CN117688231A

CN117688231A - Resource recommendation method, electronic equipment and server

Info

Publication number: CN117688231A
Application number: CN202211065615.3A
Authority: CN
Inventors: 由书凯; 黄山山; 王宝云
Original assignee: Qingdao Jukanyun Technology Co ltd
Current assignee: Qingdao Jukanyun Technology Co ltd
Priority date: 2022-09-01
Filing date: 2022-09-01
Publication date: 2024-03-12

Abstract

After receiving a resource recommendation request, if the associated features of user vector characterization carried by the resource recommendation request do not contain behavior features of a first field and the associated features of the user vector characterization contain behavior features of a second field, the server inputs the user vector into a cross-domain recommendation model to obtain a new user vector. The associated features of the new user vector representation comprise behavior features corresponding to the first domain. And finally feeding back the resources in the first field to the electronic equipment according to the new user vector. Therefore, even if the target user does not act in the first domain, the resources in the first domain can be recommended to the target user according to the behavior characteristics of the target user in other domains, so that the cross-domain resource recommendation requirement of the behavior of the target user is met.

Description

Resource recommendation method, electronic equipment and server

Technical Field

The present disclosure relates to the field of communications technologies, and in particular, to a resource recommendation method, an electronic device, and a server.

Background

With the development of internet technology, browsing network resources is becoming one of the main ways users play daily. The user cannot timely and effectively search the information required by the user when facing a large amount of information. Therefore, the resource recommendation system has been developed, and how to accurately recommend resources to users becomes a major concern for each resource platform.

The process of recommending resources by a resource platform is generally as follows: and determining the recommendation probability of each candidate resource through a resource recommendation model according to the user characteristics of the target user and the resource content characteristics of each candidate resource. And then selecting resources with higher recommendation probability from the plurality of candidate resources, and recommending the resources to the target user.

However, in the cross-domain resource recommendation scenario, the current resource recommendation method cannot perform personalized recommendation for users with single behaviors. For example, for a user who only watched a television show, it is impossible to recommend movie-related resources to it because it has no viewing behavior in the movie field. The current resource recommendation method cannot meet the cross-domain resource recommendation requirement of a single user.

Disclosure of Invention

The application provides a resource recommendation method, electronic equipment and a server, which are used for solving the problem that the current resource recommendation method cannot meet the cross-domain resource recommendation requirement of a single user.

In a first aspect, the present embodiment provides a server for performing:

receiving a resource recommendation request input by a target user from electronic equipment, wherein the resource recommendation request carries a first domain identifier and a user vector of the target user, the first domain identifier is used for representing that the resource recommendation request is directed to the first domain, and the user vector is used for representing the association characteristic of the target user;

If the associated features of the user vector representation do not contain the behavior features corresponding to the first domain, and the associated features of the user vector representation contain the behavior features corresponding to the second domain, inputting the user vector into a cross-domain recommendation model to obtain a new user vector, wherein the associated features of the new user vector representation contain the behavior features corresponding to the first domain, and feeding back the resources of the first domain to the electronic equipment according to the new user vector, the cross-domain recommendation model is a model trained according to a plurality of user vector samples, and the associated features of the user vector sample representation simultaneously contain the behavior features corresponding to the first domain and the behavior features corresponding to the second domain;

and if the associated features of the user vector representation comprise the behavior features corresponding to the first domain, feeding back the resources of the first domain to the electronic equipment according to the user vector.

In a second aspect, the present embodiment provides an electronic device, including:

a controller for performing:

a resource recommendation request input by a target user is sent to a server, wherein the resource recommendation request carries a first domain identifier and a user vector of the target user, the first domain identifier is used for representing that the resource recommendation request aims at the first domain, and the user vector is used for representing the association characteristic of the target user;

If the associated features of the user vector representation do not contain the behavior features corresponding to the first domain, and the associated features of the user vector representation contain the behavior features corresponding to the second domain, receiving resources of the first domain, which are fed back by the server and are searched according to a new user vector, wherein the new user vector is a vector generated after the user vector is input into a cross-domain recommendation model, the associated features of the new user vector representation contain the behavior features corresponding to the first domain, the cross-domain recommendation model is a model trained according to a plurality of user vector samples, and the associated features of the user vector sample representation simultaneously contain the behavior features corresponding to the first domain and the behavior features corresponding to the second domain;

and if the associated features of the user vector representation comprise the behavior features corresponding to the first domain, receiving the resources of the first domain, which are fed back by the server and are searched according to the user vector.

In a third aspect, the present embodiment provides a resource recommendation method, where the method is applied to a server, and the method includes:

After receiving the resource recommendation request, if the associated feature of the user vector representation carried by the resource recommendation request does not contain the behavior feature of the first domain, and the associated feature of the user vector representation contains the behavior feature of the corresponding second domain, the server inputs the user vector into the cross-domain recommendation model to obtain a new user vector. The associated features of the new user vector representation comprise behavior features corresponding to the first domain. And finally feeding back the resources in the first field to the electronic equipment according to the new user vector. The cross-domain recommendation model is a model obtained by training a plurality of user vector samples, and the associated features of the user vector sample representation simultaneously comprise the behavior features corresponding to the first domain and the behavior features corresponding to the second domain. If the associated features of the user vector representation comprise the behavior features corresponding to the first domain, feeding back the resources of the first domain to the electronic equipment according to the user vector. Therefore, even if the target user does not act in the first domain, the resources in the first domain can be recommended to the target user according to the behavior characteristics of the target user in other domains, so that the cross-domain resource recommendation requirement of the behavior of the target user is met.

Drawings

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

FIG. 1 illustrates an operational scenario between a display device and a control apparatus according to some embodiments;

fig. 2 shows a hardware configuration block diagram of the control device 100 according to some embodiments;

fig. 3 illustrates a hardware configuration block diagram of a display device 200 according to some embodiments;

FIG. 4 illustrates a software configuration diagram in a display device 200 according to some embodiments;

FIG. 5 illustrates a schematic diagram of a content-based personalized recommendation method, in accordance with some embodiments;

FIG. 6 illustrates a resource recommendation system framework diagram in accordance with some embodiments;

FIG. 7 illustrates a schematic diagram of vector mapping to shared space, in accordance with some embodiments;

FIG. 8 illustrates a schematic diagram of constructing a new user vector, in accordance with some embodiments;

FIG. 9 illustrates a schematic diagram of matching resources in a shared space, in accordance with some embodiments;

FIG. 10 illustrates a user interface schematic provided by display device 200 in accordance with some embodiments;

FIG. 11 illustrates yet another user interface schematic provided by display device 200 in accordance with some embodiments;

FIG. 12 illustrates yet another user interface schematic provided by display device 200 in accordance with some embodiments;

FIG. 13 illustrates yet another user interface schematic provided by display device 200 in accordance with some embodiments;

FIG. 14 illustrates yet another schematic diagram of constructing a new user vector in accordance with some embodiments;

FIG. 15 illustrates a signaling diagram of a resource recommendation method in accordance with some embodiments;

FIG. 16 illustrates a flow diagram of a resource recommendation method in accordance with some embodiments.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of some embodiments of the present application more clear, the technical solutions of some embodiments of the present application will be clearly and completely described below with reference to specific embodiments of the present application and corresponding drawings. It will be apparent that the described embodiments are only some, but not all, of the embodiments of the present application.

It should be noted that the brief description of the terms in some embodiments of the present application is only for convenience in understanding the embodiments described below, and is not intended to limit the implementation of some embodiments of the present application. Unless otherwise indicated, these terms should be construed in their ordinary and customary meaning.

The terms first, second, third and the like in the description and in the claims and in the above-described figures are used for distinguishing between similar or similar objects or entities and not necessarily for describing a particular sequential or chronological order, unless otherwise indicated. It is to be understood that the terms so used are interchangeable under appropriate circumstances.

The terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a product or apparatus that comprises a list of elements is not necessarily limited to all elements explicitly listed, but may include other elements not expressly listed or inherent to such product or apparatus.

The term "module" refers to any known or later developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware or/and software code that is capable of performing the function associated with that element.

The term "remote control" as used herein refers to a component of a display device (such as the display devices disclosed herein) that is typically capable of being controlled wirelessly over a relatively short distance. Typically, the display device is connected with infrared and/or Radio Frequency (RF) signals and/or Bluetooth, and can also comprise functional modules such as WiFi, wireless USB, bluetooth, motion sensors and the like. For example: the hand-held touch remote controller replaces most of the physical built-in hard keys in a general remote control device with a touch screen user interface.

The electronic device in the application may be a display device, or may be other electronic devices with a voice assistant function, and the scheme is described below taking the display device as an example. Fig. 1 is a schematic diagram of an operation scenario between a display device and a terminal device provided in some embodiments of the present application. As shown in fig. 1, a user may operate the display device 200 through the mobile terminal 300 and the terminal device 100.

In some embodiments, the terminal device 100 may be a remote controller, and the communication between the remote controller and the display device may include infrared protocol communication or bluetooth protocol communication, and other short-range communication modes, etc., and the display device 200 is controlled in a wireless mode or other wired mode. The user may control the display device 200 by inputting user instructions through keys on a remote control, voice input, control panel input, etc.

In some embodiments, the mobile terminal 300 may install a software application with the display device 200, implement connection communication through a network communication protocol, and achieve the purpose of one-to-one control operation and data communication. The audio/video content displayed on the mobile terminal 300 can also be transmitted to the display device 200, so as to realize the synchronous display function.

As also shown in fig. 1, the display device 200 is also in data communication with the server 400 via a variety of communication means. The display device 200 may be permitted to make communication connections via a Local Area Network (LAN), a Wireless Local Area Network (WLAN), and other networks.

The display apparatus 200 may additionally provide a smart network television function of a computer support function, including, but not limited to, a network television, a smart television, an Internet Protocol Television (IPTV), etc., in addition to the broadcast receiving television function.

Fig. 2 is a block diagram of a hardware configuration of the display device 200 of fig. 1 provided in some embodiments of the present application.

In some embodiments, display apparatus 200 includes at least one of a modem 210, a communicator 220, a detector 230, an external device interface 240, a controller 250, a display 260, an audio output interface 270, memory, a power supply, a user interface.

In some embodiments, the detector 230 is used to collect signals of the external environment or interaction with the outside.

In some embodiments, the display 260 includes a display screen component for presenting a picture, and a driving component for driving an image display, for receiving an image signal from the controller output, for displaying video content, image content, and components of a menu manipulation interface, and a user manipulation UI interface, etc.

In some embodiments, communicator 220 is a component for communicating with external devices or servers 400 according to various communication protocol types.

In some embodiments, the controller 250 controls the operation of the display device and responds to user operations through various software control programs stored on the memory. The controller 250 controls the overall operation of the display apparatus 200.

In some embodiments, a user may input a user command through a Graphical User Interface (GUI) displayed on the display 260, and the user input interface receives the user input command through the Graphical User Interface (GUI).

In some embodiments, user interface 280 is an interface that may be used to receive control inputs.

Fig. 3 is a block diagram of a hardware configuration of the terminal device in fig. 1 according to some embodiments of the present application. As shown in fig. 3, the terminal device 100 includes a controller 111, a communication interface 130, a user input/output interface, a memory, and a power supply.

The terminal device 100 is configured to control the display device 200, and can receive an input operation instruction of a user, and convert the operation instruction into an instruction recognizable and responsive to the display device 200, functioning as an interaction between the user and the display device 200.

In some embodiments, the terminal device 100 may be a smart device. Such as: the terminal device 100 may install various applications for controlling the display device 200 according to user's needs.

In some embodiments, as shown in fig. 1, a mobile terminal 300 or other intelligent display device may serve a similar function as the terminal device 100 after installing an application that manipulates the display device 200.

The controller 111 includes a processor 112 and RAM 113 and ROM 114, a communication interface 130, and a communication bus. The controller 111 is used to control the operation and operation of the terminal device 100, and communication cooperation between the internal components and external and internal data processing functions.

The communication interface 130 enables communication of control signals and data signals with the display device 200 under the control of the controller 111. The communication interface 130 may include at least one of a WiFi chip 131, a bluetooth module 132, an NFC module 133, and other near field communication modules.

A user input/output interface 140, wherein the input interface includes at least one of a microphone 141, a touchpad 142, a sensor 143, keys 144, and other input interfaces.

In some embodiments, terminal device 100 includes at least one of a communication interface 130 and an input-output interface 140. The terminal device 100 is configured with a communication interface 130, such as: the WiFi, bluetooth, NFC, etc. modules may send the user input instruction to the display device 200 through a WiFi protocol, or a bluetooth protocol, or an NFC protocol code.

A memory 190 for storing various operation programs, data and applications for driving and controlling the terminal device 100 under the control of the controller. The memory 190 may store various control signal instructions input by a user.

And a power supply 180 for providing operation power support for the respective elements of the terminal device 100 under the control of the controller.

Fig. 4 is a schematic view of software configuration in the display device in fig. 1 provided in some embodiments of the present application, in some embodiments, the system is divided into four layers, namely, an application layer (application layer), an application framework layer (Application Framework) layer (framework layer), an Android run layer (Android run layer) and a system library layer (system runtime layer), and a kernel layer from top to bottom.

In some embodiments, at least one application program is running in the application program layer, and these application programs may be a Window (Window) program of an operating system, a system setting program, a clock program, a camera application, and the like; or may be an application developed by a third party developer.

The framework layer provides an application programming interface (Aplication Pogramming Iterface, API) and programming framework for application programs of the application layer. The application framework layer includes a number of predefined functions. The application framework layer corresponds to a processing center that decides to let the applications in the application layer act.

As shown in fig. 4, the application framework layer in some embodiments of the present application includes a manager (manager), a Content Provider (Content Provider), a View System (View System), and the like.

In some embodiments, the activity manager is to: managing the lifecycle of the individual applications and typically the navigation rollback functionality.

In some embodiments, a window manager is used to manage all window programs.

In some embodiments, the system runtime layer provides support for the upper layer, the framework layer, and when the framework layer is accessed, the android operating system runs the C/C++ libraries contained in the system runtime layer to implement the functions to be implemented by the framework layer.

In some embodiments, the kernel layer is a layer between hardware and software. As shown in fig. 4, the kernel layer contains at least one of the following drivers: audio drive, display drive, bluetooth drive, camera drive, WIFI drive, USB drive, HDMI drive, sensor drive (e.g., fingerprint sensor, temperature sensor, touch sensor, pressure sensor, etc.), and the like.

In some embodiments, the kernel layer further includes a power driver module for power management.

In some embodiments, the software programs and/or modules corresponding to the software architecture in fig. 4 are stored in the first memory or the second memory shown in fig. 2 or fig. 3.

For clarity of explanation of the technical solutions of the present application, the following lists basic concepts to which the present application relates:

the Content-based recommendation algorithm (Content-Based Recommendations) constructs a recommendation algorithm model based on the resource-related information, the user-related information and the operation behavior of the user on the resource, and provides recommendation services for the user. The resource-related information can be metadata information describing the characters of the resource, labels, user comments, manually-marked information and the like. User-related information refers to demographic information (e.g., age, gender, preferences, territories, income, etc.). The user's operational behavior on the resource may be comments, collections, praise, watch, browse, click, add shopping carts, purchase, etc. Content-based recommendation algorithms generally rely solely on the user's own behavior to provide recommendations to the user, without involving the behavior of other users.

The basic principle of the content-based recommendation algorithm is to obtain interest preferences of a user according to historical behaviors of the user, and recommend resources similar to the interest preferences of the user for the user. As shown in fig. 5, content-based personalized recommendations typically require three steps: the method comprises the steps of constructing a user characteristic representation (namely a user vector) based on user information and user operation behaviors, constructing a resource characteristic representation (namely a resource vector) based on resource information, and recommending resources (namely recommended resources) for users based on the user and the resource characteristic representation.

Firstly, the similarity between the resources needs to be calculated in advance, and then the similar resources of the resources in the user history record are recommended to the user. In some embodiments, the resource characteristics are converted to a vectorized representation, after which the similarity between two resources can be calculated by cosine similarity. The resources may be represented by tags, i.e., each tag may be associated with a set of resources, and then, based on the user's tag representation, the user's interest tags may be associated with a set of resources that are associated by the tag, and the set of resources associated by the tag may be used as a candidate set of recommendations to the user.

After the user and the resource are embedded into the same vector space, the similarity between the user and the resource can be calculated, and then the resource with high similarity is recommended to the user according to the similarity between the resource and the user. User similarity may also be calculated based on the user vector representation, recommending resources to the user that are liked by similar users.

The user characteristic representation is constructed, and the user characteristic representation can be constructed based on the operation behavior (such as clicking, purchasing, collecting, playing and the like) of the user on the resource, and can be expressed based on the demographic characteristics of the user. With the user feature representation, resources matching the user feature can be recommended to the user based on the user feature. The method for constructing the user characteristics can be as follows:

User behavior records are used as display features, specifically recording the preferences of the user for resources over time. For example, in the video domain, if a user has seen A, B, C three videos for some time, the user's behavior may be scored based on the proportion of each video user's viewing time period to the total time period of the videos. In this example, the resources operated by the user history are directly used as the characteristic representation of the user, and the resources similar to the resources operated by the user can be recommended to the user during recommendation.

The interest feature of the vector type can embed the resource into the vector space based on the information of the resource, and the resource is represented by the vector. Based on the vectorized representation of the resource, the user's interest vector can be represented by an average vector of vectors of the resources he has operated on. There are a number of strategies for representing the user interest vector, and the user's weighted preference vector may be obtained based on the user's scoring of the manipulated resources and time weighting, rather than direct averaging. In addition, a plurality of interest vectors (such as clustering the resources, taking the vector average of the resources operated by the user on a certain class as the interest vector of the user on the class) can be constructed for the user according to the similarity between the resources operated by the user, so that the multi-azimuth interest preference of the user can be better expressed. Based on the interest vector of the user and the interest vector of the resource, the preference degree of the user to the resource can be calculated according to the similarity of the vectors, and the resource is recommended for the user based on the preference degree.

Demographic characteristics of the user, information about themselves provided by the user at login, registration, information filled in by the user through operation activities, conclusions drawn through user behavior using algorithm, such as age, gender, region, income, hobbies, residence, work place, etc., are very important information. Based on this information about the user dimensions, the user features can be represented in a vectorization, the dimensions of the vector being the number of user features that can be obtained. With the user feature vector, user similarity can be calculated, and similar user favorite resources are recommended to the user.

The specific generation method of the user vector and the resource vector comprises the following steps: the labels are arranged in a certain order, each label is one dimension, and each resource can be expressed as a vector of N dimensions (N is the number of labels). If the resource contains a certain tag, the vector has a value of 1 on the component of the corresponding tag, otherwise 0, i.e. one-hot coding. It is possible that N is very large (e.g., in the video domain, N may be tens of thousands, or even hundreds of thousands of millions). The labels may be obtained by an algorithm, such as by NLP (Neuro-Linguistic Programming, neuro-linguistic) techniques to extract keywords from the text information as labels. For a picture/video, a tag may be extracted from description information (title, etc.) of the picture/video, and a related object construction tag may be extracted from the picture/video by a method of object detection. The labels may be user-labeled, i.e., the resource may be labeled as the user interacts with the resource, which is a depiction of the resource. The labels may also be manually labeled.

In some embodiments, the resource is provided with structured information, such as video fields, typically with a media asset library, where for each program there is dimensional data such as title, staffs, directors, tags, scores, regions, etc., typically stored in a relational database. Such data may have a field (also a feature) as a dimension of the vector, where the vectorized value representing each dimension is not necessarily a numerical value, but is in the form of a vectorized version, i.e., a vectorized spatial model (Vector Space Model, VSM for short).

If the resource is of text type, the text resource can be converted into a vector using TF-IDF (term frequency-inverse document frequency, a common weighting technique for information retrieval and information exploration). By word segmentation of all documents (i.e. resources) to obtain a set of all different words (assuming M words), a vector of dimension M (each word is a dimension) can be constructed for each document, and the value of the dimension of a certain word in the vector can be measured by counting the importance of each word in the document, and the measure of the importance is TF-IDF. TF is the frequency of occurrence of a word in a document, and is used for measuring the importance of the word in the document, and the more the occurrence number is, the greater the importance of the word is, the more deactivated words such as "ground", "o" and the like are naturally removed in advance.

If the resource contains picture, audio or video information, the specific vectorization method is as follows: the similarity can be easily calculated by extracting features from images, videos and audios directly by using image and audio processing technology for the images or videos to carry out vectorization representation.

In some embodiments, the process by which the resource platform recommends resources is typically: and determining the recommendation probability of each candidate resource through a resource recommendation model according to the user characteristics of the target user and the resource content characteristics of each candidate resource. And then selecting resources with higher recommendation probability from the plurality of candidate resources, and recommending the resources to the target user.

However, in the cross-domain resource recommendation scenario, the resource recommendation method in the above embodiment cannot perform personalized recommendation for the user with single behavior. For example, for a user who only watched a television show, it is impossible to recommend movie-related resources to it because it has no viewing behavior in the movie field. The resource recommendation method in the above embodiment cannot meet the requirement of cross-domain resource recommendation of a single user.

In order to solve the problems in the foregoing embodiments, an embodiment of the present application provides a resource recommendation method, where the recommendation method provided in the embodiment of the present application may be applied to the recommendation system shown in fig. 6. As shown in fig. 6, the recommendation system may include: server 400 and electronic device 200 for use by a user. The server 400 may be any form of data processing server such as a cloud server, a distributed server, or the like. The electronic device 200 may be a desktop, notebook, tablet, smart phone, display device, or the like. The server 400 may execute the recommendation method provided in the embodiments of the present application to recommend resources to a user using the electronic device 200. The electronic device 200 may receive the resource recommended by the server 400 and output the resource to the user so that the user may perform network operations, such as browsing, etc., with respect to the resource.

In order to solve the technical problem that the cross-domain resource recommendation requirement of the single user cannot be met, in the recommendation system shown in fig. 6, a server 400 receives a resource recommendation request sent by an electronic device 200, where the resource recommendation request carries a first domain identifier and a user vector of the target user, the first domain identifier is used for representing that the resource recommendation request is directed against the first domain, and the user vector is used for representing the association feature of the target user.

If the associated features of the user vector representation do not contain the behavior features corresponding to the first domain, and the associated features of the user vector representation contain the behavior features corresponding to the second domain, inputting the user vector into a cross-domain recommendation model to obtain a new user vector, wherein the associated features of the new user vector representation contain the behavior features corresponding to the first domain, and feeding back the resources of the first domain to the electronic equipment according to the new user vector, wherein the cross-domain recommendation model is a model trained according to a plurality of user vector samples, and the associated features of the user vector sample representation simultaneously contain the behavior features corresponding to the first domain and the behavior features corresponding to the second domain.

Illustratively, the user inputs a resource recommendation request on the electronic device, for example, the user wants to view the resources in the first domain, and after the resource area in the first domain clicks the get resource button, the electronic device 200 sends the resource recommendation request to the server 400. The sent resource recommendation request carries a user vector of the current user and an identification for the first domain. After receiving the resource recommendation request, the server 400 first determines whether the associated features of the user vector representation include behavior features corresponding to the first domain. I.e. whether the current user has a history of actions for the first domain, e.g. whether the resources of the first domain have been looked up.

If the associated feature of the user vector representation includes a behavior feature corresponding to the first domain, that is, the current user has a historical behavior for the first domain, the server 400 may directly feed back the resource of the first domain to the electronic device 200 according to the user vector. I.e. according to the behavior characteristics of the current user aiming at the historical behavior of the first field, acquiring the resources associated with the behavior characteristics.

If the associated features of the user vector representation do not contain behavior features corresponding to the first domain, but contain behavior features corresponding to the second domain. I.e. there is no historical behaviour for the first domain for the current user, but there is a historical behaviour for the second domain. At this time, the resources of the first domain cannot be directly obtained according to the user vector. In the application, a user vector is input into a cross-domain recommendation model to obtain a new user vector. The associated features of the new user vector representation comprise behavior features corresponding to the first domain. The cross-domain recommendation model is a model obtained by training a plurality of user vector samples, and the associated features of the user vector sample representation simultaneously comprise the behavior features corresponding to the first domain and the behavior features corresponding to the second domain.

That is, the user corresponding to the user vector sample used to train the cross-domain recommendation model has both historical behavior for the first domain and historical behavior for the second domain. The cross-domain recommendation model thus obtained is associated with both the first domain and the second domain. Therefore, after the user vector of the application is input into the cross-domain recommendation model, the behavior characteristics corresponding to the first domain can be generated for the user vector. Finally, the server 400 feeds back the resources of the first domain to the electronic device 200 according to the generated new user vector. Through comprehensive analysis of the cross-domain behavior characteristics, even if the target user does not act in the first domain, resources in the first domain can be recommended to the target user according to the behavior characteristics of the target user in other domains, so that the cross-domain resource recommendation requirement of the behavior is met compared with that of a single user.

It should be noted that, the cross-domain recommendation model of the present application may be trained on the basis of the existing model DL (Deep Learning model), and the Deep Learning model may be LSTM (Long Short Term Memory network, long and short term memory neural network) or RNN (Recurrent Neural Network ), which kind of Deep Learning model is specifically adopted in the present application is not limited.

In some embodiments, the training process of the cross-domain recommendation model may specifically be the following process:

a first user vector sample and a second user vector sample are obtained, wherein the first user vector sample is a user vector of a sample user corresponding to a first field, the second user vector sample is a user vector of a sample user corresponding to a second field, and the sample user has behavior characteristics for both the first field and the second field. And mapping the first user vector sample and the second user vector sample to a shared space by using an orthogonal mapping matrix to respectively obtain a first mapping vector sample and a second mapping vector sample.

And acquiring an optimal orthogonal mapping matrix, wherein after the first user vector sample and the second user vector sample are mapped to the shared space by utilizing the optimal orthogonal mapping matrix, the obtained first mapping vector sample and second user vector sample are closest to each other in the shared space, and the second mapping vector sample and the first user vector sample are closest to each other in the shared space. The process of obtaining the optimal orthogonal mapping matrix is iterative, e.g., can be performed by iterating continuously to minimize the objective function using the following formula. When the objective function iterates to the minimum value, the matrix at the moment is the optimal orthogonal matrix:

Wherein,for the first user vector sample, +.>For the second user vector samples, X is the orthogonal mapping matrix, X ^T Transpose of X>For the first mapping vector sample, +.>And (c) obtaining the second mapping vector sample.

And training a common recommendation model by using the optimal orthogonal mapping matrix to obtain the cross-domain recommendation model.

Some embodiments of the present application are described in detail below with reference to the accompanying drawings.

In this embodiment of the present application, the first domain is a movie domain, and the second domain is a television play domain. The sample user features are first vectorized using an existing deep learning model. Sample user features may be based on user actions on resources, such as clicking on resources in the movie domain, playing resources in the television domain.

The user with user behavior in the movie field a and the movie asset played are denoted as u _a And i _a User with user behavior in the same television play field b and played television play media is represented as u _b And i _b . Meanwhile, users who have watched movies and dramas are two-domain overlapping users in the movie domain and the drama domain. User vectors for the film domain are expressed asThe resource vector of the film field is denoted +. >The user vector in the field of television shows is denoted +.>The resource vector of the film field is denoted +.>

In the movie field, a user who simultaneously views a movie and a television show is denoted as u _ab In the field of television shows, a user who has watched both a movie and a television show is denoted as u _b And satisfy u _b ＝u _b . The vectors of overlapped users in the film field and the television play field are respectivelyAnd->

Since the vector of overlapping users in the film domain is through the depth model DL of the film domain _A Training alone, while overlapping user vectors in the field of television is through the depth model DL of the field of television _B Training alone, thus overlapping users in the film domainUser vector of (a)User vectors not equal to the overlapping user in the field of television shows

As shown in the mapping schematic diagram of fig. 7, the user vector and the resource vector in the movie domain are mapped to the shared space through the orthogonal mapping matrix X, and the user vector and the resource vector in the TV play domain are simultaneously mapped through the transposed matrix X of the orthogonal matrix X ^T Mapped to shared space. User vectors in the film domain for overlapping users after mapping to the shared space ab areThe user vector of the overlapping user in the field of television series is +.>In the model training process, because the intentions of the same user in different domains are similar, after mapping to a shared space, the user vectors +. >User vector which should be in the field of television series in shared space +.>Similarly, user vectors +.>User vector which should be in the movie field in shared space +.>Close.

For example, the loss function, for example, the objective function in the above embodiment, may be used to obtain an optimal orthogonal mapping matrix through continuous iteration, and the optimal orthogonal mapping matrix is applied to the neural network of the deep learning model, so as to obtain the final cross-domain recommendation model.

When the server 400 receives a resource recommendation request, the resource request is directed to the movie domain. If the associated feature of the user vector of the current user contains a behavioral feature for the movie domain, i.e. the current user has a historical behavior for the movie domain, movie resources are recommended to the current user according to the user vector. I.e. recommending movie resources that may be of interest to the user based on movie resources that the user has watched. If the associated feature of the user vector of the current user does not contain the behavior feature for the movie domain but contains the behavior feature for the television play domain, i.e. the current user does not have a historical behavior for the movie domain but a historical behavior for the television play domain, then movie resources that may be of interest to the user cannot be recommended according to movie resources that the user has watched.

As shown in fig. 8, a schematic diagram of a new user vector is constructed, the user vector of the current user is mapped to the shared space of the movie domain and the television domain by using the optimal orthogonal mapping matrix obtained by the above embodiment, and the user vector of the current user for the movie domain is constructed by using the user vector of the current user for the television domain. The formula for constructing the new user vector is as follows:

wherein,separately movie domain and television play domainAnd new user vector, DL, of variety domain _A 、DL _B 、DL _B Recommendation models in the movie field, the television play field and the variety field are respectively provided.

Taking the training process of the recommendation model of the film field as an example, the user vector of the film field is calculatedAnd media asset vector->Inputting a film field recommendation model DL _A Training is carried out, and after one training is finished. Generating new user vector +.>Then the new user vector in the film field is +.>Inputting a film field recommendation model DL _A Recommendation model DL for film field _A Training is performed. After the iterative training process, the final recommended model DL is obtained _A 。

And then mapping the constructed vector of the movie field and the user to a shared space, and calculating a vector distance with a resource vector of the movie field in the shared space. The vector distance here may be a euclidean distance or a cosine distance. And finally, sorting the film resources from small to large according to the vector distance, and feeding back the resources in the film field N before sorting to the electronic equipment. Therefore, even if the user does not have the historical behaviors in the film field, the user can recommend the effect of film resources to the user by utilizing the historical behaviors in other fields. For example, in the shared space shown in FIG. 9, the user vector A is closer to the movie domain resource vectors B-E, and farther from the other resource vectors, the resources B-E are recommended to the user.

In the present application, in the process of mapping the user vector to the shared space by using the orthogonal mapping matrix, orthogonal transformation can be utilizedThe characteristic that the inner product of the vector is unchanged before and after the model training process, the similarity of the vectors of different users in the model training process is reserved to the greatest extent. In addition, the mapping matrix X during orthogonal transformation can be simply transposed to obtain the inverse mapping matrix X ^T Simplifying the model training learning process. Thus, a dual learning mechanism is introduced, the two fields can learn each other and are opposite source fields and target fields, and the source fields and the target fields do not need to be manually specified. For example, when the domain to which the resource recommendation request is directed is a theatrical domain, there is no historical behavior for the theatrical domain and there is a historical behavior for the movie domain for the current user. Mapping the user vector of the current user to the shared space through the optimal orthogonal mapping matrix in the embodiment to obtain the user vector for the television, and then utilizing the obtained new user vector to output television resources to the current user. I.e. recommending the resources of the television field to the user by utilizing the historical behavior of the user in the movie field.

In some embodiments, the user vectors of other users are farther away from the user vector of the current user in the shared space ab. It should be noted that, it is also necessary to map the resource vector in the field of television and the resource vector in the field of movies to the shared space. In the model training process, the resource vector corresponding to the resource watched by the same user in the movie field and the resource vector corresponding to the resource watched in the TV play field should be pulled up as much as possible. Users viewing similar movies in different fields may be clustered together.

In some embodiments, the associated features of the user vector characterization further include demographic features of the target user, and if the associated features of the user vector characterization include behavioral features corresponding to the first domain, the user vector is a splice vector of the user behavioral vector and the user demographic vector. The characteristics of the user vector representation not only do not contain the behavior characteristics of the user in a certain field, but also comprise the basic characteristics of the user, such as the age, sex and the like of the user. After the user behavior vector and the user demographic vector are spliced, a spliced vector is obtained, and resources are recommended to the user by using the spliced vector, so that the method and the device can better meet the actual demands of the user.

The above embodiments may be applied to a scenario of media asset recommendation for a user, for example, a user interface as shown in fig. 10, including movie topics and television topics on a media asset platform. If the user has historical behaviors aiming at the film field, namely, the user watches the media assets in the film theme, the server directly feeds back similar film media assets to the display equipment according to the historical behavior record of the user. And the user does not have historical behaviors aiming at the field of the television drama, namely the user does not watch the media assets given in the theme of the television drama. The server cannot directly feed back similar movie assets to the display device according to the historical behavior record of the user.

In this embodiment, if the user does not click the media asset acquisition button in the drama theme all the time, no media asset is displayed in the drama theme. If the user clicks a media asset acquisition button in the TV theme, the display device is considered to send a resource recommendation request to the server. According to the media recommendation method of the embodiment, the server maps the user vector to the shared space to obtain a new user vector of the user aiming at the field of television. And the server searches the television play vector closest to the new user vector, and then feeds back corresponding television play media to the display equipment. As shown in the user interface of fig. 11, after receiving the tv show media assets, the display device displays tv show media assets recommended by the server in the tv show theme.

In some embodiments, if the current user has historical behavior for both the movie domain and the television domain, resources may be selectively recommended to the user according to the user's needs. For example, as shown in the user interface of fig. 12, the acquire button is set on the movie theme and the tv show theme at the same time, and if the user does not click on any acquire button, only the basic recommended resources are displayed on the movie theme and the tv show theme, that is, the movie theme displays the resources recommended according to the user's historical behavior for the movie domain, and the tv show theme displays the resources recommended according to the user's historical behavior for the tv show domain.

If the user clicks the acquire button of the movie theme and the tv show theme, an integrated movie resource, which is a movie resource recommended according to the user's historical behavior for the movie field while the user is aiming at the tv show field, is displayed on the movie theme as the user interface shown in fig. 13. Similarly, a comprehensive television resource is displayed on the television theme, wherein the comprehensive television resource is recommended according to the historical behaviors of the user aiming at the movie field, and the historical behaviors of the user aiming at the television field at the same time. By the method, the resources recommended to the user can be enriched in historical behavior characteristics of the user, and the resource recommendation effect is further improved.

In some embodiments, the current domain's resources may also be recommended using a plurality of other domain's user vectors. If the current user requests resources in the movie field and the user does not have historical behaviors aiming at the movie field, the historical behaviors of the user in the TV play field and the variety field can be simultaneously utilized to recommend the movie resources. Specifically, as shown in the mapping schematic diagram of fig. 14, the user vector and the resource vector in the movie field are mapped to the shared space ab through the orthogonal mapping matrix X, and the user vector and the resource vector in the TV field are simultaneously mapped through the transposed matrix X of the orthogonal matrix X ^T Mapped to shared space ab. In addition, the user vector and the resource vector in the film field are mapped to the shared space ac through the orthogonal mapping matrix Y, and the user vector and the resource vector in the variety field are mapped to the transpose matrix Y of the orthogonal matrix Y ^T Mapped to the shared space ac.

In this embodiment, the new user vector in the movie field can be generated simultaneously by using the optimal orthogonal mapping matrix X and the optimal orthogonal mapping matrix YThe new user vector obtained in this way can simultaneously utilize the user behavior characteristics in the TV play field and the variety field, and the finally obtained resource recommendation result is more accurate. Then the new user vector in the film field is +.>Inputting a film field recommendation model DL _A For movie collarDomain recommendation model DL _A Training is performed. After the iterative training process, the final recommended model DL is obtained _A 。

And obtaining an optimal orthogonal mapping matrix X and an optimal orthogonal mapping matrix Y by using the optimal orthogonal mapping matrix obtaining method in the embodiment. And then mapping the user vector of the current user to a shared space ab and a shared space ac respectively to obtain a new user vector related to the shared space ab and a new user vector related to the shared space ac respectively. Finally, for example, the distance minimization method obtains movie resources from the shared space ab and the shared space ac, respectively. It should be noted that, the resource vectors of movie resources obtained from the shared space ab and the shared space ac may be reordered, and finally, the resources of the top with the smallest distance may be recommended to the user. Therefore, the historical behaviors of the user in the field of television drama and the historical behaviors of the user in the field of variety are simultaneously utilized, movie resources are recommended to the user, and the personalized recommendation effect of the resources is further improved.

Based on the above embodiments, the present application provides a video three-domain recommendation method, such as a signaling diagram shown in fig. 15. The full-volume user (all users) views the video and audio comprehensive three-domain media asset on the electronic equipment, and the electronic equipment records the full-volume user behaviors and uploads the full-volume user behaviors to the server. The method and the system train the cross-domain recommendation model by utilizing the full-quantity user behaviors, and a common recommendation model and a cross-domain recommendation model exist at a server side. After the user A watches the movie, the server records the browsing, clicking, playing and other actions of the user A through the electronic equipment. When the user turns on the electronic device again, since there is a historical behavior for the movie for the user a, the server recommends movie resources to the user a according to the user's historical behavior for the movie using a general recommendation model. Because the user A does not have the historical behaviors aiming at the television drama and the variety, the server recommends television drama resources and variety resources to the user A according to the historical behaviors of the user aiming at the movies by using the cross-domain recommendation model.

The application provides a resource recommendation method. FIG. 16 is a flowchart illustrating a resource recommendation method according to an example embodiment. The resource recommendation method is applicable to the server 400 of the implementation environment shown in fig. 1. As shown in fig. 16, the resource recommendation method may include the following steps.

In step S101, a media asset recommendation request input by a target user and sent by an electronic device is received, where the resource recommendation request carries a first domain identifier and a user vector of the target user, the first domain identifier is used to characterize the resource recommendation request for the first domain, and the user vector is used to characterize an association feature of the target user.

In step S102, if the associated features of the user vector representation do not include the behavior features corresponding to the first domain, and the associated features of the user vector representation include the behavior features corresponding to the second domain, the user vector is input into a cross-domain recommendation model to obtain a new user vector, the associated features of the new user vector representation include the behavior features corresponding to the first domain, and the resources of the first domain are fed back to the electronic device according to the new user vector, wherein the cross-domain recommendation model is a model obtained by training according to a plurality of user vector samples, and the associated features of the user vector sample representation include both the behavior features corresponding to the first domain and the behavior features corresponding to the second domain;

in step S103, if the associated feature of the user vector representation includes a behavior feature corresponding to the first domain, the resource of the first domain is fed back to the electronic device according to the user vector.

Those skilled in the art will appreciate that the various aspects of the invention are illustrated and described in terms of several patentable categories or circumstances, including any novel and useful procedures, machines, products, or materials, or any novel and useful modifications thereof. Accordingly, aspects of the present application may be performed entirely by hardware, entirely by software (including firmware, resident software, micro-code, etc.) or by a combination of hardware and software. The above hardware or software may be referred to as a "data block," controller, "" engine, "" unit, "" component, "or" system. Furthermore, aspects of the present application may take the form of a computer product, comprising computer-readable program code, embodied in one or more computer-readable media.

Furthermore, the order in which the elements and sequences are presented, the use of numerical letters, or other designations are used in the application and are not intended to limit the order in which the processes and methods of the application are performed unless explicitly recited in the claims. While certain presently useful inventive embodiments have been discussed in the foregoing disclosure, by way of various examples, it is to be understood that such details are merely illustrative and that the appended claims are not limited to the disclosed embodiments, but, on the contrary, are intended to cover all modifications and equivalent arrangements included within the spirit and scope of the embodiments of the present application. For example, while the system components described above may be implemented by hardware devices, they may also be implemented solely by software solutions, such as installing the described system on an existing server or mobile device.

Likewise, it should be noted that in order to simplify the presentation disclosed herein and thereby aid in understanding one or more inventive embodiments, various features are sometimes grouped together in a single embodiment, figure, or description thereof. This method of disclosure, however, is not intended to imply that more features than are presented in the claims are required for the subject application. Indeed, less than all of the features of a single embodiment disclosed above.

Claims

1. A server, wherein the server is configured to perform:

if the associated features of the user vector representation do not contain the behavior features corresponding to the first domain, and the associated features of the user vector representation contain the behavior features corresponding to the second domain, inputting the user vector into a cross-domain recommendation model to obtain a new user vector, wherein the associated features of the new user vector representation contain the behavior features corresponding to the first domain, and feeding back the resources of the first domain to the electronic equipment according to the new user vector, the cross-domain recommendation model is a model trained according to a plurality of user vector samples, and the associated features of the user vector sample representation contain the behavior features corresponding to the first domain and the behavior features corresponding to the second domain;

2. The server according to claim 1, characterized in that in the training of the cross-domain recommendation model, the server is in particular adapted to perform:

acquiring a first user vector sample and a second user vector sample, wherein the first user vector sample is a user vector of a sample user corresponding to a first field, the second user vector sample is a user vector of a sample user corresponding to a second field, and the sample user has behavior characteristics for both the first field and the second field;

mapping the first user vector sample and the second user vector sample to a shared space by using an orthogonal mapping matrix to respectively obtain a first mapping vector sample and a second mapping vector sample;

obtaining an optimal orthogonal mapping matrix, wherein after the first user vector sample and the second user vector sample are mapped to the shared space by utilizing the optimal orthogonal mapping matrix, the obtained first mapping vector sample and second user vector sample are closest to each other in the shared space, and the second mapping vector sample and the first user vector sample are closest to each other in the shared space;

3. The server according to claim 2, characterized in that in obtaining the optimal orthogonal mapping matrix, the server is in particular configured to perform:

the optimal orthogonal mapping matrix is obtained by iterating continuously to minimize the objective function:

4. The server according to claim 2, wherein in recommending the resources of the first domain to the target user based on the new user vector, the server is specifically configured to perform:

mapping the new user vector and the resource vector in the first domain to the shared space to obtain a mapped user vector and a mapped resource vector respectively;

calculating a vector distance between the mapping user vector and the mapping resource vector;

and sequencing the resources in the first domain from near to far according to the vector distance, and feeding back the resources in the first domain, which are preset in quantity before sequencing, to the electronic equipment.

5. The server of claim 1, wherein the associated features of the user vector representation further comprise demographics of the target user, and wherein the user vector is a splice vector of a user behavior vector and a user demographics vector if the associated features of the user vector representation comprise behavioral features corresponding to the first domain.

6. An electronic device, the electronic device comprising:

a controller for performing:

if the associated features of the user vector representation do not contain the behavior features corresponding to the first domain, and the associated features of the user vector representation contain the behavior features corresponding to the second domain, receiving resources of the first domain, which are fed back by the server and are searched according to a new user vector, wherein the new user vector is a vector generated after the user vector is input into a cross-domain recommendation model, the associated features of the new user vector representation contain the behavior features corresponding to the first domain, the cross-domain recommendation model is a model trained according to a plurality of user vector samples, and the associated features of the user vector sample representation contain the behavior features corresponding to the first domain and the behavior features corresponding to the second domain;

7. A resource recommendation method, wherein the method is applied to a server, and the method comprises:

8. The method of claim 7, wherein the method further comprises: acquiring a first user vector sample and a second user vector sample, wherein the first user vector sample is a user vector of a sample user corresponding to a first field, the second user vector sample is a user vector of a sample user corresponding to a second field, and the sample user has behavior characteristics for both the first field and the second field;

9. The method of claim 8, wherein obtaining the optimal orthogonal mapping matrix comprises:

10. The method of claim 8, wherein recommending the first domain of resources to the target user based on the new user vector comprises: