Commodity recommendation method and device

Abstract

The present disclosure relates to the technical field of electronic commerce, and provides a commodity recommendation method and device. The method comprises: obtaining a training data set; obtaining a first probability that a user is a new user in a unit time, a second probability that user information is missing, and a third probability that a user is a new commodity; performing data enhancement processing on the training data set based on the first probability that a user is a new user in a unit time, the second probability that user information is missing, and the third probability that a user is a new commodity; extracting data features of each group of training data of the training data set after data enhancement processing, wherein the data features of each group of training data comprise: a user identification code feature, a user information feature, a commodity identification code feature, a commodity information feature, and an interaction feature; training a commodity recommendation model using the extracted multiple groups of data features; and providing a commodity recommendation service for a target user using the trained commodity recommendation model.

Description

Technical Field

[0001] This disclosure relates to the field of e-commerce technology, and in particular to a product recommendation method and apparatus. Background Technology

[0002] In online e-commerce, recommending suitable products or services to users based on their personalized characteristics has always been a crucial challenge. In relatively low-frequency scenarios, such as online advertising, online rentals, and online home purchases, the large number of newly registered users, low overall user stickiness, low usage frequency, numerous new product listings, and frequent product delistings and relistings mean that the identifiers for new users and new products are often not trained or have only been minimally trained. Furthermore, the information of new users is often incomplete, which can negatively impact training. Therefore, traditional product recommendation models struggle to learn effectively from newly registered users or newly listed products, exhibiting weak generalization ability and limiting their ability to improve user conversion rates.

[0003] In realizing the concept disclosed herein, the inventors discovered at least the following technical problems in the related technologies: the product recommendation model has a weak generalization ability for new users or new products. Summary of the Invention

[0004] In view of this, embodiments of the present disclosure provide a product recommendation method, apparatus, electronic device, and computer-readable storage medium to solve the problem that the product recommendation model in the prior art has weak generalization ability for new users or new products.

[0005] A first aspect of this disclosure provides a product recommendation method, comprising: acquiring a training dataset, wherein the training dataset includes: multiple sets of training data, each set of training data including: user identifier code, user information, product identifier code, product information, and interaction information regarding user-product interaction behavior; acquiring a first probability that a user is a new user, a second probability that user information is missing, and a third probability that a user is a new product within a unit of time; performing data augmentation processing on the training dataset based on the first probability that a user is a new user, the second probability that user information is missing, and the third probability that a user is a new product within a unit of time; extracting data features of each set of training data from the data augmentation-processed training dataset, wherein the data features of each set of training data include: user identifier code features, user information features, product identifier code features, product information features, and interaction features; training a product recommendation model using the extracted multiple sets of data features; and providing product recommendation services to target users using the trained product recommendation model.

[0006] A second aspect of this disclosure provides a product recommendation device, comprising: a first acquisition module configured to acquire a training dataset, wherein the training dataset includes: multiple sets of training data, each set of training data including: a user identifier code, user information, a product identifier code, product information, and interaction information regarding user-product interaction behavior; a second acquisition module configured to acquire a first probability that a user is a new user within a unit of time, a second probability that user information is missing, and a third probability that a user is a new product; a processing module configured to perform data augmentation processing on the training dataset based on the first probability that a user is a new user within a unit of time, the second probability that user information is missing, and the third probability that a user is a new product; an extraction module configured to extract data features from each set of training data in the data augmentation-processed training dataset, wherein the data features of each set of training data include: user identifier code features, user information features, product identifier code features, product information features, and interaction features; a training module configured to train a product recommendation model using the extracted multiple sets of data features; and a service module configured to provide product recommendation services to target users using the trained product recommendation model.

[0007] A third aspect of this disclosure provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0008] A fourth aspect of this disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0009] The beneficial effects of this disclosure embodiment compared with the prior art are as follows: This disclosure embodiment obtains a training dataset, which includes multiple sets of training data, each set including user identification code, user information, product identification code, product information, and interaction information regarding user-product interaction behavior; obtains a first probability that a user is a new user, a second probability that user information is missing, and a third probability that a user is a new product within a unit of time; performs data augmentation processing on the training dataset based on the first probability that a user is a new user, the second probability that user information is missing, and the third probability that a user is a new product within a unit of time; extracts data features from each set of training data in the data augmentation dataset, where each set of training data features includes user identification code features, user information features, product identification code features, product information features, and interaction features; trains a product recommendation model using the extracted multiple sets of data features; and provides product recommendation services to target users using the trained product recommendation model. Therefore, by adopting the above technical means, the problem of weak generalization ability of product recommendation models for new users or new products in the prior art can be solved, thereby improving the generalization ability of product recommendation models for new users or new products. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram illustrating an application scenario of an embodiment of this disclosure;

[0012] Figure 2 This is a flowchart illustrating a product recommendation method provided in an embodiment of this disclosure;

[0013] Figure 3 This is a schematic diagram of the structure of a product recommendation device provided in an embodiment of this disclosure;

[0014] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0015] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, so as to provide a thorough understanding of the embodiments of this disclosure. However, those skilled in the art will understand that this disclosure may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this disclosure with unnecessary detail.

[0016] A product recommendation method and apparatus according to an embodiment of the present disclosure will now be described in detail with reference to the accompanying drawings.

[0017] Figure 1 This is a schematic diagram illustrating an application scenario of an embodiment of this disclosure. The application scenario may include terminal devices 101, 102, and 103, server 104, and network 105.

[0018] Terminal devices 101, 102, and 103 can be hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices with displays that support communication with server 104, including but not limited to smartphones, tablets, laptops, and desktop computers. When terminal devices 101, 102, and 103 are software, they can be installed in the aforementioned electronic devices. Terminal devices 101, 102, and 103 can be implemented as multiple software programs or software modules, or as a single software program or software module; this disclosure does not impose any limitations on this. Furthermore, various applications can be installed on terminal devices 101, 102, and 103, such as data processing applications, instant messaging tools, social platform software, search applications, shopping applications, etc.

[0019] Server 104 can be a server that provides various services, such as a backend server that receives requests sent by terminal devices with which it has established communication connections. This backend server can receive and analyze the requests sent by the terminal devices and generate processing results. Server 104 can be a single server, a server cluster consisting of several servers, or a cloud computing service center. This embodiment of the disclosure does not impose any limitations on these aspects.

[0020] It should be noted that server 104 can be either hardware or software. When server 104 is hardware, it can be various electronic devices that provide various services to terminal devices 101, 102, and 103. When server 104 is software, it can be multiple software programs or software modules that provide various services to terminal devices 101, 102, and 103, or it can be a single software program or software module that provides various services to terminal devices 101, 102, and 103. This disclosure does not limit the scope of the embodiments.

[0021] Network 105 can be a wired network using coaxial cable, twisted pair, and fiber optic connection, or it can be a wireless network that enables interconnection of various communication devices without wiring, such as Bluetooth, Near Field Communication (NFC), and Infrared. This disclosure does not limit the scope of the network.

[0022] Users can establish a communication connection with server 104 via network 105 through terminal devices 101, 102, and 103 to receive or send information, etc. It should be noted that the specific types, quantities, and combinations of terminal devices 101, 102, and 103, server 104, and network 105 can be adjusted according to the actual needs of the application scenario, and this disclosure embodiment does not impose any limitations on this.

[0023] Figure 2 This is a flowchart illustrating a product recommendation method provided in an embodiment of this disclosure. Figure 2 The product recommendation method can be provided by Figure 1 The computer or server, or the software on the computer or server, executes the command. For example... Figure 2 As shown, the product recommendation method includes:

[0024] S201, Obtain the training dataset, which includes: multiple sets of training data, each set of training data including: user identification code, user information, product identification code, product information, and interaction information about the user's interaction with the product;

[0025] S202, obtain the first probability that a user is a new user, the second probability that user information is missing, and the third probability that a user is a new product within a unit of time;

[0026] S203, based on the first probability that a user is a new user, the second probability that user information is missing, and the third probability that a user is a new product within a unit of time, data augmentation processing is performed on the training dataset;

[0027] S204, Extract the data features of each training data set in the data augmentation training dataset. The data features of each training data set include: user identification code features, user information features, product identification code features, product information features, and interaction features.

[0028] S205, Use the extracted features from multiple sets of data to train a product recommendation model;

[0029] S206 utilizes a trained product recommendation model to provide product recommendation services to target users.

[0030] This disclosure applies to the field of product recommendation. Products include: electronic goods, online advertisements, online rental room resources, and online home purchase room resources, etc. A user identifier is used to identify the user; a product identifier is used to identify the product; user information includes: the user's mobile phone model or computer model, or other electronic device model used by the user, and the user's location; product information includes: product price and sales volume, etc.; interaction between the product and the user refers to actions such as clicking, saving, and browsing the product's webpage, link, or window. A set of training data can be collected between a user and a product with which that user has interaction. The product recommendation model can be any common neural network model, such as a deep factorizer, deep neural network, convolutional neural network, or Transformer model.

[0031] New users refer to the most recently registered users. New users often lack purchase history (while existing users usually have purchase history, which helps with product recommendations), and they often lack a user identifier. New products are the most recently listed products (existing products often have sales history, which helps with product recommendations), and they often lack a product identifier. New users often register with incomplete information, resulting in frequent instances of missing user data. Due to these missing information, the product recommendation model is not sufficiently trained, leading to weak generalization ability of traditional product recommendation models for new users or new products.

[0032] According to the technical solution provided in this disclosure, a training dataset is obtained, comprising: multiple sets of training data, each set including: user identifier code, user information, product identifier code, product information, and interaction information regarding user-product interaction behavior; a first probability that a user is a new user, a second probability that user information is missing, and a third probability that a user is a new product are obtained within a unit of time; data augmentation processing is performed on the training dataset based on the first probability that a user is a new user, the second probability that user information is missing, and the third probability that a user is a new product within a unit of time; data features of each set of training data in the data augmentation dataset are extracted, wherein the data features of each set of training data include: user identifier code features, user information features, product identifier code features, product information features, and interaction features; a product recommendation model is trained using the extracted multiple sets of data features; and a product recommendation service is provided to the target user using the trained product recommendation model. Therefore, by adopting the above technical means, the problem of weak generalization ability of product recommendation models for new users or new products in the prior art can be solved, thereby improving the generalization ability of product recommendation models for new users or new products.

[0033] In step S203, based on the first probability that a user is a new user, the second probability that user information is missing, and the third probability that a user is a new product within a unit of time, data augmentation processing is performed on the training dataset, including: sampling the training dataset multiple times as follows: retaining the original value of the user identifier code in the training dataset according to the first probability, and setting the user identifier code that does not retain the original value as a mask; retaining the original value of the user information in the training dataset according to the second probability, and setting the user information that does not retain the original value as a mask; retaining the original value of the product identifier code in the training dataset according to the third probability, and setting the product identifier code that does not retain the original value as a mask.

[0034] The probability of a user being a new user, the probability of missing user information, and the probability of a user being a new product within a unit of time are results obtained by the e-commerce platform through processing historical data. The user identifier code (without retaining its original value) is masked; this portion of the user identifier code is used to represent a new user (new user with missing user identifier code). The user information (without retaining its original value) is masked; this portion of the user information represents a new user with missing user information. The product identifier code (without retaining its original value) is masked; this portion of the product identifier code is used to represent a new product (new product with missing product identifier code).

[0035] It should be noted that retaining the original value of the user identifier code in the training dataset according to the first probability means retaining the original value of the user identifier codes in the training dataset according to a proportion of the values ​​minus the first probability. The processing of user information and product identifier codes is similar.

[0036] For example: If 30% of the daily active users on an e-commerce platform are new users (per day), then the first probability is 30%; if 20% of the daily active users' information, such as gender, is unknown, then the second probability is 20%; if 40% of the products listed on an e-commerce platform each day are new products, then the third probability is 40%. The training dataset is sampled as follows:

[0037] For user identifiers in the training dataset, the original value is retained with a 70% probability, and all other data is masked. For user genders in the training dataset, the original value is retained with an 80% probability, and all other data is masked. For product identifiers in the training dataset, the original value is retained with a 60% probability, and all other data is masked.

[0038] In step S204, data features of each training data set in the data augmentation dataset are extracted, including: extracting user identification code features from user identification codes; extracting user information features from user information; extracting product identification code features from product identification codes; extracting product information features from product information; and extracting interaction features from the interaction information of user-product interaction behavior.

[0039] In step S204, data features of each training data set in the data augmentation dataset are extracted, including: processing the quantity-type data in each training data set using a mapping method to obtain the data features of each training data set; and / or processing the non-quantity-type data in each training data set using an encoding method to obtain the data features of each training data set.

[0040] For continuous information, such as user identification codes, mapping methods are often used to represent its features; for discontinuous information, such as a user's gender in user information, encoding methods are often used to represent its features. Encoding methods include one-hot encoding, integer encoding, and word embedding representation.

[0041] In step S205, the product recommendation model is trained using the extracted multiple sets of data features, including: obtaining information corresponding to each set of training data in the training dataset that indicates whether a user has conversion behavior towards a product; labeling each set of data features with tags based on the information corresponding to each set of training data that indicates whether a product corresponding to that set of training data is recommended to the user corresponding to that set of training data; and training the product recommendation model using the multiple sets of data features labeled with corresponding tags.

[0042] Users interact with product webpages, links, or windows by clicking, adding to favorites, or browsing. Whether a user will purchase or pre-order the product is uncertain given these interactions. If a user does purchase or pre-order the product, then the user has made a conversion to the product. The data features of the training data collected between the user and the product are labeled as "recommendation" or "recommend the product to the user".

[0043] In step S205, a product recommendation model is trained using the extracted multiple sets of data features, including: a product recommendation model, comprising: a feature processing network and a recommendation judgment network; the product recommendation model is trained using the extracted multiple sets of data features, such that the feature processing network learns and stores the correspondence between data features and joint features, and the recommendation judgment network learns and stores the correspondence between joint features and whether a product is recommended, wherein each set of data features corresponds to a joint feature.

[0044] The feature processing network is used to process each set of data features to obtain the joint features corresponding to each set of data features. That is, the user identification code feature, user information feature, product identification code feature, product information feature and interaction feature in a set of data features are input into the feature processing network, and the output is the joint features corresponding to each set of data features. The recommendation judgment network is used to determine whether the product corresponding to each set of data features should be recommended to the user corresponding to each set of data features based on the joint features corresponding to each set of data features.

[0045] This embodiment of the disclosure processes a set of user identification code features, user information features, product identification code features, product information features, and interaction features to obtain joint features, thereby improving the accuracy of online product recommendation.

[0046] In one optional embodiment, the method includes: masking each data feature in the extracted multiple sets of data features according to a preset probability; and training a product recommendation model using the masked multiple sets of data features.

[0047] To further improve the generalization ability of the product recommendation model, embodiments of this disclosure utilize multiple sets of data features after masking to train the product recommendation model.

[0048] In step S206, a product recommendation service is provided to the target user using a trained product recommendation model, including: obtaining the target user's target user identifier code and target user information; obtaining the target product identifier code and target product information of the target product with which the target user has interactive behavior; and obtaining target interaction information about the interaction behavior between the target user and the target product; extracting target user identifier code features from the target user identifier code; extracting target user information features from the target user information; extracting target product identifier code features from the target product identifier code; extracting target product information features from the target product information; and extracting target interaction features from the target interaction information; and using the trained product recommendation model based on the target user identifier code, target user information features, target product identifier code features, target product information features, and target interaction features, determining whether to recommend the target product to the target user.

[0049] Based on the target user identifier, target user information features, target product identifier, target product information features, and target interaction features, a trained product recommendation model is used to determine whether to recommend the target product to the target user. Alternatively, the target user identifier, target user information features, target product identifier, target product information features, and target interaction features can be input into the trained product recommendation model, and the output of the trained product recommendation model can be used to indicate whether to recommend the target product to the target user.

[0050] A trained product recommendation model can be understood as a product recommendation model after training. There can be multiple target products.

[0051] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0052] The following are embodiments of the apparatus disclosed herein, which can be used to execute embodiments of the method disclosed herein. For details not disclosed in the apparatus embodiments of this disclosure, please refer to the embodiments of the method disclosed herein.

[0053] Figure 3 This is a schematic diagram of a product recommendation device provided in an embodiment of this disclosure. Figure 3 As shown, the product recommendation device includes:

[0054] The first acquisition module 301 is configured to acquire a training dataset, wherein the training dataset includes: multiple sets of training data, each set of training data including: user identification code, user information, product identification code, product information and interaction information about the user's interaction with the product;

[0055] The second acquisition module 302 is configured to acquire the first probability that a user is a new user, the second probability that user information is missing, and the third probability that a user is a new product within a unit of time.

[0056] Processing module 303 is configured to perform data augmentation on the training dataset based on the first probability that a user is a new user, the second probability that user information is missing, and the third probability that a user is a new product within a unit of time.

[0057] The extraction module 304 is configured to extract the data features of each set of training data in the data augmentation training dataset, wherein the data features of each set of training data include: user identification code features, user information features, product identification code features, product information features and interaction features.

[0058] Training module 305 is configured to train a product recommendation model using extracted features from multiple sets of data.

[0059] Service module 306 is configured to provide product recommendation services to target users using a trained product recommendation model.

[0060] This disclosure applies to the field of product recommendation. Products include: electronic goods, online advertisements, online rental room resources, and online home purchase room resources, etc. A user identifier is used to identify the user; a product identifier is used to identify the product; user information includes: the user's mobile phone model or computer model, or other electronic device model used by the user, and the user's location; product information includes: product price and sales volume, etc.; interaction between the product and the user refers to actions such as clicking, saving, and browsing the product's webpage, link, or window. A set of training data can be collected between a user and a product with which that user has interaction. The product recommendation model can be any common neural network model, such as a deep factorizer, deep neural network, convolutional neural network, or Transformer model.

[0061] New users refer to the most recently registered users. New users often lack purchase history (while existing users usually have purchase history, which helps with product recommendations), and they often lack a user identifier. New products are the most recently listed products (existing products often have sales history, which helps with product recommendations), and they often lack a product identifier. New users often register with incomplete information, resulting in frequent instances of missing user data. Due to these missing information, the product recommendation model is not sufficiently trained, leading to weak generalization ability of traditional product recommendation models for new users or new products.

[0062] According to the technical solution provided in this disclosure, a training dataset is obtained, comprising: multiple sets of training data, each set including: user identifier code, user information, product identifier code, product information, and interaction information regarding user-product interaction behavior; a first probability that a user is a new user, a second probability that user information is missing, and a third probability that a user is a new product are obtained within a unit of time; data augmentation processing is performed on the training dataset based on the first probability that a user is a new user, the second probability that user information is missing, and the third probability that a user is a new product within a unit of time; data features of each set of training data in the data augmentation dataset are extracted, wherein the data features of each set of training data include: user identifier code features, user information features, product identifier code features, product information features, and interaction features; a product recommendation model is trained using the extracted multiple sets of data features; and a product recommendation service is provided to the target user using the trained product recommendation model. Therefore, by adopting the above technical means, the problem of weak generalization ability of product recommendation models for new users or new products in the prior art can be solved, thereby improving the generalization ability of product recommendation models for new users or new products.

[0063] Optionally, the processing module 303 is further configured to perform the following multiple samplings on the training dataset: retaining the original value of the user identification code in the training dataset according to a first probability, and setting the user identification code that does not retain the original value as a mask; retaining the original value of the user information in the training dataset according to a second probability, and setting the user information that does not retain the original value as a mask; retaining the original value of the product identification code in the training dataset according to a third probability, and setting the product identification code that does not retain the original value as a mask.

[0064] The probability of a user being a new user, the probability of missing user information, and the probability of a user being a new product within a unit of time are results obtained by the e-commerce platform through processing historical data. The user identifier code (without retaining its original value) is masked; this portion of the user identifier code is used to represent a new user (new user with missing user identifier code). The user information (without retaining its original value) is masked; this portion of the user information represents a new user with missing user information. The product identifier code (without retaining its original value) is masked; this portion of the product identifier code is used to represent a new product (new product with missing product identifier code).

[0065] It should be noted that retaining the original value of the user identifier code in the training dataset according to the first probability means retaining the original value of the user identifier codes in the training dataset according to a proportion of the values ​​minus the first probability. The processing of user information and product identifier codes is similar.

[0066] For example: If 30% of the daily active users on an e-commerce platform are new users (per unit of time is one day), then the first probability is 30%; if 20% of the user information, such as gender, is unknown among the daily active users on the same e-commerce platform, then the second probability is 20%; if 40% of the products listed on the e-commerce platform each day are new products, then the third probability is 40%; the training dataset is sampled as follows:

[0067] For user identifiers in the training dataset, the original value is retained with a 70% probability, and all other data is masked. For user genders in the training dataset, the original value is retained with an 80% probability, and all other data is masked. For product identifiers in the training dataset, the original value is retained with a 60% probability, and all other data is masked.

[0068] Optionally, the extraction module 304 is also configured to extract user identification code features from user identification codes; extract user information features from user information; extract product identification code features from product identification codes; extract product information features from product information; and extract interaction features from interaction information of user-product interaction behavior.

[0069] Optionally, the extraction module 304 is further configured to process the quantity-type data in each group of training data using a mapping method to obtain the data features of each group of training data; and / or process the non-quantity-type data in each group of training data using an encoding method to obtain the data features of each group of training data.

[0070] For continuous information, such as user identification codes, mapping methods are often used to represent its features; for discontinuous information, such as a user's gender in user information, encoding methods are often used to represent its features. Encoding methods include one-hot encoding, integer encoding, and word embedding representation.

[0071] Optionally, the training module 305 is further configured to obtain information corresponding to each set of training data in the training dataset that indicates whether a user has converted to a product; based on the information corresponding to each set of training data that indicates whether a user has converted to a product, label each set of data features with a label, wherein the label corresponding to each set of data features indicates whether the product corresponding to that set of training data is recommended to the user corresponding to that set of training data; and train a product recommendation model using multiple sets of data features labeled with corresponding tags.

[0072] Users interact with product webpages, links, or windows by clicking, adding to favorites, or browsing. Whether a user will purchase or pre-order the product is uncertain given these interactions. If a user does purchase or pre-order the product, then the user has made a conversion to the product. The data features of the training data collected between the user and the product are labeled as "recommendation" or "recommend the product to the user".

[0073] Optionally, the training module 305 is further configured to train a product recommendation model using the extracted multiple sets of data features, such that the feature processing network learns and stores the correspondence between data features and joint features, and the recommendation judgment network learns and stores the correspondence between joint features and whether a product is recommended, wherein each set of data features corresponds to a joint feature.

[0074] The feature processing network is used to process each set of data features to obtain the joint features corresponding to each set of data features. That is, the user identification code feature, user information feature, product identification code feature, product information feature and interaction feature in a set of data features are input into the feature processing network, and the output is the joint features corresponding to each set of data features. The recommendation judgment network is used to determine whether the product corresponding to each set of data features should be recommended to the user corresponding to each set of data features based on the joint features corresponding to each set of data features.

[0075] This embodiment of the disclosure processes a set of user identification code features, user information features, product identification code features, product information features, and interaction features to obtain joint features, thereby improving the accuracy of online product recommendation.

[0076] Optionally, the training module 305 is also configured to mask each of the extracted multiple sets of data features according to a preset probability; and to train a product recommendation model using the masked multiple sets of data features.

[0077] To further improve the generalization ability of the product recommendation model, embodiments of this disclosure utilize multiple sets of data features after masking to train the product recommendation model.

[0078] Optionally, service module 306 is further configured to obtain the target user identifier code and target user information of the target user, the target product identifier code and target product information of the target product with which the target user has interactive behavior, and target interaction information about the interaction behavior between the target user and the target product; extract target user identifier code features from the target user identifier code, target user information features from the target user information, target product identifier code features from the target product identifier code, target product information features from the target product information, and target interaction features from the target interaction information; and, based on the target user identifier code, target user information features, target product identifier code features, target product information features, and target interaction features, use a trained product recommendation model to determine whether to recommend the target product to the target user.

[0079] Based on the target user identifier, target user information features, target product identifier, target product information features, and target interaction features, a trained product recommendation model is used to determine whether to recommend the target product to the target user. Alternatively, the target user identifier, target user information features, target product identifier, target product information features, and target interaction features can be input into the trained product recommendation model, and the output of the trained product recommendation model can be used to indicate whether to recommend the target product to the target user.

[0080] A trained product recommendation model can be understood as a product recommendation model after training. There can be multiple target products.

[0081] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this disclosure.

[0082] Figure 4 This is a schematic diagram of the electronic device 4 provided in an embodiment of this disclosure. Figure 4As shown, the electronic device 4 of this embodiment includes a processor 401, a memory 402, and a computer program 403 stored in the memory 402 and executable on the processor 401. When the processor 401 executes the computer program 403, it implements the steps in the various method embodiments described above. Alternatively, when the processor 401 executes the computer program 403, it implements the functions of each module / unit in the various device embodiments described above.

[0083] Electronic device 4 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 4 may include, but is not limited to, processor 401 and memory 402. Those skilled in the art will understand that... Figure 4 This is merely an example of electronic device 4 and does not constitute a limitation on electronic device 4. It may include more or fewer components than shown, or different components.

[0084] The processor 401 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0085] The memory 402 can be an internal storage unit of the electronic device 4, such as a hard disk or RAM of the electronic device 4. The memory 402 can also be an external storage device of the electronic device 4, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., equipped on the electronic device 4. The memory 402 can also include both internal and external storage units of the electronic device 4. The memory 402 is used to store computer programs and other programs and data required by the electronic device.

[0086] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0087] If an integrated module / unit is implemented as 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, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in a computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0088] The above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit it. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be included within the protection scope of this disclosure.

Claims

1. A product recommendation method, characterized in that, include: Obtain a training dataset, wherein the training dataset includes: multiple sets of training data, each set of training data including: user identification code, user information, product identification code, product information, and interaction information about the user's interaction behavior with the product; The probability of a user being a new user, the probability of missing user information, and the probability of a product being a new product are obtained within a unit of time. The first, second, and third probabilities are obtained based on historical data statistics from the e-commerce platform. Based on the first probability that a user is a new user within a unit of time, the second probability that user information is missing, and the third probability that a product is a new product, data augmentation processing is performed on the training dataset. Extract the data features of each training data set in the training dataset after the data augmentation process. The data features of each training data set include: user identification code features, user information features, product identification code features, product information features, and interaction features. A product recommendation model is trained using multiple sets of extracted data features; during training, user conversion behavior information is used as the label. Use a trained product recommendation model to provide product recommendation services to target users; The data augmentation process performed on the training dataset based on the first probability that a user is a new user within the unit of time, the second probability that user information is missing, and the third probability that the product is a new product includes: The training dataset was sampled multiple times as follows: The user identification codes in the training dataset are retained according to the first probability, retaining their original values. The user identifier code that does not retain its original value is set as a mask; The user information in the training dataset is preserved in its original value according to the second probability. User information that does not retain its original value is masked; The product identification codes in the training dataset are retained according to the third probability, retaining their original values. Product identification codes that do not retain their original values ​​are masked; The extraction of data features from each group of training data in the training dataset after the data augmentation process includes: Extract the user identification code features from the user identification code; Extract the user information features from the user information; Extract the product identification code features from the product identification code; Extract the product information features from the product information; The interaction features are extracted from the interaction information of the user's interaction with the product. The extraction of data features from each group of training data in the training dataset after the data augmentation process includes: The mapping method is used to process the data types in each training data set to obtain the data features of each training data set; and / or The non-quantitative data in each training data set is processed using an encoding method to obtain the data features of each training data set.

2. The method according to claim 1, characterized in that, The process of training a product recommendation model using extracted multiple sets of data features includes: Obtain information from each group of training data in the training dataset that indicates whether a user has converted their product; Based on the information corresponding to each set of training data that indicates whether a user has conversion behavior towards a product, labels are labeled for each set of data features. The label corresponding to each set of data features indicates whether the product corresponding to that set of training data is recommended to the user corresponding to that set of training data. The product recommendation model is trained using multiple sets of data features labeled with corresponding tags.

3. The method according to claim 1, characterized in that, The process of training a product recommendation model using extracted multiple sets of data features includes: The product recommendation model includes: a feature processing network and a recommendation judgment network; The product recommendation model is trained using multiple sets of extracted data features, so that the feature processing network learns and stores the correspondence between the data features and the joint features, and the recommendation judgment network learns and stores the correspondence between the joint features and whether a product is recommended, wherein each set of data features corresponds to one joint feature.

4. The method according to claim 2 or 3, characterized in that, include: Each data feature in the extracted multiple sets of data features is masked according to a preset probability. The product recommendation model is trained using the features of multiple sets of data after the masking process.

5. A product recommendation device, characterized in that, include: The first acquisition module is configured to acquire a training dataset, wherein the training dataset includes: multiple sets of training data, each set of training data including: user identification code, user information, product identification code, product information, and interaction information about the user's interaction behavior with the product; The second acquisition module is configured to acquire the first probability that a user is a new user, the second probability that user information is missing, and the third probability that a product is a new product within a unit of time; the first probability, the second probability, and the third probability are obtained based on historical data statistics of the e-commerce platform; The processing module is configured to perform data augmentation processing on the training dataset based on a first probability that the user is a new user, a second probability that the user information is missing, and a third probability that the product is a new product within the unit of time. The extraction module is configured to extract data features of each set of training data in the training dataset after the data augmentation process, wherein the data features of each set of training data include: user identification code features, user information features, product identification code features, product information features, and interaction features. The training module is configured to train a product recommendation model using multiple sets of extracted data features; during training, user conversion behavior information for products is used as the label. The service module is configured to provide product recommendation services to target users using a pre-trained product recommendation model; The processing module is specifically configured to perform the following multiple samplings on the training dataset: retaining the original value of the user identifier code in the training dataset according to the first probability, and setting the user identifier code that does not retain the original value as a mask; retaining the original value of the user information in the training dataset according to the second probability, and setting the user information that does not retain the original value as a mask; retaining the original value of the product identifier code in the training dataset according to the third probability, and setting the product identifier code that does not retain the original value as a mask; The extraction module is specifically configured to: extract the user identification code features from the user identification code; extract the user information features from the user information; extract the product identification code features from the product identification code; extract the product information features from the product information; extract the interaction features from the interaction information of the user's interaction with the product; process the quantity-type data in each group of training data using a mapping method to obtain the data features of each group of training data; and / or process the non-quantity-type data in each group of training data using an encoding method to obtain the data features of each group of training data.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 4.

Images