the second preferred product feature.

[0079] According to the embodiments of the present disclosure, there is no limitation on the method for calculating similarity, which may be cosine similarity, Euclidean distance, etc.

[0080] According to an embodiment of the present disclosure, the product features of a preset product may be obtained by inputting the preset product into a product recommendation sub-model, and may be obtained in advance and stored in a database.

[0081] According to an embodiment of the present disclosure, the first weight value and the second weight value may be preset, and the second weight value may be greater than the first weight value. By assigning a higher weight to the second weight value of the target object, it is more likely that the needs of the target object will be met.

[0082] According to an embodiment of the present disclosure, the first weight value and the first similarities corresponding to multiple preset products can be multiplied respectively to obtain the first initial recommendation values ​​of the multiple preset products; the second weight value and the second similarities corresponding to the multiple preset products can be multiplied respectively to obtain the second initial recommendation values ​​of the multiple preset products; for each preset product, the first initial recommendation value and the second initial recommendation value are added to obtain the recommendation value of each of the multiple preset products, and the recommendation values ​​of each of the multiple preset products are sorted to obtain a product recommendation list.

[0083] According to an embodiment of the present disclosure, in some embodiments, a product recommendation list for different target objects may be as shown in Table 1 below.

[0084] Table 1

[0085]

[0086] According to an embodiment of the present disclosure, by respectively determining the first preferred product features of similar objects and the second preferred product features of the target object, the similarities corresponding to multiple preset products are determined, and combined with the preset first weight value and the second weight value, a recommendation value for each product is obtained, thereby realizing product recommendations based on the quantified recommendation values, and facilitating more effective product recommendations.

[0087] According to an embodiment of the present disclosure, the product recommendation method may further include the following operations.

[0088] Based on the node identifiers of multiple node vectors, a second sub-graph vector that matches the preset object attribute information is determined from the graph vector; the second sub-graph vector is input into the potential object prediction model to obtain the target probability that the target object is a potential object; when it is determined that the target probability is greater than or equal to a preset probability threshold, the target object is determined to be a potential object.

[0089] According to an embodiment of the present disclosure, the preset object attribute information may be information that has an impact on determining whether the target object is a potential object, wherein the preset object attribute information may include: asset information, transaction frequency, credit score, product holdings, income level, natural attribute information and other information.

[0090] According to the embodiments of the present disclosure, the object contribution of the target object, the object's stickiness relative to the enterprise, the object's transaction activity and other information can be analyzed by presetting the object attribute information, and the analysis results can also be used as one of the inputs of the potential object prediction model. That is, in some embodiments, the second sub-graph vector includes not only vectors related to the preset object attribute information, but also vectors related to the object contribution information, the object's stickiness relative to the enterprise information, etc.

[0091] According to embodiments of the present disclosure, asset information may include deposits, investment accounts, stocks, bonds, and more. Transaction frequency may include withdrawals, transfers, and investment transactions. High-frequency transactions indicate a customer's higher demand for and engagement in financial services. A credit score may be calculated based on a customer's credit history and repayment record, used to assess their credit risk and repayment ability. A high credit score is generally associated with a lower risk of default and is a key characteristic of potential customers. Product holdings may include the number and variety of products a customer purchases and uses at a business, such as savings accounts, credit cards, and loan and wealth management products. A diverse product holding indicates a higher customer loyalty and trust in the business. Income level may include a customer's annual or monthly income, which is an important indicator for assessing their financial strength and spending power. High-income customers have greater financial resources and spending potential. Natural attribute information may include age, occupation, nature of work unit, professional background, and educational background, and can be used to identify financial needs and preferences.

[0092] According to embodiments of the present disclosure, customer contribution information can be assessed by calculating the net contribution a customer brings to the bank, such as profit and revenue. Customer stickiness information relative to the enterprise can be evaluated based on factors such as the number of years the customer has been a customer, whether the customer recommends new customers, and whether the customer uses the bank's value-added services. Customers with high stickiness are more likely to become potential customers and maintain a stable partnership.

[0093] According to the embodiments of the present disclosure, there is no limitation on the implementation method of the potential object prediction model, which can be a neural network model, such as RBM, convolutional neural network, etc.

[0094] According to an embodiment of the present disclosure, by matching the preset object attribute information with a plurality of node identifiers, a second sub-graph vector matching the preset object attribute information can be extracted from the graph vector.

[0095] According to the embodiment of the present disclosure, there is no limitation on the preset probability threshold, which can be determined according to actual conditions, for example, it can be 0.6.

[0096] According to an embodiment of the present disclosure, in some embodiments, the probabilities that different target objects are potential objects may be as shown in Table 2.

[0097] Table 2

[0098]

[0099] According to embodiments of the present disclosure, when determining whether a target object is a potential object, redundant input is reduced and prediction accuracy is improved by filtering a second sub-graph vector related to object attribute information from the graph vector. Furthermore, by inputting the second sub-graph vector into a potential object prediction model, the target object is considered a potential object only when the target probability output by the object prediction sub-model is greater than or equal to a preset threshold. This achieves the screening of recommended objects and saves a certain amount of recommendation resources.

[0100] According to the embodiments of the present disclosure, by analyzing a large amount of object information and combining it with the machine learning capabilities of the restricted Boltzmann machine, individuals or groups with the potential to become high-end customers can be identified, thereby helping to more accurately position the market and formulate more targeted marketing strategies.

[0101] According to an embodiment of the present disclosure, the object information also includes: the association relationship between multiple sub-object information; converting the object information of the target object into a graph vector may include the following operations.

[0102] The object information is corrected to obtain the corrected object information; the corrected object information is extracted from multiple sub-object information and the association relationship between the multiple sub-object information is mapped to obtain a knowledge graph corresponding to the target object, wherein the knowledge graph includes nodes and edges, the nodes represent the sub-object information, and the edges represent the association relationship between the multiple sub-object information; the knowledge graph is mapped into the vector space to obtain a graph vector.

[0103] According to the embodiments of the present disclosure, information correction can be implemented in a variety of ways to address different issues. For example, duplicate information values ​​within object information can be corrected by deleting them. Missing or erroneous information values ​​can be corrected by filling in or deleting them. In other words, obvious errors or abnormal values ​​outside a reasonable range within sub-object information included in object information can be deleted.

[0104] According to embodiments of the present disclosure, entities within object information can be defined. For example, entity definitions can be defined for sub-object information such as individual customers, corporate customers, accounts, products, and services. These entities can serve as nodes in a knowledge graph. Relationships between multiple sub-objects within an object can be defined, such as: corporate representative, shareholder, transfer, guarantee, purchase, consultation, etc. Entity attributes can also be defined, such as: customer name, age, etc.

[0105] According to an embodiment of the present disclosure, the defined entities, relationships, and attributes are mapped to obtain a knowledge graph of the target object.

[0106] Figure 3 A schematic diagram of a knowledge graph according to an embodiment of the present disclosure is schematically shown.

[0107] like Figure 3 As shown, in some embodiments, the knowledge graph of the target object A includes multiple nodes and edges. The nodes can be objects, enterprises, products, etc., such as: target object A, target object B, enterprise A, enterprise B, financial management, precious metals, etc. The edges can be the association relationships between nodes, such as: purchase, transfer, shareholder, agency company, card consumption, purchase, etc.

[0108] According to an embodiment of the present disclosure, Figure 3 The knowledge graph in the figure is only illustrative. In different embodiments, a knowledge graph with different numbers, nodes and edges can be constructed.

[0109] According to the embodiments of the present disclosure, object information can be organized by constructing a knowledge graph, which facilitates subsequent model learning and thereby improves the model calculation speed.

[0110] According to the embodiments of the present disclosure, there is no limitation on the specific means of mapping the knowledge graph to the value vector space, and it may be by using the TransE algorithm, a mapping network, etc.

[0111] According to the embodiments of the present disclosure, in some embodiments, triples can be used to describe the relationship between nodes and edges in the knowledge graph. The relationship in each triple can be regarded as a translation from the head entity to the tail entity. Specifically, the scoring function can be shown as the following formula (1).

[0112] ; (1)

[0113] Among them, h is the head entity, t is the tail entity, and r is the relationship between the head entity and the tail entity.

[0114] According to an embodiment of the present disclosure, a boundary-based loss function (2) can be constructed based on the above formula (1).

[0115] ; (2)

[0116] in, is the positive sample scoring function, is the negative sample scoring function, γ>0 is the boundary parameter, is the negative sample triplet corresponding to T.

[0117] According to the embodiments of the present disclosure, since entities, attributes, and relationships in the knowledge graph are all represented using discrete, explicit symbolic representations, these discrete symbolic representations are difficult to apply to models based on continuous numerical representations. Representation learning is used to map entities and relationships in the knowledge graph into a low-dimensional, continuous vector space. These continuous vectorized representations can serve as model input, allowing the model to fully utilize the prior knowledge in the knowledge graph, thereby obtaining more accurate potential user predictions and product recommendation lists.

[0118] According to the embodiments of the present disclosure, a detailed customer profile can be created through the knowledge graph, including their purchasing behavior, preferences, social relationships, etc. Combined with the product recommendation model and the potential target prediction model, it is possible to further explore the potential value of this data, more accurately understand customer needs and behavior patterns, and thus provide more personalized services and product recommendations.

[0119] According to an embodiment of the present disclosure, performing information correction processing on object information to obtain corrected object information may include the following operations.

[0120] Determine target sub-object information of numerical type from multiple sub-object information; when it is determined that there is an abnormality in the information format of the target object information, correct the target object information based on adjacent object information adjacent to the target object information to obtain first target sub-object information; when it is determined that there is an abnormality in the information value of the target object information, determine homogeneous object information with the same information identifier as the target object information; determine a replacement interval based on the information value of the homogeneous object information; correct the information value of the target object information to a corrected value in the replacement interval to obtain second target sub-object information; determine the corrected object information based on the first target sub-object information and the second target sub-object information.

[0121] According to an embodiment of the present disclosure, there is no limitation on the target sub-object information of numerical type, and it can be a date, a transaction amount, etc.

[0122] According to an embodiment of the present disclosure, the information formats of different target objects may be different. For example, the information format of a date may be year / month / day, where the value representing the month cannot exceed 12, and the value representing the day cannot exceed 31.

[0123] According to an embodiment of the present disclosure, if an abnormality is detected in the information format of a date, the dates adjacent to the current date can be determined and used as a date interval, thereby determining a correction date from the date interval and correcting the date.

[0124] According to an embodiment of the present disclosure, if an abnormality is detected in the information value of the target object, such as a negative transaction amount or an outlier transaction amount, the transaction amount may be an outlier. For example, in the target object's bank account transaction data, most transaction amounts are between a few hundred yuan and tens of thousands of yuan. If a transaction amount of tens of millions of yuan occurs, it may be an abnormality.

[0125] According to the embodiments of the present disclosure, in some embodiments, if there is an abnormality in the transaction amount, all transaction amounts of the target object or the transaction amounts within a period of time can be counted, and the replacement interval can be obtained by calculating the mean, median, standard deviation and other statistics of the transaction amounts. Specifically, in some embodiments, the replacement interval can be set to [mean - 3 times the standard deviation, mean + 3 times the standard deviation], etc.

[0126] According to an embodiment of the present disclosure, a value may be randomly determined from the replacement interval and used as the correction value, or the mean value may be used as the correction value, which is not limited.

[0127] According to the embodiments of the present disclosure, by correcting object information, data quality can be ensured, computing efficiency can be improved, and problems such as missing values ​​and erroneous values ​​in object information can be avoided, which will seriously affect the accuracy and reliability of the knowledge graph. The corrected object information can reduce noise interference in the calculation process and improve the performance and accuracy of the model.

[0128] Figure 4 The flowchart of the product recommendation model training method according to an embodiment of the present disclosure is schematically shown.

[0129] like Figure 4 As shown, the product recommendation model includes a category determination sub-model and a product recommendation sub-model, and the method includes operations S410 to S460.

[0130] In operation S410 , a first sub-graph vector sample of each of a plurality of potential object samples is obtained.

[0131] In operation S420 , for each potential object sample, a plurality of category feature values ​​of the potential object sample are determined based on the node identifiers and node values ​​of the plurality of node vector samples included in the first sub-graph vector sample.

[0132] In operation S430 , an object category label of the potential object sample is determined based on the plurality of category feature values.

[0133] In operation S440 , the initial category determination sub-model is trained based on the first sub-graph vector samples of each of the multiple potential object samples and the object category labels of each of the multiple potential object samples to obtain a category determination sub-model.

[0134] In operation S450, an initial product recommendation sub-model is trained based on the first sub-graph vector samples of each of the multiple potential object samples and the preferred product samples of the similar object samples of each of the multiple potential object samples to obtain a product recommendation sub-model, wherein the object category of the similar object sample is the same as the object category label of the potential object sample.

[0135] In operation S460 , a product recommendation model is obtained based on the category determination sub-model and the product recommendation sub-model.

[0136] According to an embodiment of the present disclosure, a product recommendation model is used to implement the above-mentioned product recommendation method.

[0137] According to an embodiment of the present disclosure, the first sub-graph vector sample of the potential object sample is obtained by performing steps such as operations S221 to S220 on the object information sample of the potential object sample.

[0138] According to an embodiment of the present disclosure, before obtaining object information, a request to obtain object information may be issued to a potential subject sample. If the potential subject sample consents or authorizes the acquisition of object information, the object information of the target object is obtained. Similarly, before processing or making a decision on the object information of the potential subject sample, subsequent processing or decision-making is performed after obtaining the consent or authorization of the potential subject sample. This disclosure will not be further elaborated herein.

[0139] According to the embodiments of the present disclosure, the preferred products of similar objects and the preferred product samples of similar object samples are similarly acquired and processed after obtaining the consent of the similar objects and similar object samples, which will not be elaborated herein.

[0140] According to the embodiments of the present disclosure, different category feature values ​​can represent different aspects of the potential object sample. In some embodiments, the category feature values ​​may include financial management ability, purchasing power, consumption level, payment ability, stable transaction factor or capital return ability, etc.

[0141] According to an embodiment of the present disclosure, the category feature value may be obtained by performing statistical processing on the node values ​​of multiple node identifiers corresponding to the category feature value, such as addition, subtraction, multiplication, division, normalization, logarithmic processing, inverse processing, etc.

[0142] According to an embodiment of the present disclosure, in some embodiments, taking the calculation of a stable transaction factor as an example, it can be obtained by performing relevant statistical processing on a node value corresponding to the following node identifier, wherein the node identifier may include income, deposits, liabilities, financial investment, information rating, and the like.

[0143] According to embodiments of the present disclosure, there may be multiple object categories, each of which may represent the type of group to which the object belongs. In some embodiments, the object categories may include: high-value category, high-reputation category, high-purchasing-power category, high-stickiness category, etc. The high-value category may represent a customer with high purchasing power and consumption level. The high-reputation category may represent a customer with good credit and stable payment ability. The high-purchasing-power category may represent a customer who purchases a wide variety of financial products. The high-stickiness category may represent a customer with relatively stable cash flow characteristics.

[0144] In the fundamentally disclosed embodiments, different object categories may correspond to different preset weight groups, which include multiple preset category weights, each preset category weight corresponding to a category feature value. There is no limitation on the method for determining the preset weight group, which may be obtained through experiments.

[0145] According to an embodiment of the present disclosure, multiple category feature values ​​are calculated with respective weight groups of each object category to obtain calculation results corresponding to multiple category labels. The largest one is determined from the calculation results corresponding to the multiple object categories, and the object category corresponding to the largest calculation result is considered to be the object category label of the potential object.

[0146] According to an embodiment of the present disclosure, in some embodiments, the determined category feature values ​​and object category labels for different potential object samples may be as shown in Table 3.

[0147] Table 3

[0148]

[0149] Figure 5 A schematic diagram of category feature values ​​for different potential object samples according to an embodiment of the present disclosure is schematically shown.

[0150] like Figure 5 As shown, potential object sample A, potential object sample B, potential object sample C, and potential object sample D have their respective category characteristic values, which are financial management ability, purchasing power, consumption level, payment ability, stable transaction factor or capital return ability.

[0151] According to an embodiment of the present disclosure, the category feature values ​​of potential object samples are calculated by using node identifiers and node values ​​corresponding to the first sub-graph vector samples, thereby achieving quantification of relevant feature tendencies of potential object samples, thereby achieving more fine-grained classification of potential object samples, and labeling potential object samples. By labeling multiple potential object samples with multiple category feature values, the initial category determination sub-model is trained with the first sub-graph vector samples of each of the multiple potential object samples and the object category labels of each of the multiple potential object samples to obtain a category determination sub-module that can achieve more fine-grained classification, and the initial product recommendation sub-model is trained with the preferred product samples of the first sub-graph vector samples and the similar object samples to obtain a product recommendation sub-model for recommending products.

[0152] According to an embodiment of the present disclosure, in the process of training a category determination sub-model, a product recommendation sub-model, or a potential object prediction model, the training samples may be divided into a training set and a test set, such as dividing multiple potential object samples.

[0153] According to an embodiment of the present disclosure, when RBM is used to implement the above model, if the above model has m visible units and n hidden units, v and h represent the visible unit state and the hidden unit state respectively, then the energy function is shown in the following formula (3).

[0154] ; (3)

[0155] Among them, a i is the weight of the i-th visible unit, b j is the weight of the jth hidden unit, w ij is the weight between the i-th visible unit and the j-th hidden unit.

[0156] According to an embodiment of the present disclosure, the number of nodes in the visible layer and the hidden layer can be determined based on the feature dimensions of the training set. For example, if the training set contains 36 features, the number of nodes in the visible layer can be set to 36. In order to achieve the purpose of dimensionality reduction and feature extraction, the number of parameters of the hidden layer is generally less than the number of nodes in the visible layer. The number of nodes can be continuously adjusted during the model training process.

[0157] According to an embodiment of the present disclosure, the weight matrix is ​​a two-dimensional array in which rows identify visible layer units and columns identify hidden layer units. Initialization uses a uniform distribution. During the training process of the above model, the weight matrix needs to be updated based on the calculation results until the model reaches the maximum number of iterations or the performance on the validation set no longer improves. The bias terms of the visible and hidden layers correspond to the corresponding number of nodes. Like the weight matrix, the bias terms need to be continuously updated during the training process of the model to achieve a better fit.

[0158] According to the embodiments of the present disclosure, in some embodiments, taking the training of a potential object prediction model as an example, the object information of multiple initial object samples can be respectively divided into training sets and test sets, and the multiple initial object samples each have label data, and the identification data includes whether the initial object sample is a potential object. It should be noted that the object information of the initial object sample is also obtained or processed with the permission of the initial object sample, which will not be repeated here.

[0159] According to an embodiment of the present disclosure, model parameters, including weights and bias terms, can be initialized during the training process. Forward propagation: The training set is used as the input of the visible layer, and the activation value of the hidden layer is calculated through the weights and bias terms. The activation value of the hidden layer is converted into a probability distribution using an activation function such as the Sigmoid function. Backward propagation: Samples are generated based on the probability distribution of the hidden layer. These samples are used to calculate the reconstruction error, that is, the difference between the original input and the reconstructed input. The reconstruction error is used to calculate the gradient of the loss function with respect to the weights and bias terms. Based on the calculated gradient, an optimization algorithm such as gradient descent or its variants is used to update the weights and bias terms of the potential object prediction model. Forward propagation and backpropagation are iteratively performed until the model converges or the preset number of iterations is reached.

[0160] According to embodiments of the present disclosure, the performance of a potential object prediction model can be evaluated using a test set. Metrics such as precision, recall, and F1 score are calculated to assess the quality of the model. Based on the model evaluation results, model hyperparameters such as the learning rate and number of iterations can be adjusted to improve model performance.

[0161] Figure 6 The data flow diagram of the product recommendation model training method according to an embodiment of the present disclosure is schematically shown.

[0162] like Figure 6 As shown, in the process of training the product recommendation model, the object information 610 of the potential object sample can be collected first, and the object information 610 of the potential object sample can be corrected to obtain the corrected object information 620. The knowledge graph sample 630 is constructed through the corrected object information 620, and the knowledge graph sample is vectorized to obtain a graph vector sample 640. The graph vector sample 640 is divided into a training set 650 and a test set 660. The initial category determination sub-model and the initial product recommendation sub-model are trained through the training set 650 to obtain an intermediate product recommendation model 670. The product recommendation model 670 is evaluated through the test set 660. When it is determined that there is a problem with the intermediate product recommendation model, the intermediate product recommendation model 670 is optimized until it is determined that the intermediate product recommendation model meets the conditions and the product recommendation model 680 is obtained.

[0163] Based on the above product recommendation method and product recommendation model training method, the present disclosure also provides a product recommendation device and a product recommendation model training device. Figure 7-Figure 8 Describe it in detail.

[0164] Figure 7 The following schematically shows a structural block diagram of a product recommendation device according to an embodiment of the present disclosure.

[0165] like Figure 7 As shown, the product recommendation device 700 includes a conversion module 710 , a first determination module 720 , a second determination module 730 and a recommendation module 740 .

[0166] The conversion module 710 is used to convert the object information of the target object into a graph vector with the permission of the target object, wherein the object information includes multiple sub-object information, the graph vector includes multiple node vectors, and the node vector is used to represent the object sub-information.

[0167] The first determination module 720 is used to determine a first sub-graph vector that matches preset product attribute information from the graph vector based on the node identifiers of multiple node vectors when the target object is determined to be a potential object based on the graph vector.

[0168] The second determination module 730 is configured to input the first sub-graph vector into a product recommendation model to obtain the object category of the target object and a product recommendation list that matches both the object category and the target object.

[0169] The recommendation module 740 is configured to recommend the product to be recommended determined from the product recommendation list to the target object.

[0170] According to an embodiment of the present disclosure, the second determining module 730 includes the following submodules.

[0171] The first input submodule is used to input the first sub-atlas vector into the category determination submodel to obtain the object category of the target object.

[0172] The second input submodule is used to input the preferred products of similar objects with the object category and the first sub-graph vector into the product recommendation submodel to obtain a product recommendation list that matches both the object category and the target object.

[0173] According to an embodiment of the present disclosure, the second input submodule includes the following units.

[0174] The feature determination unit is used to input the preferred products of similar objects and the first sub-graph vector into the product recommendation sub-model to obtain the first preferred product features of similar objects and the second preferred product features of the target object.

[0175] The first calculation unit is configured to calculate similarities between first preferred product features of a plurality of similar objects and product features of a plurality of preset products, to obtain first similarities respectively corresponding to the plurality of preset products.

[0176] The second calculation unit is configured to calculate similarities between the second preferred product feature and the product features of each of the plurality of preset products to obtain second similarities respectively corresponding to the plurality of preset products.

[0177] A recommendation value determination unit is used to determine the recommendation value of each of multiple preset products based on a first weight value corresponding to the first preferred product feature, a first similarity corresponding to each of the multiple preset products, a second weight value corresponding to the second preferred product feature, and a second similarity corresponding to each of the multiple preset products.

[0178] The list determining unit is used to obtain a product recommendation list based on the recommendation values ​​of each of a plurality of preset products.

[0179] According to an embodiment of the present disclosure, the product recommendation device 700 further includes the following modules.

[0180] The fifth determining module is used to determine a second sub-graph vector that matches the preset object attribute information from the graph vector based on the node identifiers of the multiple node vectors.

[0181] The probability determination module is used to input the second sub-graph vector into the potential object prediction model to obtain the target probability that the target object is a potential object.

[0182] The object determination module is used to determine that the target object is a potential object when the target probability is greater than or equal to a preset probability threshold.

[0183] According to an embodiment of the present disclosure, the object information further includes: association relationships between a plurality of sub-object information. The conversion module includes the following sub-modules.

[0184] The processing submodule is used to perform information correction processing on the object information to obtain corrected object information.

[0185] The graph generation submodule is used to extract information of multiple sub-objects from the corrected object information and map the association relationships between the multiple sub-object information to obtain a knowledge graph corresponding to the target object, wherein the knowledge graph includes nodes and edges, the nodes represent the sub-object information, and the edges represent the association relationships between the multiple sub-object information.

[0186] The vector conversion submodule is used to map the knowledge graph into the vector space to obtain the graph vector.

[0187] According to an embodiment of the present disclosure, the processing submodule includes the following units.

[0188] The first information determining unit is configured to determine target sub-object information of a numerical type from a plurality of sub-object information.

[0189] The first correction unit is configured to correct the target object information based on adjacent object information adjacent to the target object information when determining that an abnormality exists in the information format of the target object information, so as to obtain first target sub-object information.

[0190] The second information determining unit is configured to determine homogeneous object information having the same information identifier as the target object information when it is determined that the information value of the target object information is abnormal.

[0191] The interval determining unit is configured to determine a replacement interval based on the information value of the homogeneous object information.

[0192] The replacement unit is configured to correct the information value of the target object information to a correction value in the replacement interval to obtain the second target sub-object information.

[0193] The third information determining unit is configured to determine the corrected object information based on the first target sub-object information and the second target sub-object information.

[0194] Figure 8 The structural block diagram of the product recommendation model training device according to an embodiment of the present disclosure is schematically shown.

[0195] like Figure 8 As shown, the product recommendation model training device 800 includes: an acquisition module 810, a third determination module 820, a fourth determination module 830, a first training module 840, a second training module 850, and a model determination module 860.

[0196] The acquisition module 810 is configured to acquire a first sub-atlas vector sample of each of a plurality of potential object samples.

[0197] The third determination module 820 is configured to determine, for each of the potential object samples, a plurality of category feature values ​​of the potential object sample based on the node identifiers and node values ​​of the plurality of node vector samples included in the first sub-graph vector sample.

[0198] The fourth determination module 830 is configured to determine the object category label of the potential object sample based on the plurality of category feature values.

[0199] The first training module 840 is used to train the initial category determination sub-model based on the first sub-graph vector samples of each of the multiple potential object samples and the object category labels of each of the multiple potential object samples to obtain the category determination sub-model.

[0200] The second training module 850 is used to train the initial product recommendation sub-model based on the first sub-graph vector samples of each of the multiple potential object samples and the preferred product samples of the same type of object samples of each of the multiple potential object samples, so as to obtain the product recommendation sub-model, wherein the object category of the same type of object samples is the same as the object category label of the potential object sample.

[0201] The model determination module 860 is configured to determine the sub-model and the product recommendation sub-model based on the category to obtain the product recommendation model.

[0202] According to an embodiment of the present disclosure, any multiple modules among the conversion module 710, the first determination module 720, the second determination module 730, the recommendation module 740, the acquisition module 810, the third determination module 820, the fourth determination module 830, the first training module 840, the second training module 850, and the model determination module 860 may be combined into a single module for implementation, or any one of these modules may be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules may be combined with at least part of the functionality of other modules and implemented in a single module. According to an embodiment of the present disclosure, at least one of the conversion module 710, the first determination module 720, the second determination module 730, the recommendation module 740, the acquisition module 810, the third determination module 820, the fourth determination module 830, the first training module 840, the second training module 850, and the model determination module 860 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or can be implemented in any one of the three implementation methods of software, hardware, and firmware, or in an appropriate combination of any of them. Alternatively, at least one of the conversion module 710, the first determination module 720, the second determination module 730, the recommendation module 740, the acquisition module 810, the third determination module 820, the fourth determination module 830, the first training module 840, the second training module 850, and the model determination module 860 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0203] Figure 9 A block diagram of an electronic device suitable for implementing a product recommendation method and a product recommendation model training method according to an embodiment of the present disclosure is schematically shown.

[0204] like Figure 9As shown, the electronic device 900 according to an embodiment of the present disclosure includes a processor 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 unit 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to the embodiment of the present disclosure.

[0205] Various programs and data required for the operation of the electronic device 900 are stored in the RAM 903. The processor 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. The processor 901 executes the various operations of the method flow according to the embodiment of the present disclosure by executing the programs in the ROM 902 and / or the RAM 903. It should be noted that the programs may also be stored in one or more memories other than the ROM 902 and the RAM 903. The processor 901 may also execute the various operations of the method flow according to the embodiment of the present disclosure by executing the programs stored in the one or more memories.

[0206] According to an embodiment of the present disclosure, electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to bus 904. Electronic device 900 may also include one or more of the following components connected to I / O interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage section 908 including a hard disk; and a communication section 909 including a network interface card such as a LAN card or modem. Communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to I / O interface 905 as needed. Removable media 911, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in drive 910 as needed, so that computer programs read from the removable media can be installed into storage section 908 as needed.

[0207] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when executed, implements the method according to the embodiments of the present disclosure.

[0208] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, and may include, for example, but is not limited to: 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), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 902 and / or RAM 903 described above, and / or one or more memories other than ROM 902 and RAM 903.

[0209] Embodiments of the present disclosure also include a computer program product, comprising a computer program containing program code for executing the method shown in the flowchart. When the computer program product is executed in a computer system, the program code is used to cause the computer system to implement the product recommendation method and product recommendation model training method provided in the embodiments of the present disclosure.

[0210] The computer program executes the above functions defined in the system / device of the embodiment of the present disclosure when the processor 901 executes the computer program. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0211] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal on a network medium, downloaded and installed via the communication portion 909, and / or installed from a removable medium 911. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0212] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 909, and / or installed from a removable medium 911. When the computer program is executed by the processor 901, the above-described functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the systems, devices, means, modules, units, etc. described above can be implemented by computer program modules.

[0213] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C" or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect via the Internet).

[0214] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order 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 flowchart, and the combination of boxes in the block diagram or flowchart, 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.

[0215] Those skilled in the art will appreciate that the features described in the various embodiments of the present disclosure may be combined and / or coupled in various ways, even if such combinations or couplings are not explicitly described in the present disclosure. In particular, the features described in the various embodiments of the present disclosure may be combined and / or coupled in various ways without departing from the spirit and teachings of the present disclosure. All such combinations and / or couplings fall within the scope of the present disclosure.

[0216] The above describes the embodiments of the present disclosure. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be advantageously used in combination. Without departing from the scope of the present disclosure, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A product recommendation method, characterized in that: The method comprises: With permission from the target object, convert the object information of the target object into a graph vector, wherein the object information includes a plurality of sub-object information, and the graph vector includes a plurality of node vectors, and the node vectors are used to represent the object sub-information; In a case where the target object is determined to be a potential object based on the graph vector, determining a first sub-graph vector matching preset product attribute information from the graph vector based on node identifiers of a plurality of node vectors; Inputting the first sub-graph vector into a product recommendation model to obtain an object category of the target object and a product recommendation list that matches both the object category and the target object; recommending the product to be recommended determined from the product recommendation list to the target object; The product recommendation model includes a category determination sub-model and a product recommendation sub-model; The step of inputting the first sub-graph vector into a product recommendation model to obtain the object category of the target object and a product recommendation list matching both the object category and the target object includes: Inputting the first sub-graph vector into the category determination sub-model to obtain the object category of the target object; Inputting the preferred products of the same type of objects and the first sub-graph vector into the product recommendation sub-model to obtain the first preferred product features of the same type of objects and the second preferred product features of the target object; Calculating similarities between the first preferred product features of the plurality of similar objects and the product features of the plurality of preset products to obtain first similarities corresponding to the plurality of preset products; Calculating similarity between the second preferred product feature and the product features of each of the plurality of preset products to obtain second similarities corresponding to the plurality of preset products; determining a recommendation value for each of the plurality of preset products based on a first weight value corresponding to the first preferred product feature, first similarities respectively corresponding to the plurality of preset products, a second weight value corresponding to the second preferred product feature, and second similarities respectively corresponding to the plurality of preset products; Based on the recommendation values ​​of each of the plurality of preset products, the product recommendation list is obtained.

2. The method according to claim 1, characterized in that The method further comprises: Based on the node identifiers of the plurality of node vectors, determining a second sub-graph vector matching the preset object attribute information from the graph vector; Inputting the second sub-graph vector into a potential object prediction model to obtain a target probability that the target object is a potential object; When it is determined that the target probability is greater than or equal to a preset probability threshold, the target object is determined to be a potential object.

3. The method according to claim 1, characterized in that The object information also includes: association relationships between multiple sub-object information; The converting the object information of the target object into a graph vector includes: performing information correction processing on the object information to obtain corrected object information; Extracting information of the plurality of sub-object information and mapping the association relationships between the plurality of sub-object information from the corrected object information to obtain a knowledge graph corresponding to the target object, wherein the knowledge graph includes nodes and edges, the nodes representing the sub-object information, and the edges representing the association relationships between the plurality of sub-object information; The knowledge graph is mapped into a vector space to obtain the graph vector.

4. The method according to claim 1, wherein The performing information correction processing on the object information to obtain corrected object information includes: Determine target sub-object information of a numerical type from the plurality of sub-object information; When it is determined that the information format of the target object information is abnormal, the target object information is modified based on the adjacent object information adjacent to the target object information to obtain first target sub-object information; When it is determined that the information value of the target object information is abnormal, Determining homogeneous object information having the same information identifier as the target object information; determining a replacement interval based on the information value of the homogeneous object information; Correcting the information value of the target object information to the correction value in the replacement interval to obtain second target sub-object information; The revised object information is determined based on the first target sub-object information and the second target sub-object information.

5. A product recommendation device, configured to execute the method according to any one of claims 1 to 4, characterized in that: The device comprises: a conversion module, configured to convert the object information of the target object into a graph vector with permission from the target object, wherein the object information includes a plurality of sub-object information, the graph vector includes a plurality of node vectors, and the node vectors are used to represent the object sub-information; A first determining module is configured to, when the target object is determined to be a potential object based on the graph vector, determine, based on the node identifiers of the plurality of node vectors, a first sub-graph vector that matches preset product attribute information from the graph vector; A second determination module is configured to input the first sub-graph vector into a product recommendation model to obtain an object category of the target object and a product recommendation list that matches both the object category and the target object; The recommendation module is used to recommend the recommended product determined from the product recommendation list to the target object.

6. An electronic device comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

8. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images