CN111695960A

CN111695960A - Object recommendation system, method, electronic device and storage medium

Info

Publication number: CN111695960A
Application number: CN201910191021.9A
Authority: CN
Inventors: 吴纲律; 戚赟炜; 王双
Original assignee: Alibaba Group Holding Ltd
Current assignee: Alibaba Group Holding Ltd
Priority date: 2019-03-12
Filing date: 2019-03-12
Publication date: 2020-09-22

Abstract

The application provides an object recommendation system, method, electronic device and storage medium, the system includes: the mobile terminal is in communication connection with the server; the mobile terminal is configured to: the method comprises the steps of obtaining real-time behaviors and behavior objects of a user to be recommended, generating corresponding dynamic feature embedded vectors based on the real-time behaviors and the behavior objects, and sending the dynamic feature embedded vectors to a server; the server is configured to: receiving the dynamic characteristic embedding vector, calculating the matching degree of the dynamic characteristic embedding vector and a static characteristic embedding vector of at least one object available for recommendation, and recommending the object to the user to be recommended based on the matching degree of the dynamic characteristic embedding vector and the static characteristic embedding vector; and the static feature embedding vector is obtained through attribute information of at least one object available for recommendation and user attribute information of the user to be recommended.

Description

Object recommendation system, method, electronic device and storage medium

Technical Field

The present disclosure relates to the internet field, and in particular, to an object recommendation system, method, electronic device, and storage medium.

Background

With the popularization of the internet and the development of electronic commerce, more and more users choose to browse, select or purchase commodities required by the users on the internet. However, with the rapid increase of the number and types of commodities, users often need to spend a lot of time browsing to find the commodities needed or interested by themselves, and in order to enable users to find the commodities needed or interested by themselves in a short time, many websites recommend the commodities to the users in various forms to different degrees by adopting various types of recommendation technologies.

In the prior art, the commodity recommendation function is usually realized by calculating the correlation between commodities and users, and personalized recommendation requires interaction of commodity information between a server and a user side.

Disclosure of Invention

In view of this, embodiments of the present disclosure provide an object recommendation system, an object recommendation method, an electronic device, and a storage medium, so as to solve technical defects in the prior art.

In one aspect, an embodiment of the present specification discloses an object recommendation system, including:

the mobile terminal is in communication connection with the server;

the mobile terminal is configured to: the method comprises the steps of obtaining real-time behaviors and behavior objects of a user to be recommended, generating corresponding dynamic feature embedded vectors based on the real-time behaviors and the behavior objects, and sending the dynamic feature embedded vectors to a server;

the server configured to: receiving the dynamic characteristic embedding vector, calculating the matching degree of the dynamic characteristic embedding vector and a static characteristic embedding vector of at least one object available for recommendation, and recommending the object to the user to be recommended based on the matching degree of the dynamic characteristic embedding vector and the static characteristic embedding vector;

and the static feature embedding vector is obtained through attribute information of at least one object available for recommendation and user attribute information of the user to be recommended.

Optionally, the server is further configured to receive the dynamic feature embedding vector, determine a static feature embedding vector of an object corresponding to the dynamic feature embedding vector from static feature embedding vectors of at least one object available for recommendation according to the dynamic feature embedding vector, and recommend an object to the user to be recommended based on a matching degree between the dynamic feature embedding vector and the static feature embedding vector of the corresponding object.

Optionally, the mobile terminal is further configured to convert the real-time behaviors and behavior objects into dynamic feature embedding vectors through a pre-trained first recommendation model.

Optionally, the server is further configured to determine an object to be recommended according to the dynamic feature embedding vector; acquiring attribute information of the object to be recommended which can be recommended; acquiring attribute information of the user to be recommended; and acquiring at least one static feature embedding vector corresponding to the attribute information of the object to be recommended and the attribute information of the user to be recommended, which are available for recommendation, through a pre-trained second recommendation model.

Optionally, the real-time behavior of the user to be recommended includes at least one of the following:

the method comprises the following steps that the times of clicking the object, the times and the duration of browsing similar or related objects, and browsing and collecting the object by a user to be recommended in a preset time period;

the attribute information of the user to be recommended comprises at least one of the following items:

gender, age, hobbies, city, education level, and income level.

Optionally, the first recommendation model is trained as follows:

acquiring historical behavior data, user attribute information, object attribute information and a selection result of the user on the object;

and training the first recommendation model by taking the historical behavior data, the user attribute information and the object attribute information of the user as training samples and taking the selection result of the user on the object as a training label, wherein the historical behavior data and the user attribute information of the user are associated with the object attribute information by the first recommendation model.

Optionally, the second recommendation model is trained as follows:

and training the second recommendation model by taking the historical behavior data, the user attribute information and the object attribute information of the user as training samples and taking the selection result of the user on the object as a training label, wherein the second recommendation model enables the historical behavior data and the user attribute information of the user to be associated with the object attribute information.

Optionally, the server is further configured to sort according to the matching degree to obtain an object recommendation list; and setting a matching degree threshold value, and recommending the objects with the matching degrees larger than the matching degree threshold value to the user to be recommended according to the sequence of the object recommendation list.

On the other hand, the embodiment of the present specification discloses an object recommendation method, which is applied to a mobile terminal and includes:

acquiring real-time behaviors and behavior objects of a user to be recommended;

and generating a corresponding dynamic feature embedded vector based on the real-time behaviors and the behavior objects, and sending the dynamic feature embedded vector to a server.

Optionally, generating the corresponding dynamic feature embedding vector based on the real-time behavior and the behavior object comprises:

and converting the real-time behaviors and the behavior objects into dynamic feature embedded vectors through a pre-trained first recommendation model.

Optionally, the first recommendation model is obtained by training using the following method:

acquiring historical behavior data of a user, user attribute information, object attribute information and a selection result of the user on the object, wherein the historical behavior data comprises historical behaviors and behavior objects;

Optionally, the real-time behavior of the user to be recommended includes at least one of:

the method comprises the steps of clicking the object by a user to be recommended within a preset time period, browsing the object for times and duration, browsing similar or related objects for times and duration, and browsing and collecting the objects.

On the other hand, the embodiment of the present specification discloses an object recommendation apparatus, including:

the processing module is configured to acquire real-time behaviors and behavior objects of the user to be recommended;

and the sending module is configured to generate a corresponding dynamic feature embedded vector based on the real-time behaviors and the behavior objects and send the dynamic feature embedded vector to a server.

On the other hand, the embodiment of the present specification discloses an object recommendation method, which is applied to a server and includes:

receiving a dynamic feature embedding vector;

calculating the matching degree of the dynamic feature embedding vector and a static feature embedding vector of at least one object available for recommendation;

and recommending the object to the user to be recommended based on the matching degree of the dynamic feature embedding vector and at least one static feature embedding vector, wherein the static feature embedding vector is obtained through the attribute information of at least one object available for recommendation and the user attribute information of the user to be recommended.

Optionally, an object recommendation method further includes:

receiving the dynamic feature embedding vector;

determining a static feature embedding vector of an object corresponding to the dynamic feature embedding vector from static feature embedding vectors of at least one object available for recommendation according to the dynamic feature embedding vector;

and recommending the object to the user to be recommended based on the matching degree of the dynamic feature embedding vector and the corresponding static feature embedding vector of the object.

Optionally, after receiving the dynamic feature embedding vector, before calculating a matching degree of the dynamic feature embedding vector and a static feature embedding vector of at least one object available for recommendation, the method further includes:

determining an object to be recommended according to the dynamic feature embedding vector;

acquiring attribute information of the object to be recommended which can be recommended;

acquiring attribute information of the user to be recommended;

and acquiring at least one static feature embedding vector corresponding to the attribute information of the object to be recommended and the attribute information of the user to be recommended, which are available for recommendation, through a pre-trained second recommendation model.

Optionally, the second recommendation model is trained as follows:

Optionally, the attribute information of the at least one object available for recommendation includes at least one of:

attribute information of at least one commodity available for recommendation, attribute information of at least one news available for recommendation, and attribute information of at least one video available for recommendation;

gender, age, hobbies, city, education level, and income level.

Optionally, the calculating the matching degree of the dynamic feature embedding vector and the static feature embedding vector of the at least one object available for recommendation includes:

and performing dot product operation on the dynamic feature embedding vector and the static feature embedding vector of the at least one object available for recommendation to obtain at least one corresponding numerical value, wherein the numerical value is the matching degree of the dynamic feature embedding vector and the static feature embedding vector of the at least one object available for recommendation.

Optionally, the recommending the object to the user to be recommended based on the matching degree of the dynamic feature embedding vector and the at least one static feature embedding vector includes:

sorting according to the matching degree to obtain an object recommendation list;

and setting a matching degree threshold value, and recommending the objects with the matching degrees larger than the matching degree threshold value to the user to be recommended according to the sequence of the object recommendation list.

a receiving module configured to receive a dynamic feature embedding vector;

a calculation module configured to calculate a degree of matching of the dynamic feature embedding vector with a static feature embedding vector of at least one object available for recommendation;

and the recommending module is configured to recommend the object to the user to be recommended based on the matching degree of the dynamic feature embedding vector and at least one static feature embedding vector, wherein the static feature embedding vector is obtained through attribute information of at least one object available for recommendation and user attribute information of the user to be recommended.

gender, age, hobbies, city, education level, and income level.

Optionally, the recommendation module comprises:

the processing submodule is configured to sort according to the matching degree to obtain an object recommendation list;

and the recommending submodule is configured to set a matching degree threshold value, and recommend the objects with the matching degrees larger than the matching degree threshold value to the user to be recommended according to the sequence of the object recommending list.

In another aspect, an embodiment of the present specification discloses an electronic device, including:

a memory and a processor;

the memory is used for storing computer-executable instructions, and the processor is used for executing the computer-executable instructions, and the instructions are executed by the processor to realize the steps of the object recommendation method applied to the mobile terminal.

a memory and a processor;

the memory is configured to store computer-executable instructions and the processor is configured to execute the computer-executable instructions, which when executed by the processor, implement the steps of the object recommendation method applied to the server.

In another aspect, an embodiment of the present specification discloses a computer-readable storage medium storing computer instructions, which when executed by a processor, implement the steps of an object recommendation method applied to a mobile terminal.

In another aspect, embodiments of the present specification disclose a computer-readable storage medium storing computer instructions which, when executed by a processor, implement the steps of an object recommendation method applied to a server.

In the object recommendation system, method, electronic device and computer-readable storage medium provided by the present specification, the mobile terminal obtains the real-time behavior and behavior object of the user to be recommended, generates the corresponding dynamic feature embedded vector based on the real-time behavior and behavior object, and sends the dynamic feature embedded vector to the server, so as to share a part of the calculation amount for the server, and the server only calculates the static data such as the attribute information of the user to be recommended and the attribute information of the product, and the like, so that the data processing efficiency of the server can be greatly improved, and a large amount of data does not need to be stored in the server, so that even if the user suddenly increases a large amount of data, the server is not congested, the real-time recommendation of the product is facilitated, and better service experience can be brought to the user.

Drawings

FIG. 1 is a schematic structural diagram of an object recommendation system according to an embodiment of the present disclosure;

FIG. 2 is an interaction diagram of an object recommendation method provided in an embodiment of the present specification;

FIG. 3 is a flowchart of an object recommendation method provided in an embodiment of the present specification;

FIG. 4 is a flowchart of an object recommendation method provided in an embodiment of the present specification;

FIG. 5 is a flowchart of an object recommendation method provided in an embodiment of the present specification;

fig. 6 is a schematic structural diagram of an object recommendation device according to an embodiment of the present disclosure;

fig. 7 is a schematic structural diagram of an object recommendation device according to an embodiment of the present disclosure;

fig. 8 is a schematic structural diagram of an object recommendation device according to an embodiment of the present disclosure;

fig. 9 is a block diagram of an electronic device according to an embodiment of the present specification;

fig. 10 is a block diagram of an electronic device according to an embodiment of the present disclosure.

Detailed Description

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present description. This description may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein, as those skilled in the art will be able to make and use the present disclosure without departing from the spirit and scope of the present disclosure.

The terminology used in the description of the one or more embodiments is for the purpose of describing the particular embodiments only and is not intended to be limiting of the description of the one or more embodiments. As used in one or more embodiments of the present specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used in one or more embodiments of the present specification refers to and encompasses any and all possible combinations of one or more of the associated listed items.

It will be understood that, although the terms first, second, etc. may be used herein in one or more embodiments to describe various information, these information should not be limited by these terms. These terms are only used to distinguish one type of information from another. For example, a first can also be referred to as a second and, similarly, a second can also be referred to as a first without departing from the scope of one or more embodiments of the present description. The word "if" as used herein may be interpreted as "at … …" or "when … …" or "in response to a determination", depending on the context.

First, the noun terms to which one or more embodiments of the present invention relate are explained.

Personalized recommendation: different commodities are recommended according to different user characteristics (such as behavior characteristics, basic attribute characteristics of the user, such as male and female, city, income and the like).

Embedding mapping: a mapping relationship that is not intuitively interpretable.

And (3) edge calculation: a process that uses a large amount of computing resources on the end (such as a cell phone).

In this specification, an object recommendation method and system, an electronic device, and a storage medium are provided, and detailed descriptions are individually made in the following embodiments.

Fig. 1 is a schematic diagram illustrating an architecture of an object recommendation system provided in an embodiment of the present specification, where the object recommendation system may include:

mobile terminal 102 and server 104, mobile terminal 102 is connected to server 104 in communication;

the mobile terminal 102 is configured to acquire real-time behaviors and behavior objects of a user to be recommended, generate corresponding dynamic feature embedded vectors based on the real-time behaviors and the behavior objects, and send the dynamic feature embedded vectors to a server;

in an embodiment of the present specification, the mobile terminal 102 is further configured to convert the real-time behaviors and behavior objects into dynamic feature embedding vectors through a pre-trained first recommendation model.

In an embodiment of the present specification, historical behavior data of a user, user attribute information, and object attribute information are used as training samples, and a selection result of the user on the object is used as a training tag to train the first recommendation model, where the first recommendation model associates the historical behavior data of the user and the user attribute information with the object attribute information, and the historical behavior data includes historical behaviors and behavior objects.

In an embodiment of the present specification, the historical behaviors and the real-time behaviors include user clicking, browsing, collecting and other behaviors, the behavior object is an object clicked, browsed and/or collected by a user, and the mobile terminal inputs the real-time behavior and the behavior object of the user with recommendation into a first recommendation model and outputs a corresponding dynamic feature embedded vector through an embedded layer of the first recommendation model.

The server 104 is configured to receive the dynamic feature embedding vector, calculate a matching degree of the dynamic feature embedding vector and a static feature embedding vector of at least one object available for recommendation, and recommend the object to the user to be recommended based on the matching degree of the dynamic feature embedding vector and the at least one static feature embedding vector.

In an embodiment of the present specification, the static feature embedding vector is obtained through attribute information of at least one object available for recommendation and user attribute information of the user to be recommended.

In an embodiment of the present specification, the server 104 is further configured to receive the dynamic feature embedding vector, determine, according to the dynamic feature embedding vector, a static feature embedding vector of an object corresponding to the dynamic feature embedding vector from static feature embedding vectors of at least one object available for recommendation, and recommend, to the user to be recommended, an object based on a matching degree between the dynamic feature embedding vector and the static feature embedding vector of the corresponding object.

In an embodiment of this specification, a server encodes and stores all objects available for recommendation in the same dictionary, and assuming that there are 100 objects available for recommendation, the 100 objects are encoded and stored from sequence numbers 001 and 100, and the storage result is shown in table 1:

TABLE 1

Serial number	Object information
		001	Brand A women's shoes
002	B brand skirt
		……	……
100	C brand mobile phone

If the dynamic feature embedding vector received by the server is a vector a, a ═ 0010, 0020, 0031, … …, 0091, … …, 1000 ], where 0 indicates that the user has not performed a behavior operation on a certain object, and 1 indicates that the user has performed a behavior operation on a certain object, the behavior operation described in this application may include at least one of the following items: and clicking the object, browsing the object or collecting the object by the user within a preset time period. The server executes behavior operation on the No. 003 and No. 009 commodities for the user to be recommended according to the information obtained by the dynamic feature embedded vectors, determines the static feature embedded vectors corresponding to the No. 003 and No. 009 commodities respectively from the static feature embedded vectors of 100 objects available for recommendation, performs dot product operation on the dynamic feature embedded vectors corresponding to the No. 003 and No. 009 commodities respectively to obtain the matching degrees of the user to be recommended and the No. 003 and No. 009 commodities, and sequentially recommends the objects to the user to be recommended according to the matching degrees.

In an embodiment of the present specification, the server 104 may be configured to determine an object to be recommended according to a dynamic feature embedding vector; acquiring attribute information of the object to be recommended which can be recommended; acquiring attribute information of the user to be recommended; and acquiring at least one static feature embedding vector corresponding to the attribute information of the object to be recommended and the attribute information of the user to be recommended, which are available for recommendation, through a pre-trained second recommendation model.

In an embodiment of the present specification, the second recommendation model is obtained by training with the historical behavior data of the user, the user attribute information, and the object attribute information as training samples and with the selection result of the user on the object as a training label, and the second recommendation model associates the historical behavior data of the user, the user attribute information, and the object attribute information, where the historical behavior data includes historical behaviors and behavior objects.

The method comprises the steps that a first recommendation model generates corresponding dynamic feature embedding vectors based on real-time behaviors and behavior objects of users to be recommended, a server determines at least one object to be recommended which can be recommended according to the dynamic feature embedding vectors, obtains attribute information of the at least one object to be recommended which can be recommended and the users to be recommended, inputs the attribute information of the at least one object to be recommended which can be recommended and the users to be recommended into a second recommendation model, and outputs corresponding at least one static feature embedding vector through an embedding layer of the second recommendation model.

Specifically, the first recommendation model and the second recommendation model can be obtained by training through the following steps:

and training the recommendation model by taking the historical behavior data, the user attribute information and the object attribute information of the user as training samples and taking the selection result of the user on the object as a training label.

In the process of recommending the model training, taking the historical behavior data of the user, the user attribute information and the object attribute information as training samples, taking the selection result of the user on the object as a training label, and inputting the training label into the recommending model for training. And for the historical behavior data, the user attribute information and the object attribute information of the user input each time, the embedding layer of the recommendation model can also output the characteristic expressions corresponding to the historical behavior data, the user attribute information and the object attribute information of the user. Of course, during the training process of the recommended model, the output of the embedding layer is not required to be obtained, and the output of the embedding layer is used during the use process of the recommended model.

By the training method of the recommendation model of the embodiment, the historical behavior data, the user attribute information and the object attribute information of the user are used as training samples, and the selection result of the user on the object is used as a training label to train the recommendation model, so that the association among the historical behavior data, the user attribute information and the object attribute information of the user is realized. In addition, the model is trained through the historical behavior data, the user attribute information and the object attribute information of the user, so that the relevance between the user and the recommended object can be reflected by the feature expression output in the using process of the recommended model.

It should be noted that the above describes preferred embodiments of the present disclosure, and some steps are not necessary for implementing the present disclosure. Once the model is built, the model can be repeatedly used on line for a period of time, and in order to reflect the preference of the user more accurately, the updated historical behavior data can be selected periodically to rebuild the model, but the steps are not necessary for object recommendation for the current online user each time.

In an embodiment of this specification, the first recommendation model and the second recommendation model may be the same or different, the recommendation model trained at the server may be configured at the server and the mobile for use, or different recommendation models may be configured for the server and the mobile according to actual needs, which is specifically determined according to actual needs, and the present invention is not limited herein.

In an embodiment of the present specification, the real-time behavior of the user to be recommended includes at least one of the following: the method comprises the following steps that the times of clicking the object, the times and the duration of browsing similar or related objects, and browsing and collecting the object by a user to be recommended in a preset time period; the attribute information of the user to be recommended comprises at least one of the following items: gender, age, hobbies, city, education level, and income level; the at least one object available for recommendation may be all objects in a platform applied by the object recommendation system or candidate objects pre-screened according to some principles, and the at least one object available for recommendation may be a commodity, news or video. The mobile terminal processes real-time behaviors and behavior objects of the user to be recommended by using a trained first recommendation model to generate corresponding dynamic feature embedded vectors, the server processes attribute information of the user to be recommended and attribute information of the user to be recommended to generate corresponding static feature embedded vectors, the dynamic feature embedded vectors and the static feature embedded vectors are subjected to point multiplication operation to obtain the matching degree of the user to be recommended and the object to be recommended, and the matching degree is sorted from large to small to obtain an object recommendation list.

In an embodiment of the present specification, data in the real-time behavior and behavior object of the user to be recommended, the attribute information of the user to be recommended, and the attribute information of the object to be recommended are converted into a vector form for replacement, that is, the data is subjected to an embedded mapping, which is intuitively and uninterpretable, and this step is mainly for the consideration of protecting user privacy. And the object which is interested by the user to be recommended can be predicted quickly through the generated object recommendation list, and personalized recommendation based on the object recommendation list is significant to improvement of user service experience and establishment of an intelligent recommendation system platform.

Specifically, an interaction diagram of an object recommendation method is shown in fig. 2, and includes steps 202 to 210.

Step 202, the mobile terminal obtains the real-time behavior and the behavior object of the user to be recommended.

In an embodiment of the present specification, the real-time behavior of the user to be recommended includes at least one of the following: the method comprises the steps of clicking the object by a user to be recommended within a preset time period, browsing the object for times and duration, browsing similar or related objects for times and duration, and browsing and collecting the objects.

Taking a video as an example, if a user to be recommended clicks the video a for three consecutive weeks and downloads the video a or the watching duration exceeds 30 minutes, obtaining related data, including data that the user to be recommended watches the video a 3 times, the watching time is weekday morning, the watching duration is more than 30 minutes, the video style is entertainment art, and the like, where the data includes real-time behaviors and behavior objects of the user to be recommended.

And 204, the mobile terminal generates a corresponding dynamic feature embedded vector based on the real-time behavior and the behavior object, and sends the dynamic feature embedded vector to a server.

In an embodiment of the present specification, after obtaining a real-time behavior of a user to be recommended, data corresponding to the real-time behavior of the user to be recommended is input into a trained first recommendation model, and a first feature expression, that is, a dynamic feature embedding vector, output by an embedding layer of the first recommendation model is obtained.

Still taking the video as an example, the data of the user to be recommended watching the video a in step 202, such as the number of times of watching the video a is 3, the watching time is weekday morning, the watching duration is more than 30 minutes, the video style is entertainment, is input into the first recommendation model, a dynamic feature embedding vector is obtained through calculation, and then the mobile terminal sends the dynamic feature embedding vector to the server.

Step 206, the server receives the dynamic feature embedded vector.

The attribute information of the user to be recommended comprises at least one of the following items: gender, age, hobbies, city, education level, and income level; the at least one object available for recommendation is not limited to physical commodities such as clothes and daily necessities, for the multimedia platform, news is recommended to the user to be recommended in a personalized mode according to a news list of the user to be recommended, or videos are recommended to the user to be recommended in a personalized mode according to videos watched by the user to be recommended or other interactive information related to the videos, and the specific news and the videos can be regarded as recommended objects.

Still taking the video as an example, assuming that the platform wants to perform video recommendation on the user to be recommended, selecting a plurality of videos from 20 recommendable videos to recommend the user to be recommended, the candidate videos at this time include the 20 recommendable videos, assuming that the attribute information of the user to be recommended is girl, age 25, city in beijing and research student reading, then combining the attribute information of the user to be recommended with the attribute information of the 20 recommendable videos respectively and inputting the combined information as a whole into a second recommendation model for calculation, and correspondingly generating 20 static feature embedded vectors.

And step 208, the server calculates the matching degree of the dynamic feature embedding vector and the static feature embedding vector of at least one object available for recommendation.

In an embodiment of the present specification, after receiving a dynamic feature embedding vector, a server determines an object to be recommended according to the dynamic feature embedding vector, obtains attribute information of the object to be recommended that is available for recommendation, obtains attribute information of a user to be recommended, obtains at least one static feature embedding vector corresponding to the attribute information of the object to be recommended that is available for recommendation and the attribute information of the user to be recommended through a pre-trained second recommendation model, and performs a point-product operation on the dynamic feature embedding vector and the at least one static feature embedding vector to obtain a corresponding matching degree, where the larger the matching degree is, the greater the probability that the user to be recommended is interested in the object is.

Still taking the video as an example for explanation, the above dynamic feature embedding vector and 20 static feature embedding vectors are subjected to dot product operation, and the corresponding result is obtained as follows: video 1:0.55, video 2: 0.93, … …, video 20: 0.80.

Step 210, the server recommends the object to the user to be recommended based on the matching degree of the dynamic feature embedding vector and the at least one static feature embedding vector.

In an embodiment of the present specification, an object recommendation list is generated by sorting according to the matching degrees from small to large, a matching degree threshold is preset, and an object of which the matching degree is greater than the matching degree threshold in the object recommendation list is recommended to the user to be recommended.

The obtained matching degrees corresponding to the 20 videos are sorted, and a generated video recommendation list is shown in table 2:

TABLE 2

If the matching degree threshold is set to 0.80, only video 1 and video 2 with matching degree greater than 0.80 are satisfied, and therefore the video 1 and the video 2 are sequentially recommended to the user to be recommended, and the recommending mode may be that names of the two videos and corresponding video detail links are displayed in a page currently browsed by the user to be recommended, or that related pictures and attribute information of the two videos are displayed in the page currently browsed by the user to be recommended.

In an embodiment of this specification, object recommendation for a user to be recommended may also be implemented by the following method steps:

receiving the dynamic feature embedding vector;

In an embodiment of this specification, the server encodes and stores all objects available for recommendation in the same dictionary, assuming that there are 100 objects available for recommendation, the 100 objects are encoded and stored from the sequence number 001-100, and it is assumed that after the dynamic feature embedding vector received by the server, the behavior operation is performed on the product of product No. 003 and product No. 009 for the user to be recommended according to the information obtained by the dynamic feature embedding vector, and then determining static feature embedded vectors corresponding to the No. 003 and the No. 009 commodities respectively from the static feature embedded vectors of the 100 objects available for recommendation, performing dot product operation on the dynamic feature embedded vectors and the static feature embedded vectors corresponding to the No. 003 and the No. 009 commodities to obtain the matching degrees of the user to be recommended and the No. 003 and No. 009 commodities, and sequentially recommending the objects to the user to be recommended according to the matching degrees.

In an embodiment of the description, the mobile terminal performs feed-forward calculation on the real-time behaviors and behavior objects of the user to be recommended, shares a part of calculation amount for the server, greatly improves the data processing efficiency of the server, and can predict the objects which are interested by the user to be recommended more quickly and accurately and perform real-time recommendation by generating the object recommendation list, so that better service experience can be brought to the user.

Fig. 3 shows a flowchart of an object recommendation method provided in an embodiment of the present specification, which includes steps 302 to 304.

Step 302, acquiring real-time behaviors and behavior objects of the user to be recommended.

In an embodiment of this specification, a general operation of a user to be recommended on an object may not well reflect whether the user to be recommended is interested in a certain object, and if the user to be recommended performs a behavior operation on the certain object, it may generally indicate that the user to be recommended is interested in the certain object, where the behavior operation includes at least one of the following items: and clicking the object, browsing the object or collecting the object by the user to be recommended in a preset time period.

The following is a simple example: taking a commodity as an example, within a preset time, if the number of times that a user to be recommended clicks a certain commodity exceeds 5 times, and the commodity is added into a shopping cart or collected, the relevant data of the commodity is obtained, wherein the relevant data comprises the number of times of clicking 5 times, the browsing duration of 10 minutes, the commodity style of a Korean suit for women, the commodity price of 300 yuan and the commodity material of pure cotton.

And 304, generating a corresponding dynamic feature embedded vector based on the real-time behaviors and the behavior objects, and sending the dynamic feature embedded vector to a server.

In an embodiment of the present specification, after a real-time behavior and a behavior object of a user to be recommended are obtained, the real-time behavior and the behavior object are input into a trained first recommendation model, and a first feature expression, that is, a dynamic feature embedding vector, output by an embedding layer (embedding layer) of the first recommendation model is obtained.

Still taking the commodity as an example, the data of 5 times of clicks, 10 minutes of browsing time, 300 yuan of commodity price, pure cotton material and the like in the step 302 are input into the first recommendation model, the dynamic feature embedding vector is obtained through calculation, and then the mobile terminal sends the dynamic feature embedding vector to the server.

In an embodiment of the present specification, the mobile terminal is configured with the first recommendation model, and the calculation of the real-time behavior and the behavior object of the user to be recommended is distributed to the mobile terminal by using edge calculation, so that the calculation amount of the server can be greatly reduced, and the work efficiency is effectively improved.

Fig. 4 shows a flowchart of an object recommendation method provided in an embodiment of the present specification, which includes steps 402 to 406.

Step 402, receiving the dynamic feature embedding vector.

In an embodiment of the present specification, after receiving a dynamic feature embedding vector, determining an object to be recommended according to the dynamic feature embedding vector, obtaining attribute information of the object to be recommended that is available for recommendation, obtaining attribute information of the user to be recommended, and obtaining at least one static feature embedding vector corresponding to the attribute information of the object to be recommended that is available for recommendation and the attribute information of the user to be recommended through a pre-trained second recommendation model.

In an embodiment of the present specification, the real-time behavior of the user to be recommended includes at least one of the following: the method comprises the following steps that the times of clicking the object, the times and the duration of browsing similar or related objects, and browsing and collecting the object by a user to be recommended in a preset time period; the attribute information of the user to be recommended comprises at least one of the following items: gender, age, hobbies, city, education level, and income level; the at least one object available for recommendation may be all objects in a platform applied by the object recommendation system or candidate objects pre-screened according to some principles, and the at least one object available for recommendation may be a commodity, news or video. The mobile terminal processes the real-time behaviors and behavior objects of the user to be recommended by using the trained first recommendation model to generate corresponding dynamic feature embedded vectors, and the server processes the attribute information of the object to be recommended and the attribute information of the user to be recommended to generate corresponding static feature embedded vectors.

And step 404, calculating the matching degree of the dynamic feature embedding vector and the static feature embedding vector of at least one object available for recommendation.

In an embodiment of the present specification, the object recommendation is not limited to commodity recommendation, but may also be service recommendation, for example, WeChat friend recommendation or news recommendation, and for a multimedia platform, when recommending news to a user to be recommended in a personalized manner according to a news list of the user to be recommended, or recommending videos to the user to be recommended in a personalized manner according to videos watched by the user to be recommended or other interactive information related to the videos, and the like, specific news and videos may be regarded as recommendation objects.

Still taking commodities as an example for explanation, assuming that a shopping platform wants to recommend new commodities to a user to be recommended, and selects a plurality of commodities from 100 new online commodities to recommend to the user to be recommended, the candidate commodities at this time include the 100 new online commodities, and it is assumed that attribute information of the user to be recommended is girl, age 25, city in beijing, and research student reading, and then the attribute information of the user to be recommended is respectively combined with attribute information of the 100 new online commodities and input to a second recommendation model as a whole for calculation, and 100 static feature embedded vectors are generated correspondingly.

The dynamic feature embedding vector and the static feature embedding vector may be M × N dimensional vectors, M and N are positive integers, and the vectors may include integers, decimal numbers or imaginary numbers.

In an embodiment of the present specification, to predict which object the user to be recommended may be interested in, a point-by-point operation needs to be performed on the dynamic feature embedded vector and the static feature embedded vector to obtain a corresponding matching degree, and the greater the matching degree is, the greater the probability that the user to be recommended is interested in the object is.

Still taking the commodity as an example, the above dynamic feature embedded vector and 100 static feature embedded vectors are subjected to dot product operation, and the corresponding result is obtained as follows: commercial 1:0.85, commercial 2: 0.66, … …, commercial product 100: 0.90.

And 406, recommending an object to the user to be recommended based on the matching degree of the dynamic feature embedding vector and the at least one static feature embedding vector, wherein the static feature embedding vector is obtained through attribute information of the at least one object available for recommendation and user attribute information of the user to be recommended.

In this step, the matching degree obtained by the operation is used to recommend the object to the user to be recommended, and the processing procedure is divided into the following two sub-steps, which are further described below with reference to fig. 5.

And 502, sequencing according to the matching degree to obtain an object recommendation list.

The matching degrees corresponding to the 100 commodities obtained in step 404 are sorted, and the generated commodity recommendation list is shown in table 3:

TABLE 3

And step 504, setting a matching degree threshold value, and recommending the objects with the matching degrees larger than the matching degree threshold value to the user to be recommended according to the sequence of the object recommendation list.

And recommending the objects with the matching degrees larger than the matching degree threshold value in the object recommendation list interested by the user to be recommended to the user to be recommended according to the matching degree threshold value and the sequence from large matching degree to small matching degree, and finishing personalized recommendation.

Still taking a commodity as an example, if the matching degree threshold is set to 0.85, then the commodity 100 and the commodity 1 and other commodities in the list between the commodity 100 and the commodity 1 with matching degrees greater than 0.85 are present, so the commodity 100 and the commodity 1 and other commodities in the list between the commodity 100 and the commodity 1 are sequentially recommended to the user to be recommended according to the order in the list, and the recommending manner may be to display names of two commodities and corresponding commodity detail links in a page currently browsed by the user to be recommended, or to display related pictures and attribute information of two commodities in a page currently browsed by the user to be recommended.

receiving the dynamic feature embedding vector;

In an embodiment of this specification, a server encodes and stores all objects available for recommendation in the same dictionary, and assuming that there are 100 objects available for recommendation, the 100 objects are encoded and stored from sequence number 001-: and clicking the object, browsing the object or collecting the object by the user within a preset time period. The server executes behavior operation on the No. 003 and No. 009 commodities for the user to be recommended according to the information obtained by the dynamic feature embedded vectors, determines the static feature embedded vectors corresponding to the No. 003 and No. 009 commodities respectively from the static feature embedded vectors of 100 objects available for recommendation, performs dot product operation on the dynamic feature embedded vectors corresponding to the No. 003 and No. 009 commodities to obtain the matching degrees of the user to be recommended and the No. 003 and No. 009 commodities, and sequentially recommends the objects to the user to be recommended according to the matching degrees.

The server directly obtains the dynamic characteristic embedded vector calculated at the mobile terminal, and performs point-forming operation on the dynamic characteristic embedded vector and the static characteristic embedded vector generated by calculation at the server to obtain the matching degree between the user to be recommended and the object to be recommended.

An embodiment of the present specification further provides an object recommendation apparatus, as shown in fig. 6, including a processing module 602 and a sending module 604.

The processing module 602 is configured to obtain real-time behaviors and behavior objects of the user to be recommended;

a sending module 604 configured to generate a corresponding dynamic feature embedding vector based on the real-time behavior and the behavior object, and send the dynamic feature embedding vector to a server.

In an embodiment of the present specification, the sending module 604 is further configured to convert the real-time behaviors and behavior objects into dynamic feature embedding vectors through a pre-trained first recommendation model.

In an embodiment of the present specification, a general operation of a user to be recommended on an object may not well reflect whether the user to be recommended is interested in a certain object, if the user to be recommended performs a behavior operation on the certain object, it may generally indicate that the user to be recommended is interested in the certain object, after acquiring a real-time behavior of the user to be recommended, data corresponding to the real-time behavior of the user to be recommended is input into a trained first recommendation model, and a first feature expression, that is, a dynamic feature embedding vector, output by an embedding layer of the first recommendation model is acquired.

An embodiment of the present specification further provides an object recommendation apparatus, as shown in fig. 7, including a receiving module 702, a calculating module 704, and a recommending module 706.

A receiving module 702 configured to receive a dynamic feature embedding vector;

a calculating module 704 configured to calculate a matching degree of the dynamic feature embedding vector and a static feature embedding vector of at least one object available for recommendation;

and the recommending module 706 is configured to recommend an object to the user to be recommended based on the matching degree of the dynamic feature embedding vector and at least one static feature embedding vector, wherein the static feature embedding vector is obtained through attribute information of at least one object available for recommendation and user attribute information of the user to be recommended.

In an embodiment of the present specification, the calculating module 704 is further configured to determine a static feature embedding vector of an object corresponding to the dynamic feature embedding vector from static feature embedding vectors of at least one object available for recommendation according to the dynamic feature embedding vector.

The attribute information of the at least one object available for recommendation includes at least one of: attribute information of at least one commodity available for recommendation, attribute information of at least one news available for recommendation, and attribute information of at least one video available for recommendation; the attribute information of the user to be recommended comprises at least one of the following items: gender, age, hobbies, city, education level, and income level.

In an embodiment of the present specification, the recommending module 706 is further configured to recommend an object to the user to be recommended based on a matching degree between the dynamic feature embedding vector and the static feature embedding vector of the corresponding object.

In an embodiment of the present specification, to predict which object the user to be recommended may be interested in, a point-by-point operation needs to be performed on the dynamic feature embedding vector and the static feature embedding vector to obtain a corresponding matching degree, and the greater the matching degree is, the greater the probability that the user to be recommended is interested in the object is.

An embodiment of the present specification further provides an object recommendation apparatus, and referring to fig. 8, the recommending module 706 includes:

the processing sub-module 802 is configured to sort according to the matching degree to obtain an object recommendation list;

and the recommending submodule 804 is configured to set a matching degree threshold value, and recommend the objects with the matching degrees larger than the matching degree threshold value to the user to be recommended according to the order of the object recommending list.

In an embodiment of the present specification, a real-time behavior and a behavior object of a user to be recommended are obtained, then, according to attribute information of the user to be recommended, the real-time behavior and behavior object, and attribute information of an object available for recommendation, a trained first recommendation model is used to obtain an object recommendation list in which the user to be recommended is currently interested, and an object in which a matching degree in the object recommendation list is greater than a matching degree threshold is recommended to the user to be recommended. By adopting the method provided by the application, the trained first recommendation model is used for calculation at the mobile terminal according to the real-time behaviors and behavior objects of the user to be recommended, and a part of calculated amount is shared for the server, so that the working efficiency of the server is improved, the information of the object which is currently interested in the user can be accurately recommended for the user in real time, the browsing time of the user is saved, and the service experience of the user can be effectively improved.

The embodiment of the electronic equipment provided by the application is as follows:

the present application provides an electronic device 900 comprising:

a memory 910 and a processor 920;

the memory 910 is configured to store computer-executable instructions, and the processor 920 is configured to execute the following computer-executable instructions:

acquiring real-time behaviors and behavior objects of a user to be recommended;

In one or more embodiments of the present description, the components of the electronic device 900 include, but are not limited to, the memory 910 and the processor 920. The processor 920 is coupled to the memory 910 by a bus 930.

The electronic device 900 also includes an access device 940, which access device 940 may include one or more of any type of network interface (e.g., a Network Interface Card (NIC)) whether wired or wireless, such as an IEEE802.11 Wireless Local Area Network (WLAN) wireless interface, a global microwave internet access (Wi-MAX) interface, an ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a bluetooth interface, a Near Field Communication (NFC) interface, and so forth.

In one embodiment of the present description, the above-mentioned components of the electronic device 900 and other components not shown in fig. 9 may also be connected to each other, for example, through a bus. It should be understood that the block diagram of the electronic device shown in fig. 9 is for exemplary purposes only and is not intended to limit the scope of the present disclosure. Those skilled in the art may add or replace other components as desired.

The electronic device 900 may be any type of stationary or mobile electronic device, including a mobile computer or mobile electronic device (e.g., tablet, personal digital assistant, laptop, notebook, netbook, etc.), a mobile phone (e.g., smartphone), a wearable electronic device (e.g., smartwatch, smartglasses, etc.), or other type of mobile device, or a stationary electronic device such as a desktop computer or PC. The electronic device 900 may also be a mobile or stationary server.

The second electronic device provided by the application has the following embodiments:

the present application provides an electronic device 1000 comprising:

a memory 1010 and a processor 1020;

the memory 1010 is configured to store computer-executable instructions, and the processor 1020 is configured to execute the following computer-executable instructions:

receiving a dynamic feature embedding vector;

and recommending the object to the user to be recommended based on the matching degree of the dynamic characteristic embedding vector and at least one static characteristic embedding vector.

gender, age, city and income level.

In one or more embodiments of the present description, the components of the electronic device 1000 include, but are not limited to, the memory 1010 and the processor 1020. The processor 1020 and the memory 1010 are connected by a bus 1030.

The electronic device 1000 also includes an access device 1040, and the access device 1040 may include one or more of any type of network interface (e.g., a Network Interface Card (NIC)) whether wired or wireless, such as an IEEE802.11 Wireless Local Area Network (WLAN) wireless interface, a global microwave internet access (Wi-MAX) interface, an ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a bluetooth interface, a Near Field Communication (NFC) interface, and so forth.

In one embodiment of the present description, the above-mentioned components of the electronic device 1000 and other components not shown in fig. 10 may also be connected to each other, for example, through a bus. It should be understood that the block diagram of the electronic device shown in fig. 10 is for exemplary purposes only and is not intended to limit the scope of the present disclosure. Those skilled in the art may add or replace other components as desired.

The electronic device 1000 may be any type of stationary or mobile electronic device, including a mobile computer or mobile electronic device (e.g., tablet, personal digital assistant, laptop, notebook, netbook, etc.), a mobile phone (e.g., smartphone), a wearable electronic device (e.g., smartwatch, smartglasses, etc.), or other type of mobile device, or a stationary electronic device such as a desktop computer or PC. The electronic device 1000 may also be a mobile or stationary server.

An embodiment of the present application discloses a computer-readable storage medium storing computer instructions which, when executed by a processor, implement the steps of an object recommendation method applied to a server.

An embodiment of the present application also discloses a computer readable storage medium, which stores computer instructions, and the instructions are executed by a processor to implement the steps of the object recommendation method applied to the mobile terminal.

In a typical configuration, an electronic device includes one or more processors, input/output interfaces, network interfaces, and memory.

The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by an electronic device. As defined herein, computer readable media does not include non-transitory computer readable media (transient media), such as modulated data signals and carrier waves.

As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

Claims

1. An object recommendation system, comprising:

the mobile terminal is in communication connection with the server;

2. The system of claim 1, wherein the server is further configured to receive the dynamic feature embedding vector, determine a static feature embedding vector of an object corresponding to the dynamic feature embedding vector from static feature embedding vectors of at least one object available for recommendation according to the dynamic feature embedding vector, and make object recommendations to the user to be recommended based on a degree of matching between the dynamic feature embedding vector and the static feature embedding vector of the corresponding object.

3. The system of claim 1, wherein the mobile terminal is further configured to convert the real-time behaviors and behavior objects into dynamic feature embedding vectors through a pre-trained first recommendation model.

4. The system of claim 1, wherein the server is further configured to determine an object to be recommended based on the dynamic feature embedding vector; acquiring attribute information of the object to be recommended which can be recommended; acquiring attribute information of the user to be recommended; and acquiring at least one static feature embedding vector corresponding to the attribute information of the object to be recommended and the attribute information of the user to be recommended, which are available for recommendation, through a pre-trained second recommendation model.

5. The system of claim 1, wherein the real-time behavior of the user to be recommended comprises at least one of:

the method comprises the following steps that the times of clicking an object, the times and duration of browsing the object, the times and duration of browsing similar or related objects, and browsing and collecting the object by a user to be recommended in a preset time period;

gender, age, hobbies, city, education level, and income level.

6. The system of claim 3, wherein the first recommendation model is trained by:

7. The system of claim 4, wherein the second recommendation model is trained by:

8. The system of claim 1, wherein the server is further configured to sort according to the matching degree to obtain an object recommendation list; and setting a matching degree threshold value, and recommending the objects with the matching degrees larger than the matching degree threshold value to the user to be recommended according to the sequence of the object recommendation list.

9. An object recommendation method is applied to a mobile terminal, and comprises the following steps:

acquiring real-time behaviors and behavior objects of a user to be recommended;

10. The method of claim 9, wherein the generating corresponding dynamic feature embedding vectors based on the real-time behaviors and behavior objects comprises:

11. The method of claim 10, wherein the first recommendation model is trained using:

12. The method of claim 9, wherein the real-time behavior of the user to be recommended comprises at least one of:

13. An object recommendation apparatus, comprising:

14. The apparatus of claim 13, wherein the real-time behavior of the user to be recommended comprises at least one of:

15. An object recommendation method applied to a server includes:

receiving a dynamic feature embedding vector;

16. The method of claim 15, further comprising:

receiving the dynamic feature embedding vector;

17. The method of claim 15, wherein after receiving the dynamic feature embedding vector and before calculating a degree of matching of the dynamic feature embedding vector to a static feature embedding vector of at least one object available for recommendation, further comprising:

acquiring attribute information of the user to be recommended;

18. The method of claim 17, wherein the second recommendation model is trained by:

acquiring historical behavior data, user attribute information, object attribute information and a selection result of the user on the object, wherein the behavior data comprises actions and the object of the actions;

19. The method of claim 15, wherein the attribute information of the at least one object available for recommendation comprises at least one of:

gender, age, hobbies, city, education level, and income level.

20. The method of claim 15, wherein said calculating a degree of matching of the dynamic feature embedding vector to a static feature embedding vector of at least one object available for recommendation comprises:

21. The method of claim 15, wherein the making object recommendations to the user to be recommended based on the degree of match of the dynamic feature embedding vector and at least one static feature embedding vector comprises:

22. An object recommendation apparatus, comprising:

a receiving module configured to receive a dynamic feature embedding vector;

23. The apparatus of claim 22, wherein the attribute information of the at least one object available for recommendation comprises at least one of:

gender, age, hobbies, city, education level, and income level.

24. The apparatus of claim 22, wherein the recommendation module comprises:

25. An electronic device, comprising:

a memory and a processor;

the memory is for storing computer-executable instructions and the processor is for executing the computer-executable instructions, which when executed by the processor, implement the steps of the method of any one of claims 9-12.

26. An electronic device, comprising:

a memory and a processor;

the memory is for storing computer-executable instructions and the processor is for executing the computer-executable instructions, which when executed by the processor, implement the steps of the method of any one of claims 15-21.

27. A computer-readable storage medium storing computer instructions, which when executed by a processor, perform the steps of the method of any one of claims 9 to 12.

28. A computer-readable storage medium storing computer instructions, which when executed by a processor, perform the steps of the method of any one of claims 15 to 21.