A product recommendation method, device, electronic device and storage medium

By dissociating and characterizing users' historical shopping behavior and social behavior, the product recommendation model is trained, and the problem of coupling social preferences and shopping preferences in the existing system is solved, and the accuracy and interpretability of product recommendations are improved.

CN114707063BActive Publication Date: 2024-12-03TSINGHUA UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210310761.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-28
Publication Date
2024-12-03
Estimated Expiration
2042-03-28

AI Technical Summary

Technical Problem

The existing product recommendation system based on user social relationships ignores the potential factors behind social information, resulting in the coupling of factors affecting social preferences and shopping preferences, reducing the accuracy of product recommendations.

Method used

By obtaining the user's historical shopping behavior and social behavior, they are represented by vectors respectively, and inputting the pre-trained product recommendation model, the shopping behavior sample set and social behavior sample set of multiple historical users are used for dissociation and characterization. According to two The product recommendation model is trained to improve the accuracy of recommendations.

Benefits of technology

Decompose the interaction between shopping behavior and social behavior at a finer granularity level, clarify the impact of users' social behavior on shopping behavior, and provide more suitable product recommendations, thereby improving the accuracy and interpretability of product recommendations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114707063B_ABST
    Figure CN114707063B_ABST
Patent Text Reader

Abstract

The present application provides a product recommendation method, device, electronic device and storage medium, which belongs to the field of data processing technology. The method includes: obtaining the historical shopping behavior and social behavior of the current user; representing the historical shopping behavior and social behavior of the current user with vectors respectively; inputting the vector representation of the historical shopping behavior and the vector representation of the social behavior of the current user into a pre-trained product recommendation model, and determining the target product recommended to the current user through the pre-trained product recommendation model; wherein the pre-trained product recommendation model is based on a shopping behavior sample set and a social behavior sample set of multiple historical users, by dissociating the shopping behavior sample set and the social behavior sample set, respectively, and training according to the influence relationship between the dissociated representation results of the shopping behavior sample set and the dissociated representation results of the social behavior sample set. The present application aims to improve the accuracy of product recommendations.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present application relate to the field of data processing technology, and in particular, to a product recommendation method, device, electronic device, and storage medium. Background Art

[0002] Nowadays, in order to improve users' shopping experience, product recommendation systems are developing rapidly. Product recommendation systems consider users' social relationships as an important factor affecting shopping choices.

[0003] However, the existing product recommendation system based on user social relationships only models users' social relationships as a factor that influences users' shopping decisions, ignoring a large number of potential factors behind social information. It is easy for the factors that affect social preferences and shopping preferences to be coupled with each other, thereby reducing the accuracy of product recommendations. Summary of the invention

[0004] The embodiments of the present application provide a product recommendation method, device, electronic device, and storage medium, aiming to improve the accuracy of product recommendations.

[0005] In a first aspect, an embodiment of the present application provides a product recommendation method, the method comprising:

[0006] Obtain the current user's historical shopping and social behaviors;

[0007] Representing the historical shopping behavior and social behavior of the current user with vectors respectively;

[0008] Inputting the vector representation of the historical shopping behavior and the vector representation of the social behavior of the current user into a pre-trained commodity recommendation model, and determining a target commodity recommended to the current user through the pre-trained commodity recommendation model;

[0009] Among them, the pre-trained product recommendation model is based on the shopping behavior sample sets and social behavior sample sets of multiple historical users. It is obtained by dissociating the shopping behavior sample sets and the social behavior sample sets respectively, and training according to the influence relationship between the dissociated representation results of the shopping behavior sample sets and the dissociated representation results of the social behavior sample sets.

[0010] Optionally, the product recommendation model is trained according to the following steps:

[0011] Obtain the historical shopping and social behaviors of multiple historical users;

[0012] Constructing the shopping behavior sample set according to the historical shopping behaviors of the plurality of historical users, and constructing the social behavior sample set according to the historical social behaviors of the plurality of historical users;

[0013] Inputting the shopping behavior sample set and the social behavior sample set into a preset model respectively, wherein the preset model includes a plurality of variational autoencoders;

[0014] Determine the dissociation characterization results of the shopping behavior sample set and the social behavior sample set respectively through a preset model;

[0015] According to the dissociated characterization result of the shopping behavior sample set, the dissociated characterization result of the social behavior sample set is corrected, and the model parameters of the preset model are updated according to the corrected result;

[0016] According to the dissociated characterization result of the social behavior sample set, the dissociated characterization result of the shopping behavior sample set is corrected, and the model parameters of the preset model are updated according to the corrected result;

[0017] After iteratively updating the preset model a preset number of times, the product recommendation model is obtained.

[0018] Optionally, after respectively determining the dissociation characterization results of the shopping behavior sample set and the social behavior sample set through a preset model, the method further includes:

[0019] Determining the real influence parameters between the multiple historical users according to the dissociation characterization results of the shopping behavior sample set and the social behavior sample set;

[0020] The dissociation characterization result of the social behavior sample set is modified according to the dissociation characterization result of the shopping behavior sample set, including:

[0021] According to the real impact parameter and the social correction coefficient, the dissociation characterization result of the social behavior sample set is corrected;

[0022] The dissociation characterization result of the shopping behavior sample set is modified according to the dissociation characterization result of the social behavior sample set, including:

[0023] The dissociation characterization result of the shopping behavior sample set is corrected according to the real influence parameter and the shopping correction coefficient.

[0024] Optionally, the dissociated characterization result of the shopping behavior sample set includes a plurality of first channels, each of which represents a shopping intention of the user; the dissociated characterization result of the social behavior sample set includes a plurality of second channels, each of which represents a social intention of the user;

[0025] Determining the real influence parameters between the multiple historical users according to the dissociation characterization results of the shopping behavior sample set and the social behavior sample set, including:

[0026] Calculating the similarity of user shopping representations of each of the plurality of first channels respectively;

[0027] Determining an average shopping representation similarity between a plurality of historical users according to the user shopping representation similarities of the plurality of first channels;

[0028] Calculating the similarity of user social representations of each of the plurality of second channels respectively;

[0029] Determining an average social representation similarity between a plurality of historical users according to the user social representation similarities of the plurality of second channels;

[0030] The real influence parameter is determined according to the average shopping representation similarity and the average social representation similarity.

[0031] Optionally, the calculation formula for the similarity of the user shopping representations of the multiple first channels is:

[0032]

[0033] The calculation formula for average shopping representation similarity is:

[0034]

[0035] In the above two formulas, X is the user's historical shopping behavior; For the kth X The similarity of user shopping representations in the first channel; X Characterize similarity for average shopping; kth after dissociation characterization X The dissociation characterization results corresponding to the first channel; 1≤k X ≤K X , K X is the total number of channels for the first channel.

[0036] Optionally, the calculation formula for the similarity of the user social representations of the multiple second channels is:

[0037]

[0038] The calculation formula for average social representation similarity is:

[0039]

[0040] In the above two formulas, A is the user's historical social behavior; For the kthA The similarity of user social representations in the second channel; A is the average social representation similarity; kth after dissociation characterization A The dissociation characterization results corresponding to the second channel; 1≤k A ≤K A , K A is the total number of channels for the second channel.

[0041] Optionally, the calculation formula of the real impact parameter is:

[0042] sim=sim X ·sim A

[0043] In the formula, sim X is the average shopping representation similarity; sim A is the average social representation similarity;

[0044] According to the real impact parameter and the shopping correction coefficient, the formula for correcting the dissociation characterization result of the shopping behavior sample set is:

[0045]

[0046] In the formula, α X is the shopping correction factor;

[0047] According to the real impact parameter and the social correction coefficient, the formula for correcting the dissociation characterization result of the social behavior sample set is:

[0048]

[0049] In the formula, α A is the social correction factor.

[0050] In a second aspect, an embodiment of the present application provides a product recommendation device, the device comprising:

[0051] The acquisition module is used to obtain the historical shopping and social behaviors of the current user;

[0052] A vector representation module, used to represent the historical shopping behavior and social behavior of the current user using vectors respectively;

[0053] A product recommendation module is used to input the vector representation of the current user's historical shopping behavior and the vector representation of the social behavior into a pre-trained product recommendation model, and determine the target product recommended to the current user through the pre-trained product recommendation model; wherein the pre-trained product recommendation model is based on a shopping behavior sample set and a social behavior sample set of multiple historical users, by dissociating the shopping behavior sample set and the social behavior sample set, respectively, and training according to the influence relationship between the dissociated representation results of the shopping behavior sample set and the dissociated representation results of the social behavior sample set.

[0054] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the product recommendation method described in the first aspect of the embodiment when executing the computer program.

[0055] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the product recommendation method described in the first aspect of the embodiment is implemented.

[0056] Beneficial effects:

[0057] The historical shopping behavior and social behavior of the current user are obtained, and the historical shopping behavior and social behavior of the current user are represented by vectors, and the vector representation of the historical shopping behavior and the vector representation of the social behavior of the current user are input into a pre-trained product recommendation model, and the target product recommended to the current user is determined through the pre-trained product recommendation model.

[0058] In this method, a product recommendation model is used to recommend target products to a user based on the current user's historical shopping behavior and social behavior. The product recommendation model is trained based on a shopping behavior sample set and a social behavior sample set of multiple historical users. During the training process, both the shopping behavior sample set and the social behavior sample set are dissociatedly represented, and then the shopping behavior and the social behavior are decomposed into a finer granularity. Then, based on the influence relationship between the dissociated representation results of the shopping behavior sample set and the dissociated representation results of the social behavior sample set, the shopping behavior and the social behavior interact with each other. The product recommendation model trained in this way determines the influence of the user's social behavior on the shopping behavior at a finer granularity level, and can provide more suitable products to the current user, thereby improving the accuracy of product recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0060] Figure 1 is a flowchart of the steps of a product recommendation method proposed in an embodiment of the present application;

[0061] Figure 2 is a flowchart of the steps of training a product recommendation model proposed in an embodiment of the present application;

[0062] Figure 3 It is a functional module diagram of a product recommendation device proposed in one embodiment of the present application. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0064] Users' shopping behaviors and social behaviors are usually reflections of their interests and hobbies. Therefore, there is a connection between their shopping behaviors and social behaviors, and the two will influence each other. For example, if a user likes to buy electronic devices, then the user will tend to follow bloggers in the technology or electronics fields; or if a user likes beauty products, then the user will also follow some beauty bloggers; similarly, if a user follows bloggers in the technology or electronics fields, it may also influence the user to buy electronic devices; or if a user follows a beauty blogger, the user is also more likely to buy beauty products.

[0065] This application dissociates the user's shopping behavior and social behavior, and mines the interactive relationship between each dissociation result between shopping behavior and social behavior at a more fine-grained level, thereby improving the accuracy of recommending products to users.

[0066] Reference Figure 1 , which shows a flowchart of the steps of a commodity recommendation method in an embodiment of the present invention, such as Figure 1 , the method may specifically include the following steps:

[0067] S101: Obtain the historical shopping behavior and social behavior of the current user.

[0068] In the actual implementation process, all or part of the current user's historical shopping behavior and social behavior can be obtained. If the current user wants to shop on a shopping client, the user's historical shopping behavior on the shopping client can be obtained. For example, by obtaining the current user's shopping records in the historical order list, and obtaining the current user's social behavior on the shopping client, such as the stores the current user follows, the stores frequently visited, the favorite products, and other shopping users or bloggers followed; with the current user's permission or authority, the user's historical shopping behavior and social behavior on other clients can also be obtained.

[0069] S102: Represent the historical shopping behavior and social behavior of the current user with vectors respectively.

[0070] After obtaining the historical shopping behavior and social behavior of the current user, the historical shopping behavior and social behavior are represented by vectors to meet the input condition requirements of the product recommendation model. For example, for any current user, his shopping behavior is a vector composed of 0 and 1, which respectively indicates that he has not purchased / purchased a product; similarly, his social behavior is also a vector composed of 0 and 1, which respectively indicates that he has not followed / followed other users, stores or bloggers.

[0071] S103: Input the vector representation of the historical shopping behavior and the vector representation of the social behavior of the current user into a pre-trained product recommendation model, and determine the target product recommended to the current user through the pre-trained product recommendation model.

[0072] Among them, the pre-trained product recommendation model is based on the shopping behavior sample sets and social behavior sample sets of multiple historical users. It is obtained by dissociating the shopping behavior sample sets and the social behavior sample sets respectively, and training according to the influence relationship between the dissociated representation results of the shopping behavior sample sets and the dissociated representation results of the social behavior sample sets.

[0073] During the training process, the product recommendation model dissociates the shopping behavior sample set and the social behavior sample set, and then further decomposes the user's preference when executing any intention in the shopping behavior and social behavior process into a finer granularity. Then, based on the influence relationship between the dissociated representation results of the shopping behavior sample set and the dissociated representation results of the social behavior sample set, the shopping behavior and the social behavior interact with each other. The product recommendation model trained in this way determines the impact of the user's social behavior on the shopping behavior at a finer granularity level, and can provide more suitable products to the current user, thereby improving the accuracy of product recommendations and making the product recommendations more explainable and controllable.

[0074] Reference Figure 2, which shows a flowchart of the steps of training a product recommendation model in an embodiment of the present invention, such as Figure 2 As shown, the product recommendation model is trained according to the following steps:

[0075] S201: Obtain historical shopping behaviors and historical social behaviors of multiple historical users.

[0076] The number of historical users is not limited in this embodiment and can be customized according to actual needs; the user's historical shopping behavior can be based on the historical shopping records of the same shopping client or based on the historical shopping records of different shopping clients, and the user's historical social behavior can be based on the same social platform or based on different social platforms.

[0077] S202: constructing the shopping behavior sample set according to the historical shopping behaviors of the plurality of historical users, and constructing the social behavior sample set according to the historical social behaviors of the plurality of historical users.

[0078] In the actual implementation process, the shopping behavior sample set and the social behavior sample set can be matrices. For example, the shopping behavior sample set is an N×M matrix X, where N rows represent N users, M columns represent M commodities, and the value x in the matrix ij is 0 or 1, x ij When it is 0, it means that a user i did not buy a certain product j, x ij When it is 1, it means that a user i has purchased a product j. The social behavior sample set is a matrix A of size N×N, where N rows and N columns represent N users. The value a in the matrix ij is 0 or 1, a ij When it is 0, it means that a user i does not follow another user j. ij When it is 1, it means that a user i follows another user j.

[0079] S203: Input the shopping behavior sample set and the social behavior sample set into a preset model respectively, wherein the preset model includes a plurality of variational autoencoders.

[0080] Variational Auto-Encoder (VAE) is a structure composed of an encoder and a decoder. The encoder is mainly used to reduce the dimension or compress the initial input data, that is, to reduce the number of features describing the data by retaining or extracting features from the features of the initial data. The decoder is used to perform the inverse process, that is, to reconstruct based on the retained or extracted features.

[0081] In this method, the encoder is used to encode the high-dimensional effective information sparse data of the shopping behavior sample set and the social behavior sample set into low-dimensional effective information dense data, that is, to perform dissociated representation on the shopping behavior sample set and the social behavior sample set. The dissociated representation process is performed in an unsupervised manner and does not require any other additional information supervision. Only the above two matrices X and A are used to train the preset model, so that the preset model restores the two initial matrices after the dissociated representation.

[0082] S204: Determine the dissociation characterization results of the shopping behavior sample set and the social behavior sample set respectively through a preset model.

[0083] In this method, the preset model includes multiple encoders, and the dissociated vector encoded by each VAE encoder is regarded as a channel, so that the shopping behavior sample set and the social behavior sample set can be represented as dissociated representations of multiple channels. The dissociated representation results of the shopping behavior sample set include multiple first channels, each of which represents a shopping intention of the user; the dissociated representation results of the social behavior sample set include multiple second channels, each of which represents a social intention of the user.

[0084] The channel represents the user's high-level intention, and the different dimensions of the dissociation vector represent the low-level preferences. For example, the high-level intention represented by the channel corresponding to the shopping behavior is the type of items purchased by the user during the shopping process. For example, channel 1 corresponding to the shopping behavior is clothing and accessories, channel 2 is sports equipment, and channel 3 is electronic products. The low-level preference mainly refers to the characteristics of the items, such as the style, size, color, etc. of the clothes; the channel corresponding to the social behavior represents the high-level intention of the user's group attributes, such as sports enthusiasts, beauty bloggers, electronics enthusiasts, etc., and the low-level preference refers to the user's personal attributes, such as age, gender, interests, etc., because a user may be paid attention to by another user because of his or her group attributes or personal attributes.

[0085] Among them, when determining the type of goods or the attributes of users, the "prototype" method is adopted, and the goods or users that are closer to a certain category prototype are regarded as this category. For example, the representation vectors of coats, sweaters, shirts, T-shirts, etc. are different. By taking the average of their representation vectors, the vector corresponding to the "midpoint" of the above multiple representation vectors is obtained, and the representation vector corresponding to this average value is regarded as the representation vector of the "clothing" category of goods; when new product data enters the model, the distance between the representation vector of the new product and the representation vector of "clothing" is judged. If the distance is less than the preset value, it represents that the new product belongs to the category of "clothing".

[0086] In real life, the shopping behavior and social behavior of users will affect each other. For example, user 1 purchases product 1 and may get to know user 2 who also purchased product 1. After user 1 and user 2 establish a social relationship, user 2's evaluation or use of product 2 may affect user 1's purchase of product 2. Therefore, after obtaining the dissociated characterization results of the shopping behavior sample set and the dissociated characterization results of the social behavior sample set, this method will also make the dissociated characterization results of the shopping behavior sample set and the dissociated characterization results of the social behavior sample set correct each other to perform alternating iterative updates.

[0087] Before the dissociated characterization results of the shopping behavior sample set and the dissociated characterization results of the social behavior sample set are mutually corrected, it is necessary to determine the real influence parameters between the multiple historical users according to the dissociated characterization results of the shopping behavior sample set and the social behavior sample set, and the specific steps are as follows:

[0088] The shopping behavior sample set is denoted as X, the social behavior sample set is denoted as A, and the dissociated representation result of the shopping behavior sample set is denoted as The dissociative representation result of the social behavior sample set is recorded as Among them, 1≤k X ≤K X , 1≤k A ≤K A , K X is the total number of channels of the first channel; K A is the total number of channels for the second channel.

[0089] A1: Calculate the user shopping representation similarity of each of the multiple first channels respectively. The user shopping representation similarity represents the shopping representation similarity of different users in the kth channel. X The similarity between the items purchased with the same intention.

[0090] kth X The calculation formula for the similarity of user shopping representations in the first channel is:

[0091]

[0092] The similarity of user shopping representations of all first channels is calculated in sequence.

[0093] A2: Determine an average shopping representation similarity between multiple historical users according to the user shopping representation similarities of each of the multiple first channels.

[0094] The calculation formula for average shopping representation similarity is:

[0095]

[0096] A3: Calculate the user social representation similarities of each of the plurality of second channels respectively, where the user social representation similarities represent the similarity between group attributes of users.

[0097] kth A The calculation formula for the similarity of user social representations in the second channel is as follows:

[0098]

[0099] The similarity of user social representations of all second channels is calculated in turn.

[0100] A4: Determine an average social representation similarity between multiple historical users according to the user social representation similarities of each of the multiple second channels.

[0101] The calculation formula for average social representation similarity is:

[0102]

[0103] A5: Determine the real impact parameter according to the average shopping representation similarity and the average social representation similarity.

[0104] Average shopping representation similarity sim X Similarity to the average social representation sim A The closeness of the connection between users in terms of shopping and social interaction is reflected in the average shopping similarity sim X Similarity to the average social representation sim A Find the intersection to get the real impact parameter sim, that is, sim = sim X ·sim A , the real influence parameter sim is used to reflect the mutual influence between users. For example, if sim X A larger value indicates that users have similar shopping behaviors; if sim A A larger value indicates that the users have similar social behaviors. If the product of the two is large, that is, the value of the real influence parameter sim is large, it means that the users are similar in both shopping and behavior, rather than just one aspect.

[0105] S205: According to the dissociated characterization result of the shopping behavior sample set, the dissociated characterization result of the social behavior sample set is corrected, and the model parameters of the preset model are updated according to the corrected result.

[0106] In a feasible implementation manner, after the real influence parameter is calculated, the dissociation characterization result of the shopping behavior sample set is corrected according to the real influence parameter and the shopping correction coefficient, and the formula is:

[0107]

[0108] In the formula, α X is the shopping correction coefficient, α X The value of can be customized. The greater the impact of social behavior on shopping behavior, the greater the X The bigger.

[0109] S206: According to the dissociated characterization result of the social behavior sample set, the dissociated characterization result of the shopping behavior sample set is corrected, and the model parameters of the preset model are updated according to the corrected result.

[0110] In a feasible implementation manner, the dissociation characterization result of the social behavior sample set is corrected according to the real influence parameter and the social correction coefficient, and the formula is:

[0111]

[0112] In the formula, α A is the social correction coefficient, α A The value of can be customized. The greater the impact of shopping behavior on social behavior, the greater the A The bigger.

[0113] S207: After iteratively updating the preset model a preset number of times, the product recommendation model is obtained.

[0114] The process of training the preset model includes a dissociation process and a mutual correction process. For example, the preset model can first dissociate the shopping behavior sample set, and preliminarily obtain the dissociated representation results of different users' shopping habits. Then, in the process of the preset model dissociating the social behavior sample set, the preset model corrects the dissociated representation of the social behavior sample set based on the average shopping representation similarity of the users obtained when dissociating the shopping behavior sample set. The above process is regarded as an iteration. In the second iteration, the dissociated representation results of the social behavior sample set will also correct the dissociated representation of the shopping behavior sample set. After the preset model is iteratively updated for a preset number of times, the product recommendation model is obtained.

[0115] In a feasible implementation, when the preset model is trained for a preset number of times, each iterative training can be based on different shopping behavior sample sets and social behavior sample sets, and each training includes the dissociation process of shopping behavior and social behavior, shopping correction, and reconstruction process. For example, the first training is based on the dissociated representation of shopping behavior sample 1 and social behavior sample 1, and the dissociated representation of social behavior sample 1 is corrected by using the average shopping representation similarity of shopping behavior sample 1. The second training is based on the dissociated representation of shopping behavior sample 2 and social behavior sample 2, and the dissociated representation of shopping behavior sample 2 is corrected by using the average social representation similarity of social behavior sample 2. After such alternating iterative updates, when the number of iterations is the preset number, a product recommendation model is obtained.

[0116] Generally speaking, the training process of the product recommendation model is to dissociate the shopping behavior sample set and the social behavior sample set, calculate the similarity of the dissociated representation results of the two, and provide a smaller weighting to each other as a correction.

[0117] After the product recommendation model is trained, when recommending to the current user, the current user's historical shopping behavior is represented by a 0 / 1 vector, and the current user's historical social behavior is represented by a 0 / 1 vector. Taking historical shopping behavior as an example, assuming that there are five products A, B, C, D, and E in the product collection, and the current user has purchased product B, then the vector of the input shopping information is [0,1,0,0,0]. The current user has not purchased the other four products A, C, D, and E.

[0118] The vector representation of the current user's historical shopping behavior and the vector representation of social behavior are input into the pre-trained product recommendation model. The product recommendation model dissociates the current user's shopping and social preferences. The dissociated low-dimensional dense vector contains the user's intentions and preferences. This dissociation result is then restored through the decoder to the user's possible shopping needs. Assume that the restored vector is [0.1, 0.9, 0.3, 0.4, 0.7], which corresponds to the possibility of the current user purchasing five items, among which B has the greatest possibility because it is a known user purchase. After ignoring the corresponding possibility of the purchased items, the items with a probability greater than the preset possibility threshold are recommended to the current user. The preset feasibility threshold and the number of recommended items can be set according to the actual application.

[0119] Reference Figure 3 , shows a functional module diagram of a commodity recommendation device in an embodiment of the present invention, the device comprising:

[0120] The acquisition module 100 is used to acquire the historical shopping behavior and social behavior of the current user;

[0121] A vector representation module 200, used to represent the historical shopping behavior and social behavior of the current user using vectors respectively;

[0122] The product recommendation module 300 is used to input the vector representation of the current user's historical shopping behavior and the vector representation of the social behavior into a pre-trained product recommendation model, and determine the target product recommended to the current user through the pre-trained product recommendation model; wherein the pre-trained product recommendation model is based on a shopping behavior sample set and a social behavior sample set of multiple historical users, by dissociating the shopping behavior sample set and the social behavior sample set, respectively, and training according to the influence relationship between the dissociated representation results of the shopping behavior sample set and the dissociated representation results of the social behavior sample set.

[0123] Optionally, the device further includes a model training module. Specifically, the model training module includes:

[0124] An acquisition unit, used to acquire the historical shopping behaviors and historical social behaviors of multiple historical users;

[0125] A sample determination unit, configured to construct the shopping behavior sample set according to the historical shopping behaviors of the plurality of historical users, and to construct the social behavior sample set according to the historical social behaviors of the plurality of historical users;

[0126] An input unit, used to input the shopping behavior sample set and the social behavior sample set into a preset model respectively, wherein the preset model includes a plurality of variational autoencoders;

[0127] A dissociation unit, used to determine the dissociation characterization results of the shopping behavior sample set and the social behavior sample set respectively through a preset model;

[0128] A first correction unit, configured to correct the dissociation characterization result of the social behavior sample set according to the dissociation characterization result of the shopping behavior sample set, and update the model parameters of the preset model according to the correction result;

[0129] A second correction unit, configured to correct the dissociation characterization result of the shopping behavior sample set according to the dissociation characterization result of the social behavior sample set, and update the model parameters of the preset model according to the correction result;

[0130] The model determination unit is used to obtain the product recommendation model after iteratively updating the preset model a preset number of times.

[0131] Optionally, the model training module also includes:

[0132] A real influence parameter determination unit, configured to determine the real influence parameters between the plurality of historical users according to the dissociation characterization results of the shopping behavior sample set and the social behavior sample set;

[0133] The first correction unit comprises:

[0134] A first correction subunit, configured to correct the dissociation characterization result of the social behavior sample set according to the real influence parameter and the social correction coefficient;

[0135] The second correction unit comprises:

[0136] The second correction subunit corrects the dissociation characterization result of the shopping behavior sample set according to the real influence parameter and the shopping correction coefficient.

[0137] Optionally, the dissociated characterization result of the shopping behavior sample set includes a plurality of first channels, each of which represents a shopping intention of the user; the dissociated characterization result of the social behavior sample set includes a plurality of second channels, each of which represents a social intention of the user;

[0138] The real impact parameter determination unit comprises:

[0139] A first similarity determination subunit, configured to respectively calculate the user shopping representation similarities of each of the plurality of first channels;

[0140] A first averaging subunit, configured to determine an average shopping representation similarity between a plurality of historical users according to the user shopping representation similarities of the plurality of first channels;

[0141] A second similarity determination subunit, configured to respectively calculate the user social representation similarities of each of the plurality of second channels;

[0142] A second averaging subunit, configured to determine an average social representation similarity between a plurality of historical users according to the user social representation similarities of the plurality of second channels;

[0143] A determination subunit is used to determine the real influence parameter according to the average shopping representation similarity and the average social representation similarity.

[0144] An embodiment of the present application also 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 implements the commodity recommendation method described in the embodiment when executing the computer program.

[0145] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the product recommendation method described in the embodiment is implemented.

[0146] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0147] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, devices, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0148] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0149] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0150] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0151] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.

[0152] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0153] Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A product recommendation method, characterized in that: The method comprises: Obtain the current user's historical shopping and social behaviors; Representing the historical shopping behavior and social behavior of the current user with vectors respectively; Inputting the vector representation of the historical shopping behavior and the vector representation of the social behavior of the current user into a pre-trained commodity recommendation model, and determining a target commodity recommended to the current user through the pre-trained commodity recommendation model; The pre-trained product recommendation model is based on a shopping behavior sample set and a social behavior sample set of multiple historical users, and is obtained by dissociating the shopping behavior sample set and the social behavior sample set, respectively, and training according to the influence relationship between the dissociated representation results of the shopping behavior sample set and the dissociated representation results of the social behavior sample set; The training of the product recommendation model includes: According to the dissociated characterization result of the shopping behavior sample set, the dissociated characterization result of the social behavior sample set is corrected, and the model parameters of the preset model are updated according to the corrected result; According to the dissociated characterization result of the social behavior sample set, the dissociated characterization result of the shopping behavior sample set is corrected, and the model parameters of the preset model are updated according to the corrected result; After iteratively updating the preset model for a preset number of times, the product recommendation model is obtained; Wherein, according to the dissociation characterization results of the shopping behavior sample set and the social behavior sample set, the real influence parameters between the multiple historical users are determined; The dissociation characterization result of the social behavior sample set is modified according to the dissociation characterization result of the shopping behavior sample set, including: According to the real impact parameter and the social correction coefficient, the dissociation characterization result of the social behavior sample set is corrected; The dissociation characterization result of the shopping behavior sample set is modified according to the dissociation characterization result of the social behavior sample set, including: According to the real influence parameter and the shopping correction coefficient, the dissociation characterization result of the shopping behavior sample set is corrected; The dissociated characterization result of the shopping behavior sample set includes a plurality of first channels, each of which represents a shopping intention of the user; the dissociated characterization result of the social behavior sample set includes a plurality of second channels, each of which represents a social intention of the user; Determining the real influence parameters between the multiple historical users according to the dissociation characterization results of the shopping behavior sample set and the social behavior sample set, including: Calculating the similarity of user shopping representations of each of the plurality of first channels respectively; Determining an average shopping representation similarity between a plurality of historical users according to the user shopping representation similarities of the plurality of first channels; Calculating the similarity of user social representations of each of the plurality of second channels respectively; Determining an average social representation similarity between a plurality of historical users according to the user social representation similarities of the plurality of second channels; Determining the real influence parameter according to the average shopping representation similarity and the average social representation similarity; The calculation formula of the real impact parameter is: In the formula, Characterize similarity for average shopping; is the average social representation similarity; According to the real impact parameter and the shopping correction coefficient, the formula for correcting the dissociation characterization result of the shopping behavior sample set is: In the formula, is the shopping correction factor; After dissociation characterization The dissociation characterization results corresponding to the first channel; , is the total number of channels of the first channel; According to the real impact parameter and the social correction coefficient, the formula for correcting the dissociation characterization result of the social behavior sample set is: In the formula, is the social correction factor; After dissociation characterization The dissociation characterization results corresponding to the second channel; , The total number of channels for the second channel.

2. The method according to claim 1, characterized in that The product recommendation model is trained according to the following steps: Obtain the historical shopping and social behaviors of multiple historical users; Constructing the shopping behavior sample set according to the historical shopping behaviors of the plurality of historical users, and constructing the social behavior sample set according to the historical social behaviors of the plurality of historical users; Inputting the shopping behavior sample set and the social behavior sample set into a preset model respectively, wherein the preset model includes a plurality of variational autoencoders; The dissociation characterization results of the shopping behavior sample set and the social behavior sample set are determined respectively through a preset model.

3. The method according to claim 1, characterized in that The calculation formula for the similarity of user shopping representations of the multiple first channels is: The calculation formula for average shopping representation similarity is: In the above two formulas, X is the user's historical shopping behavior; For the Similarity of users’ shopping representations in the first channel; Characterize similarity for average shopping; After dissociation characterization The dissociation characterization results corresponding to the first channel; , is the total number of channels for the first channel.

4. The method according to claim 3, characterized in that The calculation formula for the similarity of the user social representations of the multiple second channels is: The calculation formula for average social representation similarity is: In the above two formulas, A is the user's historical social behavior; For the The similarity of user social representations in the second channel; is the average social representation similarity; After dissociation characterization The dissociation characterization results corresponding to the second channel; , is the total number of channels for the second channel.

5. A commodity recommendation device, characterized in that: The device comprises: The acquisition module is used to obtain the historical shopping and social behaviors of the current user; A vector representation module, used to represent the historical shopping behavior and social behavior of the current user using vectors respectively; A product recommendation module, used for inputting the vector representation of the historical shopping behavior and the vector representation of the social behavior of the current user into a pre-trained product recommendation model, and determining the target product recommended to the current user through the pre-trained product recommendation model; wherein the pre-trained product recommendation model is based on a shopping behavior sample set and a social behavior sample set of multiple historical users, and is obtained by dissociating the shopping behavior sample set and the social behavior sample set, respectively, and training according to the influence relationship between the dissociated representation results of the shopping behavior sample set and the dissociated representation results of the social behavior sample set; The training of the product recommendation model includes: According to the dissociated characterization result of the shopping behavior sample set, the dissociated characterization result of the social behavior sample set is corrected, and the model parameters of the preset model are updated according to the corrected result; According to the dissociated characterization result of the social behavior sample set, the dissociated characterization result of the shopping behavior sample set is corrected, and the model parameters of the preset model are updated according to the corrected result; After iteratively updating the preset model for a preset number of times, the product recommendation model is obtained; Wherein, according to the dissociation characterization results of the shopping behavior sample set and the social behavior sample set, the real influence parameters between the multiple historical users are determined; The dissociation characterization result of the social behavior sample set is modified according to the dissociation characterization result of the shopping behavior sample set, including: According to the real impact parameter and the social correction coefficient, the dissociation characterization result of the social behavior sample set is corrected; The dissociation characterization result of the shopping behavior sample set is modified according to the dissociation characterization result of the social behavior sample set, including: According to the real influence parameter and the shopping correction coefficient, the dissociation characterization result of the shopping behavior sample set is corrected; The dissociated characterization result of the shopping behavior sample set includes a plurality of first channels, each of which represents a shopping intention of the user; the dissociated characterization result of the social behavior sample set includes a plurality of second channels, each of which represents a social intention of the user; Determining the real influence parameters between the multiple historical users according to the dissociation characterization results of the shopping behavior sample set and the social behavior sample set, including: Calculating the similarity of user shopping representations of each of the plurality of first channels respectively; Determining an average shopping representation similarity between a plurality of historical users according to the user shopping representation similarities of the plurality of first channels; Calculating the similarity of user social representations of each of the plurality of second channels respectively; Determining an average social representation similarity between a plurality of historical users according to the user social representation similarities of the plurality of second channels; Determining the real influence parameter according to the average shopping representation similarity and the average social representation similarity; The calculation formula of the real impact parameter is: In the formula, Characterize similarity for average shopping; is the average social representation similarity; According to the real impact parameter and the shopping correction coefficient, the formula for correcting the dissociation characterization result of the shopping behavior sample set is: In the formula, is the shopping correction factor; After dissociation characterization The dissociation characterization results corresponding to the first channel; , is the total number of channels of the first channel; According to the real impact parameter and the social correction coefficient, the formula for correcting the dissociation characterization result of the social behavior sample set is: In the formula, is the social correction factor; After dissociation characterization The dissociation characterization results corresponding to the second channel; , The total number of channels for the second channel.

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, the product recommendation method according to any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the commodity recommendation method according to any one of claims 1 to 4 is implemented.