Object recommendation method and device, equipment, storage medium and program product

Abstract

The embodiment of the invention provides an object recommendation method and device, equipment, a storage medium and a program product, and relates to the field of big data. The method comprises the steps of judging whether an identifier of a target first object exists in an identifier library of the first object or not; if the identifier of the target first object does not exist in the identifier library of the first object, generating a target first object feature vector of the target first object according to the attribute information of the target first object based on a graph convolutional neural network; calculating a first similarity between the target first object feature vector and each first object feature in the identification library of the first object; and determining an identifier of a second object corresponding to the first object feature of which the first similarity is greater than a first preset similarity as an identifier of a to-be-recommended second object corresponding to the target first object. According to the method, the technical effect of improving the personalized recommendation accuracy is achieved.

Description

Technical Field

[0001] The present application relates to the field of big data, and in particular to an object recommendation method, device, equipment, storage medium and program product. Background Art

[0002] With the development of financial technology, in order to increase the customer flow of financial products, personalized recommendation technology for financial products has gradually become the main way to acquire customers in the financial field. In the financial field, corresponding product trading platforms will be developed for various financial products, and corresponding products will be recommended and traded based on the trading platforms.

[0003] In the prior art, when recommending products in the financial field, machine learning and deep learning are mainly used to obtain the user's preference for different products, and based on the user's preference, recommend products that better meet the user's needs, thereby achieving the purpose of personalized recommendation.

[0004] Therefore, there is a technical problem in the prior art that the accuracy of personalized recommendations is low. Summary of the invention

[0005] The present application provides an object recommendation method, apparatus, device, storage medium and program product to solve the technical problem of low data query efficiency.

[0006] In a first aspect, the present application provides an object recommendation method, comprising:

[0007] Determine whether the identifier of the target first object exists in an identifier library of the first object, wherein the identifier library of the first object stores identifiers corresponding to each first object, first object features, and second object identifiers, wherein the first object features are generated by a graph convolutional neural network according to attribute information of the first object;

[0008] If the identifier of the target first object does not exist in the identifier library of the first object, generating a target first object feature vector of the target first object based on the graph convolutional neural network according to the attribute information of the target first object;

[0009] Calculating a first similarity between the target first object feature vector and each first object feature in the identification library of the first object;

[0010] The identifier of the second object corresponding to the first object feature whose first similarity is greater than the first preset similarity is determined as the identifier of the second object to be recommended corresponding to the target first object.

[0011] In a possible implementation, when the first object is a user, the second object is a product; when the first object is an object, the first object is a product.

[0012] In one possible embodiment, the method further comprises:

[0013] If the identifier of the target first object exists in the identifier library of the first object, the identifier of the second object corresponding to the target first object is determined as the identifier of the second object to be recommended corresponding to the target first object.

[0014] In a possible implementation manner, before determining whether the identifier of the target first object exists in the identifier library of the first object, the method further includes:

[0015] If the identifier of the target second object does not exist in the identifier library of the second object, recommending the target second object to the plurality of test first objects;

[0016] Obtaining acquisition behavior information of each test first object on the target second object;

[0017] According to the acquisition behavior information of each test first object, the test first object having an acquisition behavior on the target second object is determined as the target first object.

[0018] In a possible implementation manner, before determining whether the identifier of the target first object exists in the identifier library of the first object, the method further includes:

[0019] The user's identification and user attribute information are processed through a graph convolutional neural network to generate the user's corresponding user features;

[0020] The product identification and product attribute information are processed through the graph convolutional neural network to generate the corresponding product features;

[0021] For each user, calculating the second similarity between the user feature corresponding to the user and each product feature;

[0022] Determine the identifier of the product corresponding to the product feature whose second similarity is greater than the second preset similarity as the identifier of the product corresponding to the identifier of the user;

[0023] According to the correspondence between the product identification and the user identification, the product identification, product attribute information, product characteristics, the user identification, user attribute information and user characteristics, an identification library of the first object and an identification library of the second object are constructed.

[0024] In a possible implementation manner, before determining whether the identifier of the target first object exists in the identifier library of the first object, the method further includes:

[0025] Obtain product information of each sample product and user information of each sample user, wherein the product information of the sample product includes an identifier corresponding to the sample product, product attribute sub-information, an identifier of the sample user who has acquired the sample product, and a click volume; the user information of the sample user includes an identifier corresponding to the sample user, user attribute sub-information, an identifier of the sample product acquired by the sample user, and an identifier of the sample product clicked by the sample user;

[0026] The model is trained based on the product information of each sample product and the user information of each sample user to generate a graph convolutional neural network.

[0027] In a possible implementation, model training is performed based on product information of each sample product and user information of each sample user to generate a graph convolutional neural network, including:

[0028] According to the product information of each sample product and the user information of each sample user, an interaction matrix is ​​constructed, and the interaction matrix is ​​used to indicate whether each sample user obtains each sample product;

[0029] Generate an adjacency matrix according to the interaction matrix and the transposed matrix corresponding to the interaction matrix;

[0030] Generate initial sample product features corresponding to the sample product according to the product information of each sample product;

[0031] Generate initial sample user features corresponding to each sample user based on the user information of each sample user;

[0032] Perform graph convolution operations on the initial sample product features based on the adjacency matrix to obtain the sample product features corresponding to the sample product nodes;

[0033] Perform graph convolution operation on the initial sample user features based on the adjacency matrix to obtain the sample user features corresponding to the sample user nodes;

[0034] The model is trained based on sample product features and sample user features until the training cutoff condition is met to obtain a graph convolutional neural network.

[0035] In a second aspect, the present application provides an object recommendation device, comprising:

[0036] A first processing module is used to determine whether the identifier of the target first object exists in an identifier library of the first object, wherein the identifier library of the first object stores identifiers corresponding to each first object, first object features, and second object identifiers, wherein the first object features are generated by a graph convolutional neural network according to attribute information of the first object;

[0037] A second processing module is used to generate a target first object feature vector of the target first object based on the graph convolutional neural network according to the attribute information of the target first object when the identifier of the target first object does not exist in the identifier library of the first object;

[0038] A third processing module, used for calculating a first similarity between the target first object feature vector and each first object feature in the identification library of the first object;

[0039] The fourth processing module is used to determine the identifier of the second object corresponding to the first object feature whose first similarity is greater than the first preset similarity as the identifier of the second object to be recommended corresponding to the target first object.

[0040] In a possible implementation, when the first object is a user, the second object is a product; when the first object is an object, the first object is a product.

[0041] In one possible implementation, the second processing module is further configured to:

[0042] When the identifier of the target first object exists in the identifier library of the first object, the identifier of the second object corresponding to the target first object is determined as the identifier of the second object to be recommended corresponding to the target first object.

[0043] In a possible implementation manner, the first processing module is further configured to:

[0044] When the identifier of the target second object does not exist in the identifier library of the second object, recommending the target second object to the plurality of test first objects;

[0045] Obtaining acquisition behavior information of each test first object on the target second object;

[0046] According to the acquisition behavior information of each test first object, the test first object having an acquisition behavior on the target second object is determined as the target first object.

[0047] In a possible implementation manner, the first processing module is further configured to:

[0048] The user's identification and user attribute information are processed through a graph convolutional neural network to generate the user's corresponding user features;

[0049] The product identification and product attribute information are processed through the graph convolutional neural network to generate the corresponding product features;

[0050] For each user, calculating the second similarity between the user feature corresponding to the user and each product feature;

[0051] Determine the identifier of the product corresponding to the product feature whose second similarity is greater than the second preset similarity as the identifier of the product corresponding to the identifier of the user;

[0052] According to the correspondence between the product identification and the user identification, the product identification, product attribute information, product characteristics, the user identification, user attribute information and user characteristics, an identification library of the first object and an identification library of the second object are constructed.

[0053] In a possible implementation, the device further includes an acquisition module, configured to:

[0054] Obtain product information of each sample product and user information of each sample user, wherein the product information of the sample product includes an identifier corresponding to the sample product, product attribute sub-information, an identifier of the sample user who has acquired the sample product, and a click volume; the user information of the sample user includes an identifier corresponding to the sample user, user attribute sub-information, an identifier of the sample product acquired by the sample user, and an identifier of the sample product clicked by the sample user;

[0055] The model is trained based on the product information of each sample product and the user information of each sample user to generate a graph convolutional neural network.

[0056] In a possible implementation, the acquisition module is further used to:

[0057] According to the product information of each sample product and the user information of each sample user, an interaction matrix is ​​constructed, and the interaction matrix is ​​used to indicate whether each sample user obtains each sample product;

[0058] Generate an adjacency matrix according to the interaction matrix and the transposed matrix corresponding to the interaction matrix;

[0059] Generate initial sample product features corresponding to the sample product according to the product information of each sample product;

[0060] Generate initial sample user features corresponding to each sample user based on the user information of each sample user;

[0061] Perform graph convolution operations on the initial sample product features based on the adjacency matrix to obtain the sample product features corresponding to the sample product nodes;

[0062] Perform graph convolution operation on the initial sample user features based on the adjacency matrix to obtain the sample user features corresponding to the sample user nodes;

[0063] The model is trained based on sample product features and sample user features until the training cutoff condition is met to obtain a graph convolutional neural network.

[0064] In a third aspect, an embodiment of the present application provides an electronic device, including: a memory, a processor;

[0065] Memory stores computer-executable instructions;

[0066] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementations of the first aspect.

[0067] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the first aspect above and / or various possible implementations of the first aspect.

[0068] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements various possible implementation methods of the first aspect or the second aspect above.

[0069] The object recommendation method, device, equipment, storage medium and program product provided by the present application are based on the method to determine whether the identification of the target first object currently used for recommendation exists in the identification library of the first object; when the identification of the target first object does not exist in the identification library of the first object, the target first object feature vector is generated based on the attribute information of the target first object, the first similarity between the target first object feature vector and each first object feature in the identification library is calculated, and the identification of the second object corresponding to the first object feature whose first similarity is greater than the first preset similarity is determined as the identification of the second object to be recommended corresponding to the target first object. The identification library is used to calculate the products or users of new users or new products in the identification library whose similarity is greater than the first preset similarity, and the corresponding recommended user list is determined according to the existing products in the identification library, or the corresponding recommended product list is determined according to the existing users, so as to obtain the recommended user list corresponding to the new product or the recommended product list corresponding to the new user, and realize personalized recommendation for new products or new users. Thereby, the technical effect of improving the accuracy of personalized recommendation is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0071] Figure 1 A system architecture diagram of an object recommendation system provided in this application;

[0072] Figure 2 A flowchart of the first embodiment of the object recommendation method provided by the present application;

[0073] Figure 3 A flowchart of the second embodiment of the object recommendation method provided by this application;

[0074] Figure 4 A schematic diagram of the structure of the object recommendation device provided by this application;

[0075] Figure 5 A schematic diagram of the structure of the electronic device provided in this application.

[0076] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0077] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0078] First, the terms involved in this application are explained:

[0079] Graph Convolutional Network (GCN): is a neural network model specifically designed to process graph structured data. Traditional convolutional neural networks are very effective in processing regular grid data, while GCN expands the convolution operation to adapt to irregular graph structured data.

[0080] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application 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.

[0081] It should be noted that the object recommendation method, apparatus, equipment, storage medium and program product provided in the present application can be used in the field of big data, and can also be used in any field other than the field of big data. The application field of the object recommendation method, apparatus, equipment, storage medium and program product in the present application is not limited.

[0082] In the prior art, when making personalized recommendations for financial products, the combination of deep learning and machine learning is mainly used to achieve personalized recommendations for products. The specific method is: first determine the sample information used for model training, which includes product information and user information; extract product features and user features based on the sample information, and train the model based on the product features and user features in the sample information to obtain a recommendation model that can be used for user preference scores; input the target user group and target product group for recommendation into the recommendation model, and the preference score of each target user for each target product, and sort the target products from high to low according to the preference score, so as to determine the product recommendation list for each target user.

[0083] However, the prior art mainly calculates the product score for the user based on the user's preference for the product when making personalized recommendations. When the target user for whom product recommendations are required is a new user, the user's attribute information does not contain the usage record of the current target product. Therefore, when personalized recommendations are made based on the recommendation model, the new user's preference score for each target product cannot accurately reflect the user's true preference, and thus an accurate personalized product recommendation list cannot be obtained. At the same time, when the target product for product recommendation is a new product, the relationship between the new product and the existing multiple target products is not considered. When personalized recommendations are made, the new product will always be in a low score state and cannot achieve effective exposure; therefore, the prior art has a technical problem of low accuracy of personalized recommendations.

[0084] In view of the above technical problems, the present application proposes the following technical concept: the technical problem existing in the prior art is that the characteristic relationship between users and the characteristic relationship between products is not considered, and personalized product recommendations are made according to the preferences between users and products, which leads to the technical problem of low accuracy of personalized recommendations. Based on the technical problems existing in the prior art, the present application calculates the feature vector of the new product or new user based on the graph convolutional neural network when there is a new product or a new user, and calculates the similarity between the target first object feature vector calculated based on the target first object corresponding to the new product or the new user and each first object feature in the identification library, so as to achieve the purpose of calculating the similarity between users and users and the similarity between products, and obtains the first object with the highest similarity to the target first object by using the first similarity calculation, and obtains the identification of the second object corresponding to the first object, so as to determine the second object identification for recommendation to the target first object, so as to achieve the purpose of realizing personalized recommendations based on the similarity between users and the similarity between products; on the basis of calculating the similarity between products and users, the accuracy of personalized recommendations is further increased, so as to achieve the technical effect of improving the accuracy of personalized recommendations.

[0085] Figure 1 This is a system architecture diagram of an object recommendation system provided by this application, and the object recommendation system is a computer device. Figure 1 As shown, the above framework includes at least one of a data acquisition device 101 , a processing device 102 , and a display device 103 .

[0086] It is understandable that the structure illustrated in the embodiment of the present application does not constitute a specific limitation on the above-mentioned system architecture. In other feasible implementations of the present application, the above-mentioned architecture may include more or fewer components than those shown in the figure, or combine certain components, or split certain components, or arrange the components differently, which can be determined according to the actual application scenario and is not limited here. Figure 1 The components shown may be implemented in hardware, software, or a combination of software and hardware.

[0087] In a specific implementation process, the data acquisition device 101 may include an input / output interface and may also include a communication interface. The data acquisition device 101 may be connected to the processing device via the input / output interface or the communication interface.

[0088] The processing device 102 can determine whether the identification of the target first object exists in the identification library of the first object based on the identification of the target first object. If not, calculate the target first object feature vector corresponding to the attribute information of the target first object based on the convolutional neural network; calculate the first similarity between the target first object and each first object based on the target first object feature vector, obtain the first object corresponding to the first similarity exceeding the first preset similarity based on multiple first similarities, and determine the identification of the second object recommended to the target first object based on the identification of the second object corresponding to the first object.

[0089] The display device 103 may also be a touch display screen or a screen of a terminal device, which is used to receive user instructions while displaying the above content to achieve interaction with the user.

[0090] It should be understood that the above-mentioned processing device can be implemented by a processor reading instructions in a memory and executing the instructions, or it can be implemented by a chip circuit.

[0091] In addition, the network architecture and business scenarios described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Ordinary technicians in this field can know that with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0092] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0093] Figure 2 This is a flow chart of the first embodiment of the object recommendation method provided by this application, such as Figure 2 As shown, the method includes:

[0094] S201: Determine whether the identifier of the target first object exists in the identifier library of the first object.

[0095] In this step, the identification library of the first object stores the identification corresponding to each first object, the first object feature and the identification of the second object, and the first object feature is generated by a graph convolutional neural network according to the attribute information of the first object.

[0096] Optionally, when the first object is a user, the second object is a product; when the first object is an object, the first object is a product.

[0097] In this step, if it is determined that the first object is a user, the identifier of the first object is the user identifier, the first object feature is the user feature, and the identifier of the second object refers to the product identifier corresponding to each user identifier, and the product identifier refers to the product that can be recommended to the user; if the first object is a product, the identifier of the first object is the product identifier, the first object feature is the product feature, and the identifier of the second object refers to the user identifier corresponding to each product identifier, and the user identifier refers to the user to whom the product can be recommended. The product or user is used to determine the identifier library, and the identifier library stores the product features, product identifiers, and user identifiers corresponding to each product, or the user features, user representations, and product identifiers corresponding to each user; thereby determining an identifier library for comparison with a target user or target product. If the target user exists in the identifier library, the product identifier corresponding to the target user can be directly obtained. If the target product exists in the identifier library, the user identifier of the target product can be directly obtained, thereby achieving efficient personalized recommendation information acquisition and improving the efficiency of information acquisition.

[0098] Optionally, before determining whether the identifier of the target first object exists in the identifier library of the first object, it is necessary to determine the target first object. A possible implementation method for determining the target first object is:

[0099] S2011. If the identifier of the target second object does not exist in the identifier library of the second object, recommend the target second object to multiple test first objects.

[0100] In this step, the target second object refers to the product, and the multiple test first objects refer to the test users. When the identifier of the target second object does not exist in the identifier library of the second object, it indicates that the product corresponding to the current target second object is a new product. When recommending new products, the cold start mechanism can be used to recommend the new products to multiple test users. The feedback information of the test users can be used to determine the target users who have acquisition behavior for the new products, and further personalized recommendations can be implemented based on the target users.

[0101] S2012. Obtain acquisition behavior information of each test first object on the target second object.

[0102] In this step, a possible implementation method of obtaining the acquisition behavior information of each first object on the target second object at this time is:

[0103] a1. Recommend the target product to multiple test users.

[0104] a2. Periodically collect behavior logs of multiple test users.

[0105] a3. Determine each test user's acquisition behavior information of the target product based on the behavior logs of multiple test users.

[0106] The behavior information obtained includes: whether there is a product transaction, whether there is a click, the transaction time, the click time, the transaction product ID, and the click product ID.

[0107] S2013. According to the acquisition behavior information of each test first object, determine the test first object that has an acquisition behavior on the target second object as the target first object.

[0108] Exemplarily, the target second object is product A, and the multiple test first objects are (user A, user B, user C, user D). The acquisition behavior information of each test first object is: there is a transaction behavior of product A in the behavior log of user A, there is no transaction behavior of product A in the behavior log of user B, there is a transaction behavior of product A in the behavior log of user C, and there is no transaction behavior of product A in the behavior log of user D; according to the acquisition behavior information of each test first object, it can be determined that user A and user C have acquisition behavior for product A, so user A and user C are determined as the target first objects.

[0109] In steps S2011 to S2013, the product is cold started by recommending the target second object to multiple test first objects, ensuring that when a new product exists, the product exposure can be improved by recommending the new product to some test users, avoiding the product from not being accurately recommended to users who prefer such products due to low exposure, thereby achieving the technical effect of improving the accuracy of personalized recommendations.

[0110] It should be noted that before implementing recommendations based on the identification of the target first object, it is also necessary to build an identification library for the first object. Before building the identification library for the first object, it is necessary to train a graph convolutional neural network model for feature calculation. The specific methods for training the graph convolutional neural network model and building the identification library are explained in detail in the following embodiments and will not be elaborated here.

[0111] S202: If the identifier of the target first object does not exist in the identifier library of the first object, generating a target first object feature vector of the target first object based on the graph convolutional neural network according to the attribute information of the target first object.

[0112] Optionally, in a possible implementation, if the identifier of the target first object exists in the identifier library of the first object, the identifier of the second object corresponding to the target first object is determined as the identifier of the second object to be recommended corresponding to the target first object.

[0113] In this step, when the identifier of the target first object exists in the identifier library of the first object, the identifier of the second object to be recommended corresponding to the target first object can be determined directly based on the identifier library of the first object, thereby utilizing the identifier library to realize rapid acquisition of personalized recommendation information, thereby achieving the technical effect of improving the efficiency of personalized recommendation.

[0114] Exemplarily, if the first target object is a product, the attribute information of the first target object includes: product category, product amount, number of product transactions within a preset period, number of product clicks within a preset period, product launch time, and product purpose. If the first target object is a user, the attribute information of the first target object includes: age, education level, user level, monthly income amount, monthly expenditure amount, average daily assets, debit card binding information, credit card binding information, number of transactions within a preset period, and transaction amount.

[0115] Optionally, a method for generating a first object feature vector of a target first object based on attribute information of the target first object based on a graph convolutional neural network may be: generating an initial first object feature vector based on the attribute information of the target first object, and performing a graph convolution calculation on the initial first object feature vector based on the graph convolutional neural network to obtain the target first object feature vector.

[0116] Exemplarily, the initial feature extraction method for the first target object may be:

[0117] Assume that the first target object is product B, and the attribute information of the product includes: product category, product amount, number of transactions in a week, number of clicks in a week, product launch time, and product purpose. Use hot encoding to convert each attribute in the product attribute information into coded data, so as to obtain the initial feature vector of the product.

[0118] S203: Calculate a first similarity between the target first object feature vector and each first object feature in the identification library of the first object.

[0119] In this step, the first similarity between the target first object feature vector and each first object feature can be calculated by calculating the cosine similarity between the two vectors or calculating the Euclidean distance between the two vectors. The calculation method of the first similarity is not limited in this application, and any suitable calculation method can be used to calculate the similarity between the two vectors.

[0120] Exemplarily, the method of using the cosine similarity calculation method to calculate the similarity between the target first object feature vector and the first object feature may be:

[0121] S2031. Calculate the dot product of the target first object feature vector and the first object feature.

[0122] S2032. Calculate the norm of each vector respectively.

[0123] S2033. Calculate and determine the cosine similarity between the target first object feature vector and the first object feature based on the dot product and the norm.

[0124] S204: Determine the identifier of the second object corresponding to the first object feature whose first similarity is greater than the first preset similarity as the identifier of the to-be-recommended second object corresponding to the target first object.

[0125] In this step, there may be one or more first object features whose first similarity is greater than the first preset similarity, and there may be multiple corresponding second object identifiers.

[0126] Exemplarily, assume that there are N first objects in the identification library, the number of identifications of the second object corresponding to each first object is the same, and there are M identifications of the second object corresponding to each first object, and the total number of identifications of all second objects in the identification library is F. There are two first object features whose first similarity is currently calculated to be greater than the first preset similarity, then the identification of the second object to be recommended corresponding to the target first object is the intersection of the second object identifications corresponding to these two first objects, and the number of second object identifications in the intersection is S, and S is greater than or equal to M, and S is less than or equal to F. Among them, N, M, F and S are all positive integers greater than 0.

[0127] The object recommendation method provided in this embodiment determines whether the identifier of the target first object currently used for recommendation exists in the identifier library of the first object based on the method; when the identifier of the target first object does not exist in the identifier library of the first object, a target first object feature vector is generated based on the attribute information of the target first object, and the first similarity between the target first object feature vector and each first object feature in the identifier library is calculated, and the identifier of the second object corresponding to the first object feature whose first similarity is greater than the first preset similarity is determined as the identifier of the second object to be recommended corresponding to the target first object. The identifier library is used to calculate the products or users of new users or new products in the identifier library whose similarity is greater than the first preset similarity, and the corresponding recommended user list is determined according to the existing products in the identifier library, or the corresponding recommended product list is determined according to the existing users, so as to obtain the recommended user list corresponding to the new product or the recommended product list corresponding to the new user, and realize personalized recommendation for new products or new users. Thereby, the technical effect of improving the accuracy of personalized recommendation is achieved.

[0128] Figure 3 This is a flow chart of the second embodiment of the object recommendation method provided by this application, such as Figure 3 As shown, the method includes:

[0129] S301. Obtain product information of each sample product and user information of each sample user.

[0130] In this step, the product information of the sample product includes the corresponding identifier of the sample product, product attribute sub-information, the identifier of the sample user who has acquired the sample product, and the number of clicks; the user information of the sample user includes the corresponding identifier of the sample user, user attribute sub-information, the identifier of the sample product acquired by the sample user, and the identifier of the sample product clicked by the sample user.

[0131] Optionally, a possible implementation method of obtaining product information and user information is:

[0132] S3011. Determine a target application for collecting sample data.

[0133] S3012. Send data based on user behavior data in the target application and product.

[0134] S3013. Record user behavior data and product distribution data in the tracking log, and mark the recording time of each tracking log.

[0135] S3014. When it is necessary to obtain product information of sample products and user information of sample users, collect the tracking logs corresponding to each sample product and each sample user, parse the tracking logs and obtain product information and user information.

[0136] The method of parsing the tracking log may be: extracting product information using product identification, and extracting user information using user representation.

[0137] S302: Perform model training based on the product information of each sample product and the user information of each sample user to generate a graph convolutional neural network.

[0138] Optionally, a possible implementation of graph convolutional neural network generation is:

[0139] S3021. Construct an interaction matrix based on the product information of each sample product and the user information of each sample user. The interaction matrix is ​​used to indicate whether each sample user obtains each sample product.

[0140] For example, the method of establishing the interaction matrix can refer to Formula 1:

[0141]

[0142] Where R ij It refers to the formula representation of the interaction matrix between sample users and sample products, i refers to the i-th sample user, and j refers to the j-th sample product.

[0143] S3022. Generate an adjacency matrix according to the interaction matrix and the transposed matrix corresponding to the interaction matrix.

[0144] For example, the method of establishing the adjacency matrix can refer to Formula 2:

[0145]

[0146] Among them, R refers to the interaction matrix between sample products and sample users, R T It refers to the transposed matrix of the interaction matrix, and A refers to the adjacency matrix composed of sample users and sample products.

[0147] S3023. Generate initial sample product features corresponding to the sample product based on the product information of each sample product.

[0148] In this step, a possible implementation method of generating initial sample product features based on the product information of the sample product is:

[0149] b1. Perform data processing based on product information, supplement missing values ​​in product information, and standardize the formats of multiple types of information in product information.

[0150] b2. Convert the product attribute sub-information and click volume in the product information into numerical features using a preset encoding method.

[0151] b3. Combine the numerical features, product identifiers, and click volumes corresponding to the multiple attributes to generate initial sample product features for each sample product.

[0152] S3024. Generate initial sample user features corresponding to each sample user based on the user information of each sample user.

[0153] In this step, a possible implementation method of generating initial sample features based on the user information of the sample user is:

[0154] c1. Perform data processing based on user information, supplement missing values ​​in user information, and standardize the formats of multiple types of information in user information.

[0155] c2. Convert the user attribute sub-information in the user information into numerical features using a preset encoding method.

[0156] c3. Combine the numerical features corresponding to the multiple attributes and the user ID to generate the initial sample user features for each sample user.

[0157] It should be noted that the preset encoding method mentioned in S3023 and S3024 can be all-hot encoding, or any encoding method that can extract data features.

[0158] S3025. Perform graph convolution operation on the initial sample product features based on the adjacency matrix to obtain sample product features corresponding to the sample product nodes.

[0159] S3026. Perform graph convolution operation on the initial sample user features based on the adjacency matrix to obtain sample user features corresponding to the sample user nodes.

[0160] In S3025 to S3026, the method of using graph convolution operation to obtain the sample product features and sample user features corresponding to the sample product nodes and the sample user nodes respectively may be:

[0161] In this step, the method of performing graph convolution operation on the initial sample product features based on the adjacency matrix is ​​as follows:

[0162] d1. Use the diagonal matrix to iterate the adjacency matrix to obtain an iterative matrix.

[0163] The calculation of the iteration matrix is ​​shown in Formula 3:

[0164]

[0165] Where D is a (M+N)*(M+N) diagonal matrix, the elements on the diagonal are the sum of the elements in the i-th row of matrix A, M refers to the number of sample users, N refers to the number of sample products; A refers to the adjacency matrix of sample products and sample users; A' refers to the iteration matrix.

[0166] d2. Perform graph convolution propagation calculation on the initial sample product features and the initial sample user features based on the iteration matrix. The propagation calculation process is shown in Formula 4:

[0167]

[0168] Among them, k refers to the number of layers of propagation calculation, A' refers to the iteration matrix, and E refers to the concatenated features of the initial sample product features and the initial sample user features.

[0169] d3. Perform weighted average calculation based on the splicing features of each propagation layer to obtain the splicing feature results of each sample user and sample product, as shown in Formula 5:

[0170]

[0171] Where E' is the calculated concatenated features of each sample user and sample product; k is the number of layers of propagation calculation; a j =1 / (k+1),j=1,……,k,a j Refers to the weight of a propagation layer.

[0172] d4. Split based on the concatenated features of each sample user and sample product to determine the sample user features of the sample user node and the sample product features of the sample product node.

[0173] S3027. Perform model training based on sample product features and sample user features until a training cutoff condition is met to obtain a graph convolutional neural network.

[0174] In this step, the model training is performed based on the sample product features and sample user features, and the method of determining the graph convolutional neural network can be:

[0175] e1. Calculate and obtain the similarity score between the sample product features and the sample user features.

[0176] e2. Calculate the loss value based on the similarity score. When the loss value is lower than the preset loss value, stop the iterative training and determine the graph convolutional neural network.

[0177] The loss value calculation function may be a Bayesian personalized ranking loss function. It should be noted that this example does not impose specific restrictions on the loss calculation function, and in actual use, a suitable loss calculation function may be selected based on different data types.

[0178] In S3021 to S3027 of this embodiment, user features and product features are fused by using a graph convolutional neural network training method, and the fused features are propagated using an adjacency matrix to ultimately obtain a feature representation of the user node and a feature representation of the product node. The correlation between user features and product features is enhanced through an interaction matrix and an adjacency matrix, and the representation of user features and product features is optimized through propagation calculations, thereby improving the accuracy of similarity calculations between user features and product features, thereby achieving the effect of improving the accuracy of personalized recommendations.

[0179] In S301 to S302 of this embodiment, the graph convolutional neural network is trained using the product information of the sample products and the user information of the sample users, thereby obtaining a graph convolutional neural network that can be used to calculate sample features, product features, and the similarity between samples and products, thereby achieving feature calculations at user nodes and product nodes, providing a basis for subsequent personalized recommendations, and facilitating the creation of a product identification library or a user identification library.

[0180] S303: Process the user's identification and user attribute information through a graph convolutional neural network to generate user features corresponding to the user.

[0181] For example, there is a user B, whose user ID is B. The user attribute information of the user includes: age, education, user level, monthly income amount, monthly expenditure amount, average daily assets, debit card binding information, credit card binding information, number of transactions within a preset period, and transaction amount. Initial user feature extraction is performed based on the user attribute information to obtain the initial user features of user B; the user representation B and the user features are input into the graph convolutional neural network, and the user features of user B are obtained through multiple propagation calculations.

[0182] It should be noted that the graph convolutional neural network contains feature representations of user nodes and feature representations of product nodes.

[0183] S304: Process the product identification and product attribute information through a graph convolutional neural network to generate product features corresponding to the product.

[0184] For example, there is a product C, the product identifier of which is C, and the product attribute information of which includes: product category, product amount, number of product transactions within a preset period, number of product clicks within a preset period, product launch time, and product purpose. Initial product features are extracted based on the product attribute information to obtain the initial product features of product C; the product identifier C and the product features are input into the graph convolutional neural network, and the product features of product C are obtained through multiple propagation calculations.

[0185] S305: For each user, calculate the second similarity between the user feature corresponding to the user and each product feature.

[0186] In this step, the similarity calculation between users and products can be performed using a graph convolutional neural network model, or other similarity calculation methods, such as the Euclidean distance calculation method or the Manhattan distance calculation method.

[0187] For example, when the Euclidean distance calculation method is used, the similarity calculation method between the user features and the product features can be:

[0188] S3051. Calculate the difference between each corresponding element of two features to obtain a difference vector.

[0189] S3052. Calculate the sum of the squares of each element in the difference vector.

[0190] S3053. Calculate the square root corresponding to the sum of squares, and determine it as the similarity between the user feature and the product feature.

[0191] S306: Determine the identifier of the product corresponding to the product feature whose second similarity is greater than the second preset similarity as the identifier of the product corresponding to the identifier of the user.

[0192] In this step, each user's ID may correspond to one or more product IDs, and the product IDs corresponding to multiple user IDs may overlap. The product IDs corresponding to each user's ID may be determined as a product ID list corresponding to the user, and one product ID may exist in the product ID lists of multiple user IDs.

[0193] S307: Construct an identification library of the first object and an identification library of the second object according to the correspondence between the identification of the product and the identification of the user, the identification of the product, the product attribute information, the product characteristics, the identification of the user, the user attribute information and the user characteristics.

[0194] For example, the first object refers to a user, and the second object refers to a product; the identification library of the first object includes: the user's identification, the user's attribute information, the user's characteristics, and the product identification corresponding to each user's identification. The identification library of the second object includes: the product's identification, the product's attribute information, the product's characteristics, and the user identification corresponding to each product's identification.

[0195] In this embodiment, a graph convolutional neural network is used to calculate the features corresponding to users and products for personalized recommendations, and a user identification library and a product identification library are created based on the user features and product features, thereby obtaining an identification library that can quickly acquire recommendation information, thereby improving the efficiency of personalized recommendations.

[0196] Figure 4 A schematic diagram of the structure of the object recommendation device provided in this application, such as Figure 4 As shown, the object recommendation device 40 provided in this embodiment includes:

[0197] A first processing module 401 is used to determine whether the identifier of the target first object exists in an identifier library of the first object, wherein the identifier library of the first object stores identifiers corresponding to each first object, first object features, and second object identifiers, wherein the first object features are generated by a graph convolutional neural network according to attribute information of the first object;

[0198] A second processing module 402 is used to generate a target first object feature vector of the target first object based on the graph convolutional neural network according to the attribute information of the target first object when the identifier of the target first object does not exist in the identifier library of the first object;

[0199] The third processing module 403 is used to calculate the first similarity between the target first object feature vector and each first object feature in the identification library of the first object;

[0200] The fourth processing module 404 is configured to determine the identifier of the second object corresponding to the first object feature whose first similarity is greater than the first preset similarity as the identifier of the second object to be recommended corresponding to the target first object.

[0201] In a possible implementation, when the first object is a user, the second object is a product; when the first object is an object, the first object is a product.

[0202] In one possible implementation, the second processing module 402 is further configured to:

[0203] When the identifier of the target first object exists in the identifier library of the first object, the identifier of the second object corresponding to the target first object is determined as the identifier of the second object to be recommended corresponding to the target first object.

[0204] In a possible implementation, the first processing module 401 is further configured to:

[0205] When the identifier of the target second object does not exist in the identifier library of the second object, recommending the target second object to the plurality of test first objects;

[0206] Obtaining acquisition behavior information of each test first object on the target second object;

[0207] According to the acquisition behavior information of each test first object, the test first object having an acquisition behavior on the target second object is determined as the target first object.

[0208] In a possible implementation, the first processing module 401 is further configured to:

[0209] The user's identification and user attribute information are processed through a graph convolutional neural network to generate the user's corresponding user features;

[0210] The product identification and product attribute information are processed through the graph convolutional neural network to generate the corresponding product features;

[0211] For each user, calculating the second similarity between the user feature corresponding to the user and each product feature;

[0212] Determine the identifier of the product corresponding to the product feature whose second similarity is greater than the second preset similarity as the identifier of the product corresponding to the identifier of the user;

[0213] According to the correspondence between the product identification and the user identification, the product identification, product attribute information, product characteristics, the user identification, user attribute information and user characteristics, an identification library of the first object and an identification library of the second object are constructed.

[0214] In a possible implementation, the device further includes an acquisition module 405, which is used to:

[0215] Obtain product information of each sample product and user information of each sample user, wherein the product information of the sample product includes an identifier corresponding to the sample product, product attribute sub-information, an identifier of the sample user who has acquired the sample product, and a click volume; the user information of the sample user includes an identifier corresponding to the sample user, user attribute sub-information, an identifier of the sample product acquired by the sample user, and an identifier of the sample product clicked by the sample user;

[0216] The model is trained based on the product information of each sample product and the user information of each sample user to generate a graph convolutional neural network.

[0217] In a possible implementation, the acquisition module 405 is further configured to:

[0218] According to the product information of each sample product and the user information of each sample user, an interaction matrix is ​​constructed, and the interaction matrix is ​​used to indicate whether each sample user obtains each sample product;

[0219] Generate an adjacency matrix according to the interaction matrix and the transposed matrix corresponding to the interaction matrix;

[0220] Generate initial sample product features corresponding to the sample product according to the product information of each sample product;

[0221] Generate initial sample user features corresponding to each sample user based on the user information of each sample user;

[0222] Perform graph convolution operations on the initial sample product features based on the adjacency matrix to obtain the sample product features corresponding to the sample product nodes;

[0223] Perform graph convolution operation on the initial sample user features based on the adjacency matrix to obtain the sample user features corresponding to the sample user nodes;

[0224] The model is trained based on sample product features and sample user features until the training cutoff condition is met to obtain a graph convolutional neural network.

[0225] The object recommendation device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be described in detail here.

[0226] Figure 5 This is a schematic diagram of the structure of the electronic device provided in this application. Figure 5 As shown, the electronic device 50 provided in this embodiment includes: at least one processor 501 and a memory 502. Optionally, the device further includes a communication component 503. The processor 501, the memory 502 and the communication component 503 are connected via a bus 504.

[0227] In a specific implementation process, at least one processor 501 executes the computer-executable instructions stored in the memory 502, so that at least one processor 501 executes the above method.

[0228] The specific implementation process of the processor 501 can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.

[0229] In the above embodiments, it should be understood that the processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the invention can be directly implemented as a hardware processor, or can be implemented by a combination of hardware and software modules in the processor.

[0230] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (NVM), such as at least one disk storage.

[0231] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of this application is not limited to only one bus or one type of bus.

[0232] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0233] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0234] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0235] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0236] The division of units is only a logical function division, and there may be other divisions in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0237] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0238] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0239] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0240] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0241] Finally, it should be noted that those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention, which follow the general principles of the present invention and include common knowledge or customary technical means in the art not disclosed by the present invention, are not limited to the precise structure described above and shown in the drawings, and may be modified and changed in various ways without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. An object recommendation method, characterized in that: include: Determine whether the identifier of the target first object exists in an identifier library of the first object, wherein the identifier library of the first object stores identifiers corresponding to each first object, first object features, and identifiers of second objects, wherein the first object features are generated by a graph convolutional neural network according to attribute information of the first object; If the identifier of the target first object does not exist in the identifier library of the first object, generating a target first object feature vector of the target first object based on the graph convolutional neural network according to the attribute information of the target first object; Calculating a first similarity between the target first object feature vector and each first object feature in the identification library of the first object; The identifier of the second object corresponding to the first object feature whose first similarity is greater than the first preset similarity is determined as the identifier of the to-be-recommended second object corresponding to the target first object.

2. The method according to claim 1, characterized in that When the first object is a user, the second object is a product; when the first object is an object, the first object is a product.

3. The method according to claim 1 or 2, characterized in that: The method further comprises: If the identifier of the target first object exists in the identifier library of the first object, the identifier of the second object corresponding to the target first object is determined as the identifier of the second object to be recommended corresponding to the target first object.

4. The method according to claim 1 or 2, characterized in that: Before determining whether the identifier of the target first object exists in the identifier library of the first object, the method further includes: If the identifier of the target second object does not exist in the identifier library of the second object, recommending the target second object to the plurality of test first objects; Acquire acquisition behavior information of each test first object on the target second object; According to the acquisition behavior information of each test first object, the test first object having an acquisition behavior on the target second object is determined as the target first object.

5. The method according to claim 2, characterized in that: Before determining whether the identifier of the target first object exists in the identifier library of the first object, the method further includes: Processing the user's identification and user attribute information through the graph convolutional neural network to generate user features corresponding to the user; Processing the product identification and product attribute information through the graph convolutional neural network to generate product features corresponding to the product; For each user, calculating a second similarity between a user feature corresponding to the user and each product feature; Determine the identifier of the product corresponding to the product feature whose second similarity is greater than a second preset similarity as the identifier of the product corresponding to the identifier of the user; According to the correspondence between the product identification and the user identification, the product identification, product attribute information, product characteristics, the user identification, user attribute information and user characteristics, the identification library of the first object and the identification library of the second object are constructed.

6. The method according to claim 2, characterized in that Before determining whether the identifier of the target first object exists in the identifier library of the first object, the method further includes: Acquire product information of each sample product and user information of each sample user, wherein the product information of the sample product includes an identifier corresponding to the sample product, product attribute sub-information, an identifier of a sample user who has acquired the sample product, and a click volume; the user information of the sample user includes an identifier corresponding to the sample user, user attribute sub-information, an identifier of a sample product acquired by the sample user, and an identifier of a sample product clicked by the sample user; Model training is performed based on the product information of each sample product and the user information of each sample user to generate the graph convolutional neural network.

7. The method according to claim 6, characterized in that The performing model training based on the product information of each sample product and the user information of each sample user to generate the graph convolutional neural network includes: Constructing an interaction matrix based on the product information of each sample product and the user information of each sample user, wherein the interaction matrix is ​​used to indicate whether each sample user obtains each sample product; Generate an adjacency matrix according to the interaction matrix and a transposed matrix corresponding to the interaction matrix; Generating initial sample product features corresponding to each sample product according to the product information of the sample product; Generating initial sample user features corresponding to each sample user according to the user information of the sample user; Performing a graph convolution operation on the initial sample product features based on the adjacency matrix to obtain sample product features corresponding to sample product nodes; Performing a graph convolution operation on the initial sample user features based on the adjacency matrix to obtain sample user features corresponding to sample user nodes; Model training is performed based on the sample product features and the sample user features until a training cutoff condition is met, thereby obtaining the graph convolutional neural network.

8. An object recommendation device, characterized in that: include: A first processing module is used to determine whether the identifier of the target first object exists in an identifier library of the first object, wherein the identifier library of the first object stores identifiers corresponding to each first object, first object features, and identifiers of second objects, wherein the first object features are generated by a graph convolutional neural network according to attribute information of the first object; A second processing module is configured to generate a target first object feature vector of the target first object based on the graph convolutional neural network according to the attribute information of the target first object when the identifier of the target first object does not exist in the identifier library of the first object; A third processing module, configured to calculate a first similarity between the target first object feature vector and each first object feature in the identification library of the first object; The fourth processing module is used to determine the identifier of the second object corresponding to the first object feature whose first similarity is greater than the first preset similarity as the identifier of the second object to be recommended corresponding to the target first object.

9. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed by a processor.

11. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 7 when being executed by a processor.

Images