CN115033777A - Data recommendation method, electronic device and storage medium - Google Patents

Data recommendation method, electronic device and storage medium Download PDF

Info

Publication number
CN115033777A
CN115033777A CN202210436960.7A CN202210436960A CN115033777A CN 115033777 A CN115033777 A CN 115033777A CN 202210436960 A CN202210436960 A CN 202210436960A CN 115033777 A CN115033777 A CN 115033777A
Authority
CN
China
Prior art keywords
recommendation
sample
target
recommended
resource
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210436960.7A
Other languages
Chinese (zh)
Inventor
王国瑞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tencent Music Entertainment Technology Shenzhen Co Ltd
Original Assignee
Tencent Music Entertainment Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tencent Music Entertainment Technology Shenzhen Co Ltd filed Critical Tencent Music Entertainment Technology Shenzhen Co Ltd
Priority to CN202210436960.7A priority Critical patent/CN115033777A/en
Publication of CN115033777A publication Critical patent/CN115033777A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9538—Presentation of query results
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25—Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/251—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25—Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/258—Client or end-user data management, e.g. managing client capabilities, user preferences or demographics, processing of multiple end-users preferences to derive collaborative data
    • H04N21/25866—Management of end-user data
    • H04N21/25891—Management of end-user data being end-user preferences
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/4508—Management of client data or end-user data
    • H04N21/4532—Management of client data or end-user data involving end-user characteristics, e.g. viewer profile, preferences
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H04N21/4668—Learning process for intelligent management, e.g. learning user preferences for recommending movies for recommending content, e.g. movies

Abstract

The embodiment of the application discloses a data recommendation method, electronic equipment and a storage medium, and is applied to the technical field of machine learning. The method comprises the following steps: the method comprises the steps of obtaining a recommendation information set, generating a recommendation feature set according to the recommendation information set, determining a first recommendation index of a target object for a resource to be recommended according to the recommendation feature set, determining a second recommendation index of the target object for the resource to be recommended under target scene information according to object attribute features and resource attribute features, determining a target recommendation index of the target object for the resource to be recommended according to the first recommendation index and the second recommendation index, and pushing the resource to be recommended to an object terminal of the target object according to the target recommendation index. By adopting the method and the device, the recommendation prediction efficiency and accuracy of the resource to be predicted can be improved.

Description

Data recommendation method, electronic device and storage medium
Technical Field
The present application relates to the field of machine learning technologies, and in particular, to a data recommendation method, an electronic device, and a storage medium.
Background
At present, the purpose of recommending tasks is mainly to predict resources to be recommended, which are interested by a user, in a target application and push the resources to be recommended to a user terminal, and the purpose is to improve the effective recommendation rate for the user. For example, a main broadcast which is interested by a user is predicted and a live broadcast room of the main broadcast is recommended, so that the click rate of the user on the live broadcast room is improved. The existing prediction mode is generally to construct a prediction model to predict the user-related attributes and the resource-related attributes to be predicted so as to realize accurate recommendation. However, in this manner, the features learned by the model are fewer only due to the foregoing two attributes, which results in low efficiency and accuracy of recommendation prediction. Therefore, how to improve the recommendation prediction efficiency and accuracy of resources to be recommended becomes a problem to be solved urgently.
Disclosure of Invention
The embodiment of the application provides a data recommendation method, an electronic device and a storage medium, which can improve the recommendation prediction efficiency and accuracy of resources to be recommended under a multi-recommendation scene.
In one aspect, an embodiment of the present application provides a data recommendation method, including:
acquiring a recommendation information set; the recommendation information set comprises object attribute information of a target object, resource attribute information of resources to be recommended and recommendation scene information of the resources to be recommended;
generating a recommendation feature set according to the recommendation information set; the recommendation feature set comprises object attribute features of the object attribute information, resource attribute features of the resource attribute information and recommendation scene features of the recommendation scene information;
determining a first recommendation index of the target object for the resource to be recommended according to the recommendation feature set; the first recommendation index represents the probability of the target object responding to the recommendation behavior of the resource to be recommended;
determining a second recommendation index of the target object for the resource to be recommended under the target scene information according to the object attribute characteristics and the resource attribute characteristics; the target scene information is at least one scene information in a plurality of scene information associated with the resource to be recommended, and the second recommendation index represents the probability that the target object responds to the recommendation behavior for the resource to be recommended;
determining a target recommendation index of the target object for the resource to be recommended according to the first recommendation index and the second recommendation index, and pushing the resource to be recommended to an object terminal of the target object according to the target recommendation index; the target recommendation index represents a target probability that a target object responds to a recommendation behavior of the resource to be recommended.
In one aspect, an embodiment of the present application provides a data recommendation device, where the device includes:
the acquisition module is used for acquiring a recommendation information set; the recommendation information set comprises object attribute information of a target object, resource attribute information of resources to be recommended and recommendation scene information of the resources to be recommended;
the processing module is used for generating a recommendation feature set according to the recommendation information set; the recommendation feature set comprises object attribute features of the object attribute information, resource attribute features of the resource attribute information and recommendation scene features of the recommendation scene information;
the determining module is used for determining a first recommendation index of the target object for the resource to be recommended according to the recommendation feature set; the first recommendation index represents the probability that the target object responds to the recommendation behavior of the resource to be recommended;
the determining module is used for determining a second recommendation index of the target object for the resource to be recommended under the target scene information according to the object attribute characteristics and the resource attribute characteristics; the target scene information is at least one scene information in a plurality of scene information associated with the resource to be recommended, and the second recommendation index represents the probability that the target object responds to the recommendation behavior for the resource to be recommended;
the determining module is used for determining a target recommendation index of the target object for the resource to be recommended according to the first recommendation index and the second recommendation index; the target recommendation index represents a target probability that a target object responds to a recommendation behavior of a resource to be recommended;
and the processing module is used for pushing the resources to be recommended to the object terminal of the target object according to the target recommendation index.
In one aspect, an embodiment of the present application provides an electronic device, which includes a processor and a memory, where the memory is used to store a computer program, and the computer program includes program instructions, and the processor is configured to call the program instructions to perform some or all of the steps in the above method.
In one aspect, the present application provides a computer-readable storage medium, which stores a computer program, where the computer program includes program instructions, and the program instructions, when executed by a processor, are used to perform some or all of the steps of the above method.
Accordingly, according to an aspect of the present application, there is provided a computer program product or computer program comprising program instructions stored in a computer readable storage medium. The processor of the computer device reads the program instructions from the computer-readable storage medium, and the processor executes the program instructions to cause the computer device to execute the data recommendation method provided above.
In the embodiment of the application, a recommendation information set can be obtained, a recommendation feature set is generated, and a first recommendation index of a target object for a resource to be recommended is determined according to the recommendation feature set; the first recommendation index reflects the interaction of various recommendation characteristics in the recommendation characteristic set; determining a second recommendation index of the target object under the target scene information according to the object attribute characteristics and the resource attribute characteristics; the second recommendation index is obtained by utilizing learning characteristics under the target scene information; determining a target recommendation index according to the first recommendation index and the second recommendation index, and pushing the resource to be recommended to an object terminal of a target object according to the target recommendation index; the target recommendation index obtained in the way can be deeply combined with various recommendation characteristics of the target recommendation index, and can be combined with learning characteristics under various recommendation scenes, so that the recommendation prediction efficiency and accuracy of resources to be predicted can be improved, and the subsequent recommendation effect based on the target recommendation index can be improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a schematic diagram of an application architecture according to an embodiment of the present application;
fig. 2 is a schematic flowchart of a data recommendation method according to an embodiment of the present application;
fig. 3a is a schematic diagram of recommended scene information provided in an embodiment of the present application;
fig. 3b is a schematic diagram of recommended scene information provided in an embodiment of the present application;
fig. 3c is a schematic diagram of recommended scene information provided in an embodiment of the present application;
fig. 4 is a schematic flowchart of a data recommendation method according to an embodiment of the present application;
fig. 5 is a scene schematic diagram of generating a recommended feature set according to an embodiment of the present application;
fig. 6a is a schematic view of a scenario for determining a fusion attribute feature according to an embodiment of the present application;
fig. 6b is a schematic view of a scenario for determining fusion attribute characteristics according to an embodiment of the present application;
fig. 7 is a schematic view of a pushing scenario of a resource to be recommended based on a target prediction model according to an embodiment of the present application;
fig. 8 is a schematic flowchart of a data recommendation method according to an embodiment of the present application;
fig. 9a is a schematic view of a model training scenario provided in an embodiment of the present application;
FIG. 9b is a diagram of a prediction model provided in an embodiment of the present application;
fig. 10 is a schematic structural diagram of a data recommendation device according to an embodiment of the present application;
fig. 11 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present application without making any creative effort belong to the protection scope of the present application.
The data recommendation method provided by the embodiment of the application is realized in the electronic equipment, and the electronic equipment can be a server or a terminal. The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing basic cloud computing services such as cloud service, a cloud database, cloud computing, a cloud function, cloud storage, Network service, cloud communication, middleware service, domain name service, security service, CDN (Content Delivery Network), big data, an artificial intelligence platform, and the like. The terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart sound box, a smart watch, a smart voice interaction device, a smart home appliance, a vehicle-mounted terminal, an aircraft, and the like, but is not limited thereto.
In some embodiments, please refer to fig. 1, where fig. 1 is a schematic diagram of an application architecture provided in the present embodiment, and the data recommendation method provided in the present application can be executed through the application architecture. As shown in fig. 1, an object terminal that may include an electronic device, a target object; the resource to be recommended belongs to a target application, and the target object can be a user (such as a user) oriented to the target application; the electronic equipment can acquire a recommendation information set according to a target object and a resource to be recommended and generate a recommendation feature set, wherein the recommendation feature set comprises a first recommendation feature of the target object, a second recommendation feature of the resource to be recommended and a third recommendation feature of a recommendation scene where the resource to be recommended is located, a first recommendation index of the target object for the resource to be recommended is acquired according to the three recommendation features, a second recommendation index is determined according to the first recommendation feature and the second recommendation feature and by combining the recommendation scene where the resource to be recommended is located, and the target recommendation index is determined according to the first recommendation index and the second recommendation index; and subsequently, a recommendation strategy of the resource to be recommended in the object terminal can be determined according to the target recommendation index, and the resource to be recommended is pushed to the object terminal based on the recommendation strategy. Here, the number of resources to be recommended may be one or more, which is merely an example. Optionally, the above process may also be performed by a target prediction model, that is, the target prediction model may be deployed in the electronic device, a recommendation feature set is generated in the target prediction model by a recommendation information set, and the first recommendation index, the second recommendation index, and the target recommendation index are obtained according to the recommendation feature set.
It should be understood that fig. 1 merely illustrates a possible application architecture of the present application, and does not limit the specific architecture of the present application, that is, the present application may also provide other forms of application architectures.
Optionally, in some embodiments, the electronic device may execute the data recommendation method according to an actual service requirement to improve the prediction efficiency and accuracy of the resource to be predicted. The electronic equipment can obtain a corresponding recommendation information set according to the target object and the object to be recommended to generate a recommendation feature set, the recommendation feature set not only comprises relevant attribute features of the target object and relevant attribute features of the resource to be predicted, but also comprises recommendation scene features corresponding to recommendation scene information of the resource to be predicted, a first recommendation index and a second recommendation index are determined according to the recommendation feature set and the recommendation scene where the object to be recommended is located, the target recommendation index is further determined, the object to be recommended can be pushed to an object terminal of the target object according to the target recommendation index, and therefore accurate pushing and recommendation effect improvement can be achieved.
The recommendation scene information may be any scene information in the scene information associated with the resource to be recommended. For example, the scene information may refer to a scene in which the resource to be recommended is recommended in an application top page of a target application to which the resource to be recommended belongs; for another example, the scene information may also refer to a scene in which the resource to be recommended is recommended on a presentation page of another resource, and the like.
Optionally, the technical scheme of the application can be applied to any recommended task. For example, the method and the device can be applied to a live broadcast recommendation task, the object to be recommended at this time can be a current online anchor, the target object can be a user using live broadcast application, the electronic device can determine a target recommendation index of the user for the anchor based on the technical scheme of the application, and the anchor is pushed to a user terminal of the user based on the target recommendation index, so that the click rate, the watching duration and the like of the user for the recommended anchor are improved. For another example, the method can also be applied to e-commerce recommendation tasks, the object to be recommended at this time can be a commodity, the target object can be a user using e-commerce applications, the electronic device can determine a target recommendation index of the user for the commodity based on the technical scheme of the application, and push the commodity to a user terminal of the user based on the target recommendation index, so as to improve the click rate, purchase rate and the like of the user for the recommended commodity.
Optionally, data related to the present application, such as a recommendation information set, a recommendation feature set, and the like, may be stored in a database, or may be stored in a blockchain, such as by a blockchain distributed system, which is not limited in the present application.
It should be understood that the foregoing scenarios are only examples, and do not constitute a limitation on application scenarios of the technical solutions provided in the embodiments of the present application, and the technical solutions of the present application may also be applied to other scenarios. For example, as can be known by those skilled in the art, with the evolution of system architecture and the emergence of new service scenarios, the technical solution provided in the embodiments of the present application is also applicable to similar technical problems.
Based on the above description, the present application embodiment proposes a data recommendation method, which may be executed by the above-mentioned electronic device. Referring to fig. 2, fig. 2 is a schematic flowchart of a data recommendation method according to an embodiment of the present application. As shown in fig. 2, a flow of the data recommendation method according to the embodiment of the present application may include the following steps:
s201, acquiring a recommendation information set.
The recommendation information set may include object attribute information of the target object, resource attribute information of the resource to be recommended, and recommendation scene information of the resource to be recommended.
In some embodiments, the target object may be a user, the object attribute information may include one or more user attributes, which may be used to characterize the relevant features of the user, and the resource attribute information of the resource to be recommended may include one or more resource attributes, which may be used to characterize the relevant features of the resource to be recommended. The specific attribute types included in the object attribute information and the resource attribute information may be set by the relevant service personnel according to the recommended tasks of the practical application, and are not limited herein. For example, taking the application to a live broadcast recommendation task as an example, the object attribute information of the target object may include basic portrait attributes (such as the age, sex, and academic calendar of the user), statistical attributes (such as the total duration of live broadcast viewing within one day or three days of the user, the total number of main broadcast flowers within three days or one week of the user, the total number of main broadcast awards within three days or one week of the user, and the like); the resource to be recommended may be an online anchor to be recommended, and thus the resource attribute information of the resource to be recommended may include a base portrait attribute (such as the age, sex, and academic calendar of the anchor), a real-time attribute (such as the total number of viewing users or the maximum number of viewing users in a live broadcast room of the anchor within one day or three days, the type of the live broadcast room, the live broadcast duration of the live broadcast room, and the like), and the like.
It is understood that in the specific implementation of the present application, related data of user information such as user age, etc. is referred to, when the above embodiments of the present application are applied to specific products or technologies, user permission or consent needs to be obtained, and the collection, use and processing of related data need to comply with related laws and regulations and standards of related countries and regions.
In some embodiments, the resource to be recommended belongs to a target application in the object terminal, the resource to be recommended may be recommended in multiple application interfaces in the target application, the multiple application interfaces may be regarded as multiple pieces of scene information that can be recommended, which is associated with the resource to be recommended in the target application, and when the resource to be recommended is recommended, the resource to be recommended may be displayed in the multiple application interfaces in different presentation forms. And the scene information related to the resource to be recommended is related to the application to which the resource to be recommended belongs.
Therefore, the recommendation scene information of the resource to be recommended is any scene information in the multiple scene information related to the resource to be recommended, the recommendation prediction processes of the resource to be recommended under different recommendation scene information are the same, and the prediction resources (such as related parameters participating in the process) used by the different recommendation scene information during recommendation prediction can be partially shared and unique, so that the recommendation prediction under the multiple recommendation scenes can be realized by the technical scheme of the application, and compared with the case that each recommendation scene is modeled independently for recommendation prediction, the recommendation scene information of the resource to be recommended can be combined with the characteristics of multiple recommendation scenes to improve the recommendation prediction effect. For example, the recommended scene information may be an application interface currently displayed in the target application, that is, the scene information that can be recommended where the resource to be recommended is currently located. The electronic device can be a background device to which the target application belongs, when the target terminal detects that the target application displays a target application interface capable of being recommended, a recommendation instruction is generated and sent to the electronic device, when the electronic device receives the recommendation instruction, the object attribute information of the target object and the resource attribute information of the resource to be recommended can be obtained according to the instruction of the recommendation instruction, and the scene information represented by the target application interface is determined as the recommendation scene information of the resource to be recommended.
The acquired resources to be recommended may be one or more, and when the number of the acquired resources to be recommended is multiple, the electronic device may construct a recommendation information set for each resource to be recommended, where each recommendation information set includes the object attribute information of the target object, the resource attribute information of one resource to be recommended, and the current recommendation scene information. And the prediction process and principle of each recommendation information set are the same, and the following description takes the recommendation information set as an example.
For example, as shown in fig. 3a to 3b, fig. 3a to 3b are schematic diagrams of recommended scenario information provided by an embodiment of the present application; taking the application to a live broadcast recommendation task as an example, the resource to be recommended is a anchor in a live broadcast application, and the anchor can be recommended on a live broadcast channel page (namely, scene information 1) of the live broadcast application, and at this time, a live broadcast room where the anchor is located is represented as a live broadcast tag (tab), and thumbnail display is performed on the live broadcast channel page, as shown in fig. 3 a; when a plurality of anchor broadcasts to be recommended exist, sequentially displaying live broadcast room thumbnails of the anchor broadcasts on the live broadcast channel page;
for another example, the anchor can also be recommended on a recommended channel page (i.e., scene information 2) of the live application, a live room detail screen of the anchor is directly displayed on the recommended channel page, and a user can enter live rooms of different recommended anchors by sliding up and down, as shown in fig. 3b, one-time sliding indicates that one-time anchor is recommended;
for another example, the recommendation of the anchor may also be made on a live broadcast viewing page (that is, scene information 3) of the live broadcast application, the target application displays a live broadcast room detail screen of the anchor a in response to a trigger operation of the user on the anchor 1, and the insertion recommendation of a live broadcast information stream (feed) may be made on the detail interface, that is, the recommendation of other anchors may be made in a specified area of the detail screen, as shown in fig. 3c, a live broadcast room thumbnail of the anchor 2 may be displayed in a floating window floating layer at a lower left corner, and the live broadcast room thumbnail may be the same as or different from the thumbnail in fig. 3 a; in this case, the recommendation method may be to make a recommendation of a anchor once within a specified time, such as recommending once every 10s, recommending 10 anchors at most, and the like.
It can be understood that the technical scheme of the application can be applied to any recommended task, the target object, the resource to be recommended and the recommended scene information can be correspondingly different according to different recommended tasks, and the recommended tasks are not limited here. For convenience of explanation, the live recommended task is taken as an example in the following description.
And S202, generating a recommendation feature set according to the recommendation information set.
The recommendation feature set may include an object attribute feature of the object attribute information, a resource attribute feature of the resource attribute information, and a recommendation scene feature of the recommendation scene information.
In one possible embodiment, the generation process and principle of each recommended feature in the recommended feature set are the same, and the object attribute feature is taken as an example here. The electronic device may specifically generate the object attribute feature according to the object attribute information by performing One-Hot encoding (One-Hot encoding) on the object attribute information to obtain an encoded feature vector, and using the encoded feature vector as the object attribute feature. For example, since the object attribute information includes an age attribute, and the age attribute is classified into a plurality of attribute categories of [ < 18,19-30,31-40, 41-50,51-60, > 60], and if the age of the target object is 24, the classified target attribute category is [19-30], the encoding feature vector corresponding to the age attribute obtained by the thermal unique encoding can be represented as [0,1,0,0,0,0 ].
Therefore, corresponding encoding feature vectors can be respectively generated by one or more attributes contained in the object attribute information, and the object attribute features can be composed of the one or more encoding feature vectors. When the hot unique code is carried out, the dividing mode aiming at the attribute can be set by related service personnel according to experience values. The specific manner in which the electronic device generates the resource attribute features and the recommended scene features may be the same as the specific manner in which the object attribute features are generated, and details are not described here.
In a possible embodiment, the electronic device may specifically generate the object attribute feature according to the object attribute information, and specifically, the electronic device may further perform One-Hot encoding (One-Hot encoding) on the object attribute information to obtain an encoded feature vector, acquire an embedded feature vector corresponding to the encoded feature vector, and use the embedded feature vector as the object attribute feature. The electronic device may obtain the embedded feature vector corresponding to the encoded feature vector by obtaining an embedded feature matrix constructed for the object attribute information, and determining the corresponding embedded feature vector from the embedded feature matrix according to the encoded feature vector and in a preset manner.
In some embodiments, the electronic device may determine the corresponding embedded feature vector from the embedded feature matrix according to the encoded feature vector and in a preset manner, specifically, multiply the encoded feature vector and the embedded feature matrix, and use the multiplication result as the corresponding embedded feature vector. Since the coded feature vector is composed of 0 and 1, the obtained embedded feature vector indicates that the column value of the element of 1 in the coded feature vector corresponds to the target row vector of the embedded feature matrix. Therefore, corresponding embedded feature vectors can be respectively generated by one or more attributes contained in the object attribute information, and the object attribute features can be composed of the one or more embedded feature vectors.
For example, since the object attribute information includes an age attribute, and a coded feature vector obtained by performing hot-unique coding based on the age attribute is [0,1,0,0,0,0], and a column value of an element having a value of 1 in the coded feature vector is 2, a vector in the 2 nd row is obtained from the embedded feature matrix as an embedded feature vector corresponding to the age attribute. The embedded characteristic matrix constructed when the embedded characteristic vector is determined can be set by related service personnel according to experience values, the embedded characteristic matrices corresponding to different attribute information can be the same or different, and the size of the embedded characteristic matrix is not limited. The specific manner in which the electronic device generates the resource attribute features and the recommended scene features may be the same as the specific manner in which the object attribute features are generated, and details are not described here.
Optionally, the object attribute feature, the resource attribute feature, and the recommended scene feature in the recommendation information set may include a coding feature vector and an embedded feature vector respectively corresponding to the object attribute feature, the resource attribute feature, and the recommended scene feature.
S203, determining a first recommendation index of the target object for the resource to be recommended according to the recommendation feature set.
The first recommendation index can represent the probability of the target object responding to the recommendation behavior of the resource to be recommended.
In some embodiments, when the target object is interested in the recommended resource to be recommended, a recommendation action may be responded to the resource to be recommended, the recommendation action may be understood as making the resource to be recommended generate an effective recommendation for the target object, and the first recommendation index may also represent the interest level of the target object in the resource to be recommended. Therefore, when the value indicated by the first recommendation index is larger, the probability that the target object responds to the recommendation action is larger, the probability that the resource to be recommended generates effective recommendation is larger, and the interest degree of the target object in the resource to be recommended is larger. The specific type of the recommended behavior is related to recommended scene information of the resource to be recommended, the recommended scene information is different, the recommended behavior may be different, and the type of the recommended behavior is not limited here.
For example, taking application to a live broadcast recommendation task as an example, if the recommended scene information is the scene information indicated in fig. 3a, the recommendation behavior may be a click behavior to an anchor live broadcast room in a recommended channel page, and at this time, the first recommendation index may represent a predicted click rate of the target object to the recommended anchor; if the recommended scene information is the scene information indicated in fig. 3b, the recommended behavior may be a viewing behavior to a live broadcast room of a main broadcast in the recommended channel page, and if the live broadcast room of the main broadcast a is watched for 3s, it indicates that the main broadcast a generates effective recommendation, and at this time, the first recommendation index may represent a predicted viewing rate of the target object to the recommended main broadcast; if the recommended scene information is the scene information indicated in fig. 3c, the recommended behavior may be a click behavior for viewing other anchor recommended on the live broadcast page, and at this time, the first recommendation index may represent a predicted click rate of the target object for the recommended anchor.
In some embodiments, the electronic device may obtain the first recommendation index according to the multiple independent recommendation features and the fused recommendation features in the recommendation feature set, so that the first recommendation index includes not only directly displayed feature information obtained based on the multiple independent recommendation features, but also implicitly displayed feature information based on the multiple fused recommendation features. The specific manner of obtaining the first recommendation index according to the multiple independent recommendation features and the fused recommendation features in the recommendation feature set may be referred to in the following description of the embodiments.
In some embodiments, the electronic device may obtain, by using a target prediction model, a first recommendation index of the target object for the resource to be recommended according to the recommendation feature set. Specifically, a target prediction model is obtained, and a first recommendation index is obtained by predicting in the target prediction model according to the object attribute characteristics, the resource attribute characteristics and the recommended scene characteristics; at this time, the object attribute feature, the resource attribute feature and the recommended scene feature all include the corresponding encoding feature vector and the corresponding embedding feature vector. When a plurality of recommended feature sets are provided, the model parameters used in the prediction of the recommended feature sets in the target prediction model may be the same, that is, when the first recommendation index is obtained through the prediction, each recommended feature set shares a part of the model network of the target prediction model to share model information. The target prediction model is trained based on the procedures described in steps S801-S804 in the following embodiments.
S204, determining a second recommendation index of the target object for the resource to be recommended under the target scene information according to the object attribute characteristics and the resource attribute characteristics.
The target scene information is at least one scene information in a plurality of scene information associated with the resource to be recommended, and the second recommendation index can represent the probability that the target object responds to the recommendation behavior associated with the target scene information on the resource to be recommended.
In some embodiments, the second recommendation index may represent the interest level of the target object in the target context information on the resource to be recommended, so that when the value indicated by the second recommendation index is larger, the probability that the target object responds to the recommendation behavior in the target context information is larger, the probability that the resource to be recommended generates an effective recommendation in the target context information is larger, and the interest level of the target object in the target context information on the resource to be recommended is larger.
In one possible implementation manner, the recommended scenario information is scenario information where the resource to be recommended is currently located in the target application, the multiple scenario information associated with the resource to be recommended includes the recommended scenario information, and the target scenario information may be determined based on the recommended scenario information.
In some embodiments, the target scenario information may be recommended scenario information, and may also be recommended scenario information and associated scenario information of the recommended scenario information in the plurality of scenario information. The related scene information of the recommended scene information may be all the scene information except the recommended scene information in the plurality of scene information, or may be designated scene information in the plurality of scene information, and the designated scene information may be set by the relevant service person according to an experience value.
For example, if the recommended scene information is the scene information 2 as indicated in fig. 3b, the recommended scene information is an application interface playing a live view, and the application interface represented by the scene information 3 as indicated in fig. 3c also plays a live view, so that the scene information 3 can be set as the associated scene information of the scene information 2.
In some embodiments, the second recommendation index may be derived from the target prediction model described above, which may include a plurality of scene prediction networks. One scene prediction network corresponds to one scene information, namely, each scene information has a unique scene prediction network.
When the target scene information is recommendation scene information, the electronic device determines, according to the object attribute features and the resource attribute features, a second recommendation index of the target object for the resource to be recommended under the target scene information, specifically, the object attribute features and the resource attribute features are predicted according to a scene prediction network corresponding to the recommendation scene information, so as to obtain the second recommendation index. The object attribute feature and the resource attribute feature which are predicted here are both corresponding embedded feature vectors.
When the target scene information is the recommended scene information and the associated scene information, the electronic device determines, according to the object attribute feature and the resource attribute feature, a second recommendation index of the target object for the resource to be recommended under the target scene information, which may be specifically, predicting the object attribute feature and the resource attribute feature according to a scene prediction network corresponding to the recommended scene information and a scene prediction network corresponding to the associated scene information, respectively, to obtain the second recommendation index. The specific manner of obtaining the second recommendation index according to the scene prediction network corresponding to the recommended scene information and the scene prediction network corresponding to the associated scene information may be referred to in the following description of the embodiments. The object attribute feature and the resource attribute feature which are predicted here are both corresponding embedded feature vectors.
It can be understood that the scene prediction network is obtained by training using the object attribute features and the resource attribute features under the respective corresponding scene information, and therefore includes the learning features unique to the respective corresponding scene information. When prediction is performed, the object attribute features and the resource attribute features can be divided into corresponding exclusive scene prediction networks based on the target scene information to obtain a second recommendation index, and the second recommendation index includes feature information of the object attribute features and the resource attribute features on the premise of learning features of the target scene information. When the target prediction model is used for predicting different recommendation feature sets, the recommendation scene information in the recommendation feature sets is different, and the used scene prediction networks are also different, namely, the model parameters used in the prediction can be different and are different from the model parameters used in the first recommendation index. Therefore, the first recommendation index and the second recommendation index are obtained based on the recommendation feature set but obtained through different model parameters under different prediction angles. Based on the above description, the target prediction model may include at least a common prediction network for predicting the first recommendation index and a plurality of scene prediction networks for predicting the second recommendation index.
S205, determining a target recommendation index of the target object for the resource to be recommended according to the first recommendation index and the second recommendation index, and pushing the resource to be recommended to an object terminal of the target object according to the target recommendation index.
The target recommendation index can be obtained through the first recommendation index and the second recommendation index in a comprehensive mode, and the prediction accuracy of the target recommendation index can be improved. The target recommendation index represents a target probability of a target object for responding to a recommendation action on resources to be recommended.
In some embodiments, the determining, by the electronic device, the target recommendation index of the target object for the resource to be recommended according to the first recommendation index and the second recommendation index may specifically be that the first recommendation index and the second recommendation index are subjected to weighted summation to obtain the target recommendation index. Wherein the coefficients of the weighted sum can be set by the relevant service personnel.
Optionally, the target recommendation index determined according to the first recommendation index and the second recommendation index may also be obtained through a target prediction model, and the target recommendation index may be output in the target prediction model through the predicted first recommendation index and the predicted second recommendation index. For example, the weighting coefficients may be obtained by training as model parameters in the target prediction model.
In some embodiments, the target recommendation index may represent the interest degree of the target object in the resource to be recommended under the current recommendation scene information, so that the recommendation policy of the resource to be recommended may be determined based on the target recommendation index, and the resource to be recommended may be pushed to the object terminal based on the recommendation policy. The recommendation strategy can be set by the relevant service personnel according to experience values. For example, when there are multiple resources to be recommended, the recommendation strategy may be sequentially pushed to the object terminal according to the descending order of the target recommendation index; for another example, the recommendation policy may be to push the resource to be recommended when the target recommendation index is greater than or equal to a preset index, and not push the resource when the target recommendation index is less than the preset index.
In the embodiment of the application, a recommendation information set can be obtained, a recommendation feature set is generated according to the recommendation information set, and a first recommendation index of a target object for a resource to be recommended is determined according to the recommendation feature set; the first recommendation index reflects the interaction of various recommendation characteristics in the recommendation characteristic set; determining a second recommendation index of the target object for the resource to be recommended under the target scene information according to the object attribute characteristics and the resource attribute characteristics; the second recommendation index is obtained by utilizing learning features under target scene information, wherein the learning features under the target scene information can be obtained by comprehensively obtaining the learning features of all scene information; determining a target recommendation index of the target object for the resource to be recommended according to the first recommendation index and the second recommendation index, and pushing the resource to be recommended to an object terminal of the target object according to the target recommendation index; the target recommendation index obtained in the way can be deeply combined with various recommendation characteristics of the target recommendation index, learning characteristics under various recommendation scenes can be combined, the recommendation prediction efficiency and accuracy of resources to be predicted can be improved, and the follow-up effect of recommending based on the target recommendation index can be improved.
Referring to fig. 4, fig. 4 is a schematic flowchart of a data recommendation method according to an embodiment of the present application, where the method may be executed by the above-mentioned electronic device. As shown in fig. 4, the flow of the data recommendation method in the embodiment of the present application may include the following steps:
s401, acquiring a recommendation information set. For a specific implementation of step S401, reference may be made to the relevant description of the foregoing embodiments, which is not described herein again.
And S402, generating a recommendation feature set according to the recommendation information set.
The recommendation feature set comprises object attribute features of the object attribute information, resource attribute features of the resource attribute information and recommendation scene features of the recommendation scene information.
In one possible embodiment, the generation process and principle of each recommended feature in the recommended feature set are the same, and each recommended feature may include an encoded feature vector and an embedded feature vector obtained from each recommended information in the recommended information set. For a specific method for generating the recommendation feature set, reference may be made to the related description of the foregoing embodiments.
For example, as shown in fig. 5, fig. 5 is a schematic view of a scenario for generating a recommended feature set according to an embodiment of the present application; wherein, the object attribute information includes an object attribute 1, and the encoding eigenvector corresponding to the object attribute 1 is [0,1,0,0,0,0] (is set as U1), so that the row vector of the 2 nd row can be obtained from the corresponding embedded matrix 1 as the embedded eigenvector corresponding to the object attribute 1 (is set as V1), and the corresponding object attribute eigenvectors are obtained as the encoding eigenvector [ U1, U2.,. Un ] and the embedded eigenvector [ V1, V2,. Vn ] based on the object attribute information (including the object attribute 1, the object attribute 2.,. object attribute n); and obtaining corresponding resource attribute characteristics as an encoding characteristic vector [ Un +1, Un +2,. once, UN ] and an embedding characteristic vector [ Vn +1, Vn +2,. once, VN ] based on the resource attribute information (including the resource attribute N +1, the resource attribute N +2,. once, and the resource attribute N); and obtaining corresponding recommended scene features as a coding feature vector [ Um ] and an embedded feature vector [ Vm ] based on the recommended scene information (set as m).
Optionally, the above process may also be obtained by using a target prediction model, that is, the target prediction model may include a feature generation layer, and the feature generation layer generates a recommended feature set according to the recommended information set, that is, the embedded feature matrix may be used as a model parameter in the target prediction model. The target prediction model is trained based on the procedures described in steps S801-S804 in the following embodiments.
And S403, determining fusion attribute features and recommendation influence values according to the recommendation feature set, and determining a first recommendation index according to the fusion attribute features and the recommendation influence values.
In a possible implementation manner, the electronic device may determine a recommendation influence value according to multiple independent recommendation features in the recommendation feature set, and perform feature fusion according to the multiple independent recommendation features to obtain a fusion attribute feature, where the fusion attribute feature may deepen feature interaction among the multiple recommendation features to obtain more feature information.
In one possible implementation, the recommendation influence value may characterize the degree of influence of various recommendation features in the recommendation information set on the target object response recommendation behavior. The electronic device determining the recommendation influence value according to the recommendation feature set may be implemented by the target prediction model. The target prediction model can also comprise a public prediction network, the public prediction network can comprise an influence prediction network, the electronic equipment can acquire all coding feature vectors from the recommended feature set, and the influence prediction network carries out prediction according to all the coding feature vectors to obtain recommended influence values. The coded feature vector is composed of 0 and 1, and the element value corresponding to only the target attribute category divided based on the attribute is 1, so that the prediction of the recommendation influence value can be performed through the coded feature vector, which means that the prediction is performed only based on the divided target attribute category.
Optionally, the recommendation influence value may be a positive number or a negative number, and when the recommendation influence value is a positive value and is larger, the positive influence generated by the plurality of recommendation characteristics is larger; conversely, when the recommended influence value is negative and larger, the negative influence generated by the various recommended features is larger. For example, if the recommendation influence value of the target object on the resource a to be recommended is determined to be 1 based on the recommendation information set 1, and the recommendation influence value of the target object on the resource B to be recommended is determined to be 2 based on the recommendation information set 2, the target object is more likely to respond to the recommendation behavior of the resource B to be recommended under the influence of the recommendation information set 2.
In one possible implementation, the fusion attribute feature represents that a new fusion recommendation feature is generated by a plurality of recommendation features of the recommendation feature set; the method can obtain the most different information from a plurality of original recommended features involved in the fusion process, and can eliminate redundant information generated by correlation among different recommended features. The electronic device determines the fusion attribute feature according to the recommended feature set, and the fusion attribute feature can be realized by the target prediction model. The target prediction model may include a common prediction network, which may include a feature fusion network.
In some embodiments, the electronic device performs feature fusion according to the recommended feature set by using a feature fusion network, and obtaining the fusion attribute feature may specifically be obtaining all embedded feature vectors in the recommended feature set, and performing feature fusion on all the embedded feature vectors.
In some embodiments, T, which is the number of the total embedded feature vectors (set to V), includes a first embedded feature vector (set to Vi) and a second embedded feature vector (set to Vj), so feature fusion of the total embedded feature vectors may specifically be according to the following fusion formula:
P ij ＝V i *W*V j T *w ij formula 1.1;
in the above, W is a common weight matrix; w is a ij Is a unique weight parameter for the first embedded feature vector and the second embedded feature vector; v j T A transposed vector that is the second embedded feature vector; i and j are positive integers from 1 to T, and i can be equal to j or not equal to j; when i is equal to j, P ij Default parameters that may be set for the associated business personnel.
If the number of elements of the first embedded feature vector is n and the number of elements of the second embedded feature vector is m, the size of the public weight matrix is nxm; and because the size of the embedded feature matrix corresponding to each attribute can be different, the obtained embedded feature vector elements can be different, a plurality of public weight matrices can be used in feature fusion, and the size of each public weight matrix is different and is determined based on the two fused embedded feature matrices. When the first embedded feature vector and the second embedded feature vector are fused, a matched common weight matrix can be selected from a plurality of common weight matrices based on the respective element numbers of the first embedded feature vector and the second embedded feature vector;
optionally, only one target weight matrix may be provided, where the size of the target weight matrix is determined based on the number of elements of a target embedded feature vector, the target embedded feature vector is a feature vector with the largest number of elements in all embedded feature vectors, and if the number of elements is Nmax, the size of the target weight matrix may be a matrix of Nmax rows and Nmax columns; when the first embedded characteristic vector and the second embedded characteristic vector are fused, nxm weight value groups are sequentially extracted from the upper left corner of the target weight matrix to form a required public weight matrix.
Based on the fusion formula, each two embedded feature vectors can obtain a fusion element, a plurality of fusion elements can be obtained through all the embedded feature vectors, and a feature matrix formed in sequence based on the plurality of fusion elements is the fusion feature matrix; the fusion formula refers to the idea of WKFM, a weighted FM (Factorization Machine). Optionally, the public weight matrix and the unique weight parameter may be set by a relevant service worker according to an empirical value, or may be both network parameters in the feature fusion network, and obtained by training a target prediction model.
For example, as shown in fig. 6a to 6b, fig. 6a to 6b are schematic diagrams of a scenario for determining a fusion attribute feature according to an embodiment of the present application; wherein, the recommended feature set includes an embedded feature vector 1 (set as V1, the number of elements is 3), an embedded feature vector 2 (set as V2, the number of elements is 4), and an embedded feature vector 3 (set as V3, the number of elements is 5); thus:
as in fig. 6a, it is assumed that there are multiple common weight matrices, the size of which is determined based on the number of elements of each embedded eigenvector, and the matrix size can be 3x4, 3x5, · and so on, respectively; when the features are fused: when the embedded characteristic vector 1 and the embedded characteristic vector 2 are fused, determining to select a public weight matrix 1 with the size of 3x4 from a plurality of public weight matrices according to the number of elements of the embedded characteristic vector 1 and the embedded characteristic vector 2; when the embedded characteristic vector 1 and the embedded characteristic vector 3 are fused, determining to select a public weight matrix 2 with the size of 3x5 from a plurality of public weight matrices according to the number of elements of the embedded characteristic vector 1 and the embedded characteristic vector 3;
as another example, as shown in fig. 6b, when there is an objective weight matrix, the objective weight matrix may be a matrix of 5 × 5; when the features are fused: when the embedded eigenvector 1 and the embedded eigenvector 2 are fused, sequentially extracting 3x4 weighted values from the target weight matrix according to the number of elements of the embedded eigenvector 1 and the embedded eigenvector 2 to form a common weight matrix 1; when the embedded eigenvector 1 and the embedded eigenvector 3 are fused, sequentially extracting 3x5 weighted values from the target weight matrix according to the number of elements of the embedded eigenvector 1 and the embedded eigenvector 3 to form a public weight matrix 2;
therefore, the embedded eigenvector 1 is sequentially associated with the public weight matrix 1, the transposed vector of the embedded eigenvector 2, the embedded eigenvector 1 and the unique weight parameter (w) corresponding to the embedded eigenvector 2 according to the above fusion formula 12 ) Multiplying to obtain a fused element (set as P) 12 ) (ii) a The embedded eigenvector 1 is sequentially associated with the unique weight parameters (w) corresponding to the common weight matrix 2, the transposed vector of the embedded eigenvector 3, the embedded eigenvector 1, and the embedded eigenvector 3 according to the above fusion formula 13 ) Multiplying to obtain a fused element (set as P) 13 ) (ii) a 9 fusion elements can be obtained based on V1, V2 and V3, and a 3x 3 fusion feature matrix is formed in sequence; the fusion feature matrix is the fusion attribute feature.
In some embodiments, the electronic device may determine the first recommendation index in a preset determination manner based on the fusion attribute feature and the recommendation impact value. The preset determination mode can be set by related service personnel when constructing the target prediction model.
For example, the preset determination mode may be that the sum of each element in the fusion attribute feature and the recommendation influence value are weighted and summed to obtain a first recommendation index; for another example, the preset determining manner may be that the fusion attribute feature is input into a fusion feature prediction network to obtain a predicted value, and the sum of the predicted value and the recommended influence value is determined as a first recommendation index, where the target prediction model may further include the fusion feature prediction network; for another example, the preset determining manner may be that the fusion attribute feature and the recommendation influence value are input to an index prediction network together to obtain the first recommendation index, where the target prediction model may further include the index prediction network; and so on. The predetermined determination method is not limited herein.
S404, determining a second recommendation index of the target object for the resource to be recommended under the target scene information according to the object attribute characteristics and the resource attribute characteristics.
The target scene information is at least one scene information in a plurality of scene information associated with the resource to be recommended, and the plurality of scene information may include recommended scene information of the resource to be recommended. And the second recommendation index represents the probability of the target object responding to the recommendation behavior of the resource to be recommended.
In a possible implementation manner, the electronic device may determine, according to the object attribute feature and the resource attribute feature, a second recommendation index at the target scene information, which may be obtained by the target prediction model, where the target prediction model may include scene prediction networks for a plurality of pieces of scene information, and one scene prediction network corresponds to one piece of scene information.
In some embodiments, the target context information may be recommendation context information; the electronic device can obtain a plurality of scene prediction networks in the target prediction model, and predict the embedded feature vector which is the object attribute feature and the embedded feature vector which is the resource attribute feature according to the scene prediction network corresponding to the recommended scene information in the plurality of scene prediction networks to obtain a second recommendation index.
In some embodiments, the target context information may include recommended context information and associated context information of the recommended context information in the plurality of context information, and the associated context information may be one or more, and may be set by the relevant service person according to an experience value. The electronic equipment can obtain a plurality of scene prediction networks in the target prediction model, and predict object attribute characteristics and resource attribute characteristics according to the scene prediction networks corresponding to the recommended scene information in the plurality of scene prediction networks and the scene prediction networks corresponding to the associated scene information, so as to obtain a first initial recommendation index of the target object for the resource to be recommended in the recommended scene information and a second initial recommendation index of the target object for the resource to be recommended in the associated scene information, wherein when the associated scene information is multiple, the corresponding second initial recommendation indexes are multiple; and obtaining a second recommendation index according to the first initial recommendation index and the second initial recommendation index.
In some embodiments, the obtaining, by the electronic device, the second recommendation index according to the first initial recommendation index and the second initial recommendation index may specifically be that the first initial recommendation index and the second initial recommendation index are subjected to weighted summation to obtain the second recommendation index. Wherein, the weighting coefficient can be set by the related service personnel according to experience values.
S405, determining a target recommendation index of the target object for the resource to be recommended according to the first recommendation index and the second recommendation index. For a specific implementation of step S405, reference may be made to the related description of the foregoing embodiments, which is not described herein again.
S406, pushing the resource to be recommended to the object terminal of the target object according to the target recommendation index.
In some embodiments, the electronic device may determine the resource to be recommended that is most likely to be interested by the target object based on the target recommendation index, thereby improving the effective recommendation rate. Therefore, the resource to be recommended is pushed to the object terminal of the target object according to the target recommendation index, wherein the target recommendation index is larger than the preset index; or the resources to be recommended can be pushed in sequence according to the sequence of the target recommendation indexes from large to small; and so on.
For example, fig. 7 is a schematic diagram of a pushing scenario of a resource to be recommended based on a target prediction model according to an embodiment of the present application; taking application to live broadcast recommendation tasks as an example; wherein:
when a target application iN a user terminal displays an application interface indicated by any scene information iN a plurality of scene information, the electronic equipment acquires a plurality of anchor casts (i1, i2, i3, and anchor.) which are on line currently, and forms a prediction pair ([ u, i1, s ], [ u, i2, s ], [ u, i3, s ], [ u, iN, s ]) with each anchor and recommended scene information(s) which are located currently respectively by the target user (u), and extracts user attribute information of the target user and anchor attribute information of each anchor from an attribute storage platform (such as a database) based on the plurality of prediction pairs, so as to form an information pair by the user attribute information and each anchor attribute information respectively, and input each information pair into a target prediction model iN sequence; generating a feature pair corresponding to each information pair (including a user attribute feature of the user attribute information, a anchor attribute feature corresponding to the anchor attribute information, and a recommended scene feature of the recommended scene information) in a target prediction model, and respectively obtaining a first recommended index and a second recommended index corresponding to each feature pair, so as to determine a target recommended index of each feature pair based on the corresponding first recommended index and second recommended index; and sequencing each anchor according to the target recommendation index of each characteristic pair to obtain sequenced anchors, and sequentially pushing the sequenced anchors or the sequenced anchors with target quantity to the user terminal of the target user.
Specifically, as shown in fig. 3a, the recommending based on the ordered anchor may be implemented by sequentially acquiring live broadcast thumbnail images of the ordered anchor (such as anchor 1, anchor 3, anchor 2, and.... that is), and sequentially displaying the live broadcast thumbnail images on a live broadcast channel page of the target application; as shown in fig. 3b, the sorted live broadcast room detail pictures of the anchor (such as anchor 1, anchor 3, anchor 2, say...) are sequentially obtained, the live broadcast room detail picture of the anchor 1 is displayed on a recommended channel page of the target application, a "click to enter the live broadcast room" touch is used to indicate that the live broadcast watching scene of the anchor 1 can be entered, and when a sliding operation of the target user is detected, the live broadcast room detail picture of the next anchor 3 is sequentially displayed on the recommended channel page to indicate that a recommendation action is executed.
In the embodiment of the application, a recommendation information set can be obtained, a recommendation feature set is generated according to the recommendation information set, a fusion attribute feature and a recommendation influence value are determined according to the recommendation feature set, and a first recommendation index is determined according to the fusion attribute feature and the recommendation influence value; the first recommendation index is obtained based on the fusion attribute characteristics and can reflect interaction of various recommendation characteristics in the recommendation characteristic set, and the first recommendation index is also obtained based on the recommendation influence value and can reflect the influence degree of the various recommendation characteristics on response recommendation behaviors; determining a second recommendation index of the target object for the resource to be recommended under the target scene information according to the object attribute characteristics and the resource attribute characteristics; the second recommendation index is obtained by using learning characteristics under the target scene information, and the learning characteristics under the target scene information can be obtained by comprehensively using the learning characteristics of all scene information; determining a target recommendation index of the target object for the resource to be recommended according to the first recommendation index and the second recommendation index, and pushing the resource to be recommended to an object terminal of the target object according to the target recommendation index; the target recommendation index obtained in the way can be deeply combined with various recommendation characteristics of the target recommendation index, learning characteristics under various recommendation scenes can be combined, the recommendation prediction efficiency and accuracy of resources to be predicted can be improved, the subsequent recommendation effect based on the target recommendation index can be improved, the recommendation scene information can be any scene information in the associated scene information, and therefore recommendation prediction under the multi-recommendation scenes can be realized.
Referring to fig. 8, fig. 8 is a flowchart illustrating a data recommendation method according to an embodiment of the present application, where the method may be executed by the electronic device mentioned above. As shown in fig. 8, a flow of the data recommendation method in the embodiment of the present application may include the following:
s801, obtaining a plurality of sample information sets.
The sample information set comprises sample object attribute information of a sample object, sample resource attribute information of a sample recommended resource and sample scene information of the sample recommended resource. The sample objects, sample recommendation resources, and sample scene information in the plurality of sample information sets may all be different.
In some embodiments, the target recommendation index may be generated by a target prediction model. Therefore, the electronic device may perform batch training on the initial prediction model by using the multiple sample information sets as a batch (batch) of sample data, and obtain a final target prediction model based on multiple rounds of batch training until convergence. The process and principle of each batch training are the same, and the kth training is taken as an example for explanation. The sample objects, sample recommendation resources, and sample scene information in the plurality of sample information sets of the kth round may all be different. The sample data included in the sample information set may be the same as the data used in the application process described in the above embodiment.
S802, inputting each sample information set into an initial prediction model, and generating a sample feature set corresponding to each sample information set based on the initial prediction model.
The sample feature set comprises a sample object attribute feature of sample object attribute information, a sample resource attribute feature of sample resource attribute information, and a sample recommendation scene feature of sample scene information of the sample resource attribute information.
In some embodiments, the initial predictive model may include a feature generation layer to be trained, and the model parameters in the feature generation layer may include an embedded feature matrix corresponding to each attribute, which may be trained by the model. The electronic device may sequentially input each sample information set into an initial prediction model, generate a coding feature vector and an embedded feature vector corresponding to each sample information set based on a feature generation layer in the initial prediction model, and use the corresponding coding feature vector and embedded feature vector as a sample feature set corresponding to each sample information set. The specific manner of generating the coded feature vector and the embedded feature vector corresponding to each sample information set based on the embedded feature matrix corresponding to each attribute in the feature generation layer may be the same as that described in the above embodiment.
And S803, respectively determining a target sample recommendation index of the sample object corresponding to each sample feature set for the sample recommendation resource corresponding to the sample object according to each sample feature set.
In some embodiments, the process and principle of determining, by the electronic device, the target sample recommendation index corresponding to each sample information set according to each sample feature set are the same, where any sample feature set in each sample feature set is taken as an example for description, where any sample feature set is taken as a target feature set, a target sample object is a sample object corresponding to the target feature set, a target sample recommendation resource is a sample recommendation resource corresponding to the target sample object, and target sample scene information is sample scene information corresponding to the target sample recommendation resource.
The electronic device may specifically determine, according to each sample feature set, a target sample recommendation index of the sample object corresponding to each sample feature set for the sample recommendation resource corresponding to the sample object, by generating, in the initial prediction model, a first sample recommendation index of the target sample object for the target sample recommendation resource according to the target feature set, and determining, according to the sample object attribute features and the sample resource attribute features in the target feature set, a second sample recommendation index of the target sample object for the target sample recommendation resource; and determining a target sample recommendation index of the target sample object for the target sample recommended resource according to the first sample recommendation index and the second sample recommendation index.
The first sample recommendation index can represent the initial sample probability of the target sample object responding to the target sample recommendation resource recommendation behavior; the second sample recommendation index can represent the initial sample probability of the target sample object responding to the recommendation behavior of the target sample recommendation resource under the target sample scene information; the target sample recommendation index may characterize a target sample probability that the target sample object responds to the target sample recommendation resource recommendation behavior.
In one possible implementation, the initial prediction model may include an initial common prediction network and a plurality of initial scene prediction networks, one initial scene prediction network corresponding to one sample scene information. The electronic device may specifically generate, in the initial prediction model, the first sample recommendation index of the target sample object for recommending resources for the target sample according to the target feature set, where the first sample recommendation index is generated according to the target feature set in the initial public prediction network. The electronic device determines, according to the sample object attribute features and the sample resource attribute features in the target feature set, a second sample recommendation index of the target sample object for the target sample recommended resources, specifically, in the target scene prediction network, the second sample recommendation index is determined according to the embedded feature vector which is the sample object attribute feature and the embedded feature vector which is the sample resource attribute feature in the target feature set; the target scene prediction network is an initial scene prediction network corresponding to target sample scene information in a plurality of initial scene prediction networks.
In some embodiments, the initial public prediction network may include an initial feature fusion network and an initial impact prediction network, and the generating, by the electronic device, the first sample recommendation index in the initial public prediction network according to the target feature set may be that, in the initial feature fusion network, feature fusion is performed on all embedded feature vectors in the target feature set to obtain a sample fusion attribute feature of the target information set; and in the initial impact prediction network, performing feature fusion on all coding feature vectors in the target feature set to obtain a sample recommendation impact value of the target information set, summing all element values in the sample fusion attribute features to obtain a sample fusion value, and taking the sum of the sample fusion value and the sample recommendation impact value as a first sample recommendation index.
S804, model parameters of the initial prediction model are corrected according to the recommended indexes of each target sample, and a target prediction model is obtained.
In some embodiments, the electronic device corrects the model parameter of the initial prediction model according to each target sample recommended index to obtain the target prediction model specifically, and corrects the network parameter of the initial public prediction network according to each target sample recommended index to obtain the public prediction network; correcting network parameters of a plurality of initial scene prediction networks according to the recommendation index of each target sample and the sample scene information of each sample recommended resource to obtain a plurality of scene prediction networks; and determining a target prediction model according to the public prediction network and the plurality of scene prediction networks. And obtaining a final target prediction model by correcting the model parameters for multiple times. If the model parameters of the initial prediction model further include an embedded feature matrix in the feature generation layer and a weighting coefficient for performing weighted summation, the model parameters of the previous part can be corrected through each target sample recommendation index, and a final target prediction model is obtained through a common prediction network, a plurality of scene prediction networks, a trained embedded feature matrix and the weighting coefficient.
In some embodiments, the electronic device corrects the network parameters of the multiple initial scene prediction networks according to each target sample recommendation index and the sample scene information of each sample recommendation resource to obtain the multiple scene prediction networks, specifically, the electronic device divides each target sample recommendation index into the multiple initial scene prediction networks according to the sample scene information of each sample recommendation resource, and sequentially corrects the network parameters of the divided initial scene prediction networks according to each target sample recommendation index to obtain the multiple scene prediction networks in the target prediction model.
For example, the plurality of initial scene prediction networks include a prediction network a, a prediction network B, and a prediction network C, where the sample scene information a in the sample information set a corresponds to the prediction network a, the sample scene information B in the sample information set B corresponds to the prediction network B, and the sample scene information C in the sample information set C corresponds to the prediction network C; a target sample recommendation index A corresponding to the sample information set A is obtained according to a first sample recommendation index A and a second sample recommendation index A generated based on the prediction network A, a target sample recommendation index B corresponding to the sample information set B is obtained according to the first sample recommendation index B and a second sample recommendation index B generated based on the prediction network B, and a target sample recommendation index C corresponding to the sample information set C is obtained according to the first sample recommendation index C and a second sample recommendation index C generated based on the prediction network C; thus, an initial common prediction network may be trained with the target sample recommendation index A, B, C, and a prediction network a may be trained with the target sample recommendation index a, a prediction network B may be trained with the target sample recommendation index B, and a prediction network C may be trained with the target sample recommendation index C.
In some embodiments, the electronic device may obtain a recommendation index label for each sample information set separately and determine a loss value for each sample information set according to the target sample recommendation index and recommendation index label for each sample information set. Therefore, the network parameter of the initial public prediction network is corrected according to the recommended index of each target sample, namely, the network parameter of the initial public prediction network is corrected according to the loss value of each sample information set. The network parameters of the plurality of initial scene prediction networks are corrected according to the target sample recommendation index and the sample scene information of the sample recommendation resource, and the network parameters of the initial scene prediction networks corresponding to the sample scene information of each sample information set are corrected according to the loss value of each sample information set. And the electronic device may determine target loss values for the plurality of sample information sets from the loss value for each sample information set and determine whether the trained model converges based on the target loss values.
For example, as shown in fig. 9a, fig. 9a is a scene schematic diagram of model training provided in the embodiment of the present application; the initial prediction model may include a feature generation layer, a first recommended index generation layer, a second recommended index generation layer, and a target recommended index generation layer; the first recommendation index generation layer may include an initial public prediction network, the second recommendation index generation layer may include a plurality of initial scene prediction networks, each initial scene prediction network corresponds to one sample scene information, the initial scene prediction networks may be formed of one or more Fully Connected Layers (FCs), the number of the Fully Connected Layers of different initial public prediction networks may be different, and the number may be set by a relevant service person according to an experience value; the initial public prediction network may include an initial feature fusion network and an initial impact prediction network, which may be comprised of one or more fully connected layers; wherein:
when a plurality of sample information sets are obtained, generating a sample feature set corresponding to each sample information set in a feature generation layer, wherein the sample feature set comprises corresponding coding feature vectors and embedding feature vectors;
respectively obtaining sample fusion attribute characteristics corresponding to each sample information set according to the embedded characteristic vectors in each sample characteristic set in the initial characteristic fusion network;
respectively obtaining a sample recommendation influence value corresponding to each sample information set according to the embedded characteristic vector in each sample characteristic set in the initial influence prediction network, and determining the sum of all elements in the sample fusion attribute characteristics corresponding to each sample information set and the sample recommendation influence value corresponding to each sample information set as a first sample recommendation index corresponding to each sample information set;
predicting sample object attribute characteristics and sample resource attribute characteristics in each sample information set according to an initial scene prediction network corresponding to sample scene information in each sample information set to obtain a second sample recommendation index corresponding to each sample information set;
respectively carrying out weighted summation on the first sample recommendation index and the second sample recommendation index corresponding to each sample information set in a target recommendation index generation layer to obtain a target sample recommendation index corresponding to each sample information set;
parameter correction is carried out on the characteristic generation layer, the first recommendation index generation layer and the target recommendation index generation layer according to the target sample recommendation index corresponding to each sample information set, and parameter correction is carried out on the initial scene prediction network corresponding to each sample information set according to the target sample recommendation index corresponding to each sample information set to obtain a target prediction model; in the application stage, if the target scene information is multiple, determining the second recommendation index further requires a weighting coefficient used for weighted summation of the first initial recommendation index and the second initial recommendation index, where the weighting coefficient may be specified in the target prediction model by a relevant service person, or may be based on further training of the target prediction model, and only the weighting coefficient of the part is modified, so as to obtain the final target prediction model.
Optionally, a gate unit may be added to the initial prediction model to distinguish different initial scene prediction networks based on different scene information, and the gate unit may determine a corresponding target scene prediction network according to the target sample scene information, introduce the sample object attribute features and the sample resource attribute features in the target feature set into the target scene prediction network, and hide the initial scene prediction networks except the target scene prediction network from the target feature set.
It is to be understood that the model structure of the initial prediction model described above is merely an example, and other forms are also possible. For example, there may be differences in the model structure constructed based on the different ways in which the first sample recommendation index, and/or the second sample recommendation index, and/or the target sample recommendation index are determined. For example, the initial public prediction network may further include an initial index prediction network, the initial index prediction network may be formed by one or more fully-connected layers, and after the sample fusion attribute feature and the sample recommendation influence value are obtained, a first sample recommendation index may be determined based on the initial index prediction network, as shown in fig. 9b, where fig. 9b is a schematic diagram of a prediction model provided in an embodiment of the present application.
Through the above process, model prediction of multiple recommended scenes can be achieved, and the obtained target prediction model may include a common network (common tower) for all recommended scenes and an individual network (individual tower) specific to a certain recommended scene, and the final output is determined by the output of the common tower and the output of the individual tower. Namely, for the prediction resource of the target prediction model, a part of resources are shared by a plurality of recommendation scenes, and a part of resources are exclusive to each recommendation scene information, so that learning features blended in during recommendation prediction can be more comprehensive, and the prediction effect is better. During training, a batch of samples can comprise a sample information set of a plurality of sample scene information, so as to realize scene tower-splitting training based on the batch data, and the batch data is not required to be only the same sample scene information, namely, the sample data is subjected to tower-splitting training in the training process of the batch, so that the samples aiming at different recommended scenes are split to train different scene prediction networks, and meanwhile, the samples of the different recommended scenes can be jointly trained to a public prediction network, so that the characteristic difference among the different recommended scenes can be fully learned, and the characteristics among the different recommended scenes can be fully utilized and cooperated.
In addition, the recommendation targets of different recommendation scenes can be aligned through combined modeling aiming at a plurality of recommendation scenes, and the recommendation targets of different recommendation scenes can be different, so that equivalent mapping can be performed on the recommendation target of each recommendation scene, the correlation of different recommendation scenes is enhanced, and the model prediction effect is better. For example, the recommendation target of the recommendation scene shown in fig. 3a is to make the target object click the resource to be recommended to generate an effective recommendation, and the recommendation target of the recommendation scene shown in fig. 3b is to make the target object watch the resource to be recommended for a specified duration to generate an effective recommendation.
For another example, taking application to a live broadcast recommendation task as an example, a great deal of tests are performed on the target prediction model obtained through the training, so that the prediction accuracy and efficiency of the target prediction model are greatly improved compared with those of the prior art, and through the recommendation mode, the click rate and the watching duration of a user in a main broadcast live broadcast room are greatly improved;
as seen in table 1 below:
data set Exposure method Clicking Duration of time
Training set 7152 thousands of 551 ten thousand 874 million
Test set 361 ten thousand 27 ten thousand 43 ten thousand
TABLE 1
The initial prediction model is trained by using a training set as a sample information set to obtain a target prediction model, and the target prediction model is tested by using a test set, so that when the exposure (aiming at the push of the anchor) is about 361 ten thousand times, the click rate of the anchor is about 27 ten thousand times, and the effective watching time is about 43 ten thousand times.
By comparing the existing prediction model, the model evaluation index AUC (area under the curve) of the target prediction model provided by the technical scheme of the application is found to be greatly improved compared with the existing prediction model; the existing prediction model is a model trained by using a plurality of batch sample data, and the batch sample data comprises a plurality of sample information sets with the same sample recommendation information. The larger the AUC, the better the training effect of the model and the higher the prediction accuracy. As seen in table 2 below:
model (model) AUC
Existing prediction models 0.8115
Object prediction model 0.8246
TABLE 2
And comparing 10% of users on the line, and finding that the click rate and the watching time obtained when the target prediction model provided by the technical scheme of the application is applied are superior to those obtained when the existing prediction model is applied;
as seen in table 3 below:
Figure BDA0003612202870000261
TABLE 3
Wherein, the click rate when the target prediction model is applied is about 1.28 percent higher than the click rate when the existing prediction model is applied; the viewing duration when the target predictive model is applied is about 2.47% higher than when the existing predictive model is applied.
In the embodiment of the application, a plurality of sample information sets can be obtained, each sample information set is input into an initial prediction model, a sample feature set corresponding to each sample information set is generated based on the initial prediction model, a target sample recommendation index of a sample object corresponding to each sample feature set for a sample recommendation resource corresponding to the sample object is determined according to each sample feature set, and a model parameter of the initial prediction model is corrected according to each target sample recommendation index to obtain a target prediction model. By the method, the target prediction model can have the prediction function of multiple recommended scenes by performing joint training by using a plurality of batch sample data containing different sample scene information, and learning characteristics under multiple recommended scenes can be combined during training, so that the prediction efficiency and accuracy of the model are improved.
While the method of the embodiments of the present application has been described in detail above, to facilitate better implementation of the above-described aspects of the embodiments of the present application, the apparatus of the embodiments of the present application is provided below accordingly.
Please refer to fig. 10, fig. 10 is a schematic structural diagram of a data recommendation device provided in the present application; the data recommendation device may be a computer program (including program code) running in the electronic equipment, for example, the data recommendation device may be an application program (such as a program capable of making data recommendations) in the electronic equipment. It should be noted that the data recommendation apparatus shown in fig. 10 is configured to perform some or all of the steps in the methods according to the embodiments shown in fig. 2, fig. 4, and fig. 8 of the present application. The data recommendation device 1000 may include: an acquisition module 1001, a processing module 1002, and a determination module 1003. Wherein:
an obtaining module 1001, configured to obtain a recommendation information set; the recommendation information set comprises object attribute information of the target object, resource attribute information of the resource to be recommended and recommendation scene information of the resource to be recommended;
the processing module 1002 is configured to generate a recommendation feature set according to the recommendation information set; the recommendation feature set comprises object attribute features of the object attribute information, resource attribute features of the resource attribute information and recommendation scene features of the recommendation scene information;
a determining module 1003, configured to determine, according to the recommendation feature set, a first recommendation index of the target object for the resource to be recommended; the first recommendation index represents the probability of the target object responding to the recommendation behavior of the resource to be recommended;
the determining module 1003 is configured to determine, according to the object attribute feature and the resource attribute feature, a second recommendation index of the target object for the resource to be recommended under the target scene information; the target scene information is at least one scene information in a plurality of scene information associated with the resource to be recommended, and the second recommendation index represents the probability that the target object responds to the recommendation behavior for the resource to be recommended;
the determining module 1003 is configured to determine a target recommendation index of the target object for the resource to be recommended according to the first recommendation index and the second recommendation index; the target recommendation index represents a target probability that a target object responds to a recommendation behavior of a resource to be recommended;
the processing module 1002 is configured to push the resource to be recommended to an object terminal of the target object according to the target recommendation index.
In a possible embodiment, the determining module 1003, when configured to determine, according to the recommendation feature set, a first recommendation index of the target object for the resource to be recommended, is specifically configured to:
performing feature fusion on the recommended feature set to obtain fusion attribute features;
determining a recommendation influence value according to the recommendation feature set; the recommendation influence value represents the influence degree of the recommendation information set on the target object response recommendation behavior;
and determining a first recommendation index according to the fusion attribute characteristics and the recommendation influence value.
In a possible implementation manner, the plurality of pieces of scene information include recommended scene information of the resource to be recommended, and the target scene information is recommended scene information;
the determining module 1003, when configured to determine, according to the object attribute feature and the resource attribute feature, a second recommendation index of the target object for the resource to be recommended under the target scene information, is specifically configured to:
acquiring a plurality of scene prediction networks; a scene prediction network corresponds to scene information;
and predicting the object attribute characteristics and the resource attribute characteristics according to the scene prediction network corresponding to the recommended scene information in the plurality of scene prediction networks to obtain a second recommendation index.
In a possible implementation manner, the plurality of pieces of scene information include recommended scene information of the resource to be recommended, and the target scene information includes the recommended scene information and associated scene information of the recommended scene information in the plurality of pieces of scene information;
the determining module 1003, when configured to determine, according to the object attribute feature and the resource attribute feature, a second recommendation index of the target object for the resource to be recommended under the target scene information, is specifically configured to:
acquiring a plurality of scene prediction networks; a scene prediction network corresponds to scene information;
predicting object attribute characteristics and resource attribute characteristics according to a scene prediction network corresponding to recommended scene information in a plurality of scene prediction networks and a scene prediction network corresponding to associated scene information respectively to obtain a first initial recommendation index of a target object for a resource to be recommended under the recommended scene information and a second initial recommendation index of the target object for the resource to be recommended under the associated scene information;
and carrying out weighted summation on the first initial recommendation index and the second initial recommendation index to obtain a second recommendation index.
In one possible embodiment, the target recommendation index is generated by a target prediction model; the processing module 1002 is further configured to:
acquiring a plurality of sample information sets; a sample information set comprises sample object attribute information of a sample object, sample resource attribute information of a sample recommended resource and sample scene information of the sample recommended resource;
inputting each sample information set into an initial prediction model, and generating a sample characteristic set corresponding to each sample information set based on the initial prediction model; a sample feature set comprises a sample object attribute feature of sample object attribute information, a sample resource attribute feature of sample resource attribute information, and a sample recommendation scene feature of sample scene information of the sample resource attribute information;
respectively determining a target sample recommendation index of the sample object corresponding to each sample feature set for the sample recommendation resource corresponding to the sample object according to each sample feature set;
and correcting the model parameters of the initial prediction model according to the recommended index of each target sample to obtain a target prediction model.
In a possible embodiment, the initial prediction model includes an initial common prediction network and a plurality of initial scene prediction networks, and one initial scene prediction network corresponds to one sample scene information;
the processing module 1002, when configured to respectively determine, according to each sample feature set, a target sample recommendation index of the sample object corresponding to each sample feature set for the sample recommendation resource corresponding to the sample object, is specifically configured to:
in an initial public prediction network, generating a first sample recommendation index of a target sample object for recommending resources to a target sample according to a target feature set; the target feature set is any one of the sample feature sets, the target sample object is a sample object corresponding to the target feature set, and the target sample recommendation resource is a sample recommendation resource corresponding to the target sample object;
in a target scene prediction network, determining a second sample recommendation index of a target sample object for a target sample recommendation resource according to the sample object attribute features in the target feature set and the sample resource attribute features in the target feature set; the target scene prediction network is an initial scene prediction network corresponding to target sample scene information in a plurality of initial scene prediction networks, and the target sample scene information is sample scene information corresponding to target sample recommended resources;
and determining a target sample recommendation index of the target sample object for the target sample recommended resource according to the first sample recommendation index and the second sample recommendation index.
In one possible embodiment, the initial prediction model includes an initial common prediction network and a plurality of initial scene prediction networks; the processing module 1002 is specifically configured to, when being configured to correct the model parameters of the initial prediction model according to each target sample recommendation index to obtain a target prediction model:
correcting network parameters of the initial public prediction network according to the recommended index of each target sample to obtain a public prediction network;
correcting network parameters of a plurality of initial scene prediction networks according to each target sample recommendation index and the sample scene information of each sample recommendation resource to obtain a plurality of scene prediction networks;
and determining a target prediction model according to the public prediction network and the scene prediction networks.
In one possible implementation, one initial scene prediction network corresponds to one sample scene information; the processing module 1002 is specifically configured to, when being configured to correct network parameters of multiple initial scene prediction networks according to each target sample recommendation index and sample scene information of each sample recommendation resource to obtain multiple scene prediction networks:
according to the sample scene information of each sample recommended resource, dividing each target sample recommended index into a plurality of initial scene prediction networks;
and correcting the network parameters of the divided initial scene prediction networks according to the recommended index of each target sample to obtain a plurality of scene prediction networks.
According to an embodiment of the present application, the units in the data recommendation device shown in fig. 10 may be respectively or entirely combined into one or several other units to form the data recommendation device, or some unit(s) thereof may be further split into multiple functionally smaller units to form the data recommendation device, which may achieve the same operation without affecting the achievement of the technical effect of the embodiment of the present application. The units are divided based on logic functions, and in practical application, the functions of one unit can be realized by a plurality of units, or the functions of a plurality of units can be realized by one unit.
In other embodiments of the present application, the data recommendation device may also include other units, and in practical applications, these functions may also be implemented by assistance of other units, and may be implemented by cooperation of multiple units. According to another embodiment of the present application, the data recommendation apparatus shown in fig. 10 may be constructed by running a computer program (including program codes) capable of executing the steps involved in the respective methods shown in fig. 2, fig. 4 and fig. 8 on a general-purpose computing device such as a computer including a processing element such as a Central Processing Unit (CPU), a random access storage medium (RAM), a read-only storage medium (ROM) and a storage element, and implementing the data recommendation method of the embodiment of the present application. The computer program may be recorded on a computer-readable recording medium, for example, and loaded and executed in the electronic apparatus described above via the computer-readable recording medium.
In the embodiment of the application, an acquisition module acquires a recommendation information set; the processing module generates a recommendation feature set according to the recommendation information set; the determining module determines a first recommendation index of the target object for the resource to be recommended according to the recommendation feature set; the determining module determines a second recommendation index of the target object for the resource to be recommended under the target scene information according to the object attribute characteristics and the resource attribute characteristics; the determining module determines a target recommendation index of the target object for the resource to be recommended according to the first recommendation index and the second recommendation index; and the processing module pushes the resource to be recommended to the object terminal of the target object according to the target recommendation index. By the aid of the device, the interaction of various recommendation characteristics in the recommendation characteristic set is reflected by the obtained first recommendation index, the obtained second recommendation index is obtained by learning characteristics under target scene information, the learning characteristics under the target scene information can be obtained by the learning characteristics of all scene information comprehensively, the obtained target recommendation index can be combined with various recommendation characteristics of the target recommendation index more deeply, the learning characteristics under various recommendation scenes can be combined, the recommendation prediction efficiency and accuracy of resources to be predicted can be improved, and the subsequent recommendation effect based on the target recommendation index can be improved.
The functional modules in the embodiments of the present application may be integrated into one processing module, or each module may exist alone physically, or two or more modules are integrated into one module. The integrated module may be implemented in a form of hardware, or may be implemented in a form of software functional module, which is not limited in this application.
Referring to fig. 11, fig. 11 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure. As shown in fig. 11, the electronic device 1100 includes: a processor 1101, a communication interface 1102, and a computer-readable storage medium 1103. The processor 1101, the communication interface 1102, and the computer-readable storage medium 1103 may be connected by a bus or other means. The communication interface 1102 is used for receiving and transmitting data, among other things. A computer readable storage medium 1103 may be stored in the memory of the electronic device, the computer readable storage medium 1103 being used to store a computer program comprising program instructions, the processor 1101 being used to execute the program instructions stored by the computer readable storage medium 1103. The processor 1101 (or CPU) is a computing core and a control core of the electronic device, and is adapted to implement one or more instructions, and in particular, is adapted to load and execute the one or more instructions so as to implement a corresponding method flow or a corresponding function.
Embodiments of the present application also provide a computer-readable storage medium (Memory), which is a Memory device in an electronic device and is used for storing programs and data. It is understood that the computer readable storage medium herein may include both a built-in storage medium in the electronic device and, of course, an extended storage medium supported by the electronic device. The computer readable storage medium provides a memory space that stores a processing system of the electronic device. Also stored in the memory space are one or more instructions, which may be one or more computer programs (including program code), suitable for loading and execution by the processor 1101. It should be noted that the computer-readable storage medium may be a high-speed RAM memory, or may be a non-volatile memory (non-volatile memory), such as at least one disk memory; optionally, there may also be at least one computer readable storage medium located remotely from the aforementioned processor.
In one embodiment, the computer-readable storage medium has one or more instructions stored therein; one or more instructions stored in the computer-readable storage medium are loaded and executed by the processor 1101 to implement the corresponding steps in the above-described document processing method embodiments; in particular implementations, one or more instructions in the computer-readable storage medium are loaded by processor 1101 and perform the following steps:
acquiring a recommendation information set; the recommendation information set comprises object attribute information of a target object, resource attribute information of resources to be recommended and recommendation scene information of the resources to be recommended;
generating a recommendation feature set according to the recommendation information set; the recommendation feature set comprises object attribute features of the object attribute information, resource attribute features of the resource attribute information and recommendation scene features of the recommendation scene information;
determining a first recommendation index of the target object for the resource to be recommended according to the recommendation feature set; the first recommendation index represents the probability of the target object responding to the recommendation behavior of the resource to be recommended;
determining a second recommendation index of the target object for the resource to be recommended under the target scene information according to the object attribute characteristics and the resource attribute characteristics; the target scene information is at least one scene information in a plurality of scene information associated with the resource to be recommended, and the second recommendation index represents the probability that the target object responds to the recommendation behavior for the resource to be recommended;
determining a target recommendation index of the target object for the resource to be recommended according to the first recommendation index and the second recommendation index, and pushing the resource to be recommended to an object terminal of the target object according to the target recommendation index; the target recommendation index represents a target probability that a target object responds to a recommendation behavior of the resource to be recommended.
In a possible embodiment, the processor 1101, when being configured to determine the first recommendation index of the target object for the resource to be recommended according to the recommendation feature set, is specifically configured to:
performing feature fusion on the recommended feature set to obtain fusion attribute features;
determining a recommendation influence value according to the recommendation feature set; the recommendation influence value represents the influence degree of the recommendation information set on the target object response recommendation behavior;
and determining a first recommendation index according to the fusion attribute characteristics and the recommendation influence value.
In a possible implementation manner, the plurality of pieces of scene information include recommended scene information of the resource to be recommended, and the target scene information is recommended scene information;
when the processor 1101 is configured to determine, according to the object attribute feature and the resource attribute feature, a second recommendation index of the target object for the resource to be recommended under the target scene information, specifically:
acquiring a plurality of scene prediction networks; a scene prediction network corresponds to scene information;
and predicting the object attribute characteristics and the resource attribute characteristics according to the scene prediction network corresponding to the recommended scene information in the plurality of scene prediction networks to obtain a second recommendation index.
In a possible implementation manner, the plurality of pieces of scene information include recommended scene information of the resource to be recommended, and the target scene information includes the recommended scene information and associated scene information of the recommended scene information in the plurality of pieces of scene information;
when the processor 1101 is configured to determine, according to the object attribute feature and the resource attribute feature, a second recommendation index of the target object for the resource to be recommended under the target scene information, specifically:
acquiring a plurality of scene prediction networks; a scene prediction network corresponds to scene information;
predicting object attribute characteristics and resource attribute characteristics according to a scene prediction network corresponding to recommended scene information in a plurality of scene prediction networks and a scene prediction network corresponding to associated scene information respectively to obtain a first initial recommendation index of a target object for a resource to be recommended under the recommended scene information and a second initial recommendation index of the target object for the resource to be recommended under the associated scene information;
and carrying out weighted summation on the first initial recommendation index and the second initial recommendation index to obtain a second recommendation index.
In one possible embodiment, the target recommendation index is generated by a target prediction model; the processor 1101 is further configured to:
acquiring a plurality of sample information sets; a sample information set comprises sample object attribute information of a sample object, sample resource attribute information of a sample recommended resource and sample scene information of the sample recommended resource;
inputting each sample information set into an initial prediction model, and generating a sample characteristic set corresponding to each sample information set based on the initial prediction model; a sample feature set comprises a sample object attribute feature of sample object attribute information, a sample resource attribute feature of sample resource attribute information, and a sample recommendation scene feature of sample scene information of the sample resource attribute information;
respectively determining a target sample recommendation index of a sample object corresponding to each sample feature set aiming at the sample recommendation resource corresponding to the sample object according to each sample feature set;
and correcting the model parameters of the initial prediction model according to the recommended index of each target sample to obtain a target prediction model.
In a possible embodiment, the initial prediction model includes an initial common prediction network and a plurality of initial scene prediction networks, and one initial scene prediction network corresponds to one sample scene information;
when the processor 1101 is configured to determine, according to each sample feature set, a target sample recommendation index of the sample object corresponding to each sample feature set for the sample recommendation resource corresponding to the sample object, specifically:
in an initial public prediction network, generating a first sample recommendation index of a target sample object for recommending resources to a target sample according to a target feature set; the target feature set is any one of the sample feature sets in each sample feature set, the target sample object is a sample object corresponding to the target feature set, and the target sample recommendation resource is a sample recommendation resource corresponding to the target sample object;
in a target scene prediction network, determining a second sample recommendation index of a target sample object for a target sample recommendation resource according to the sample object attribute features in the target feature set and the sample resource attribute features in the target feature set; the target scene prediction network is an initial scene prediction network corresponding to target sample scene information in a plurality of initial scene prediction networks, and the target sample scene information is sample scene information corresponding to target sample recommended resources;
and determining a target sample recommendation index of the target sample object for the target sample recommended resource according to the first sample recommendation index and the second sample recommendation index.
In one possible embodiment, the initial prediction model includes an initial common prediction network and a plurality of initial scene prediction networks; when the processor 1101 is configured to modify the model parameters of the initial prediction model according to each target sample recommendation index to obtain a target prediction model, the processor 1101 is specifically configured to:
correcting network parameters of the initial public prediction network according to the recommended index of each target sample to obtain a public prediction network;
correcting network parameters of a plurality of initial scene prediction networks according to each target sample recommendation index and the sample scene information of each sample recommendation resource to obtain a plurality of scene prediction networks;
and determining a target prediction model according to the public prediction network and the plurality of scene prediction networks.
In one possible implementation, one initial scene prediction network corresponds to one sample scene information; when the processor 1101 is configured to modify network parameters of a plurality of initial scene prediction networks according to each target sample recommendation index and the sample scene information of each sample recommendation resource to obtain a plurality of scene prediction networks, the processor is specifically configured to:
according to the sample scene information of each sample recommended resource, dividing each target sample recommended index into a plurality of initial scene prediction networks;
and correcting the network parameters of the divided initial scene prediction networks according to the recommended index of each target sample to obtain a plurality of scene prediction networks.
In the embodiment of the application, the processor can obtain a recommendation information set, generate a recommendation feature set according to the recommendation information set, and determine a first recommendation index of a target object for a resource to be recommended according to the recommendation feature set; the first recommendation index represents the probability of the target object responding to the recommendation behavior of the resource to be recommended, a second recommendation index of the target object for the resource to be recommended under the target scene information is determined according to the object attribute characteristics and the resource attribute characteristics, a target recommendation index of the target object for the resource to be recommended is determined according to the first recommendation index and the second recommendation index, and the resource to be recommended is pushed to an object terminal of the target object according to the target recommendation index. According to the scheme, the interaction of various recommendation characteristics in the recommendation characteristic set is reflected by the obtained first recommendation index, the obtained second recommendation index is obtained by utilizing the learning characteristics under the target scene information, the learning characteristics under the target scene information can be obtained by comprehensively combining the learning characteristics of all scene information, the obtained target recommendation index can be deeply combined with various recommendation characteristics of the target recommendation index, the learning characteristics under various recommendation scenes can be combined, the recommendation prediction efficiency and accuracy of resources to be predicted can be improved, and the subsequent recommendation effect based on the target recommendation index can be improved.
Embodiments of the present application also provide a computer program product, which includes program instructions, and when the program instructions are executed by a processor, part or all of the steps of the above method can be implemented.
Those of ordinary skill in the art would appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. The procedures or functions described in accordance with the embodiments of the invention are all or partially effected when the computer program instructions are loaded and executed on a computer. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored on or transmitted over a computer-readable storage medium. The computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optics, digital object line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that incorporates one or more of the available media. The available media may be magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., DVDs), or semiconductor media (e.g., Solid State Disks (SSDs)), among others.
The above description is only for the specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present disclosure, and all the changes or substitutions should be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims (10)

1. A method for recommending data, the method comprising:
acquiring a recommendation information set; the recommendation information set comprises object attribute information of a target object, resource attribute information of resources to be recommended and recommendation scene information of the resources to be recommended;
generating a recommendation feature set according to the recommendation information set; the recommendation feature set comprises object attribute features of the object attribute information, resource attribute features of the resource attribute information and recommendation scene features of the recommendation scene information;
determining a first recommendation index of the target object for the resource to be recommended according to the recommendation feature set; the first recommendation index represents the probability of the target object responding to the recommendation behavior of the resource to be recommended;
determining a second recommendation index of the target object for the resource to be recommended under the target scene information according to the object attribute characteristics and the resource attribute characteristics; the target scene information is at least one scene information in a plurality of scene information associated with the resource to be recommended, and the second recommendation index represents the probability of the target object responding to the recommendation behavior of the resource to be recommended;
determining a target recommendation index of the target object for the resource to be recommended according to the first recommendation index and the second recommendation index, and pushing the resource to be recommended to an object terminal of the target object according to the target recommendation index; the target recommendation index represents a target probability of the target object responding to the recommendation behavior of the resource to be recommended.
2. The method according to claim 1, wherein the determining a first recommendation index of the target object for the resource to be recommended according to the recommendation feature set comprises:
performing feature fusion on the recommended feature set to obtain fusion attribute features;
determining a recommendation influence value according to the recommendation feature set; the recommendation influence value represents the influence degree of the recommendation information set on the target object response recommendation behavior;
and determining the first recommendation index according to the fusion attribute characteristics and the recommendation influence value.
3. The method according to claim 1, wherein the plurality of scenario information includes recommended scenario information of a resource to be recommended, and the target scenario information is the recommended scenario information;
determining a second recommendation index of the target object for the resource to be recommended under the target scene information according to the object attribute feature and the resource attribute feature, wherein the second recommendation index comprises:
acquiring a plurality of scene prediction networks; a scene prediction network corresponds to scene information;
and predicting the object attribute characteristics and the resource attribute characteristics according to a scene prediction network corresponding to the recommended scene information in the plurality of scene prediction networks to obtain the second recommendation index.
4. The method according to claim 1, wherein the plurality of scenario information includes recommended scenario information of a resource to be recommended, and the target scenario information includes the recommended scenario information and associated scenario information of the recommended scenario information in the plurality of scenario information;
determining a second recommendation index of the target object for the resource to be recommended under the target scene information according to the object attribute feature and the resource attribute feature, wherein the second recommendation index comprises:
acquiring a plurality of scene prediction networks; a scene prediction network corresponds to scene information;
predicting the object attribute characteristics and the resource attribute characteristics according to a scene prediction network corresponding to the recommended scene information and a scene prediction network corresponding to the associated scene information in the plurality of scene prediction networks respectively to obtain a first initial recommendation index of the target object for the resource to be recommended under the recommended scene information and a second initial recommendation index of the target object for the resource to be recommended under the associated scene information;
and carrying out weighted summation on the first initial recommendation index and the second initial recommendation index to obtain the second recommendation index.
5. The method of claim 1, wherein the target recommendation index is generated by a target prediction model; the method comprises the following steps:
acquiring a plurality of sample information sets; a sample information set comprises sample object attribute information of a sample object, sample resource attribute information of a sample recommended resource and sample scene information of the sample recommended resource;
inputting each sample information set into an initial prediction model, and generating a sample feature set corresponding to each sample information set based on the initial prediction model; a sample feature set comprises a sample object attribute feature of sample object attribute information, a sample resource attribute feature of sample resource attribute information, and a sample recommendation scene feature of sample scene information of the sample resource attribute information;
respectively determining a target sample recommendation index of the sample object corresponding to each sample feature set for the sample recommendation resource corresponding to the sample object according to each sample feature set;
and correcting the model parameters of the initial prediction model according to the recommended index of each target sample to obtain the target prediction model.
6. The method of claim 5, wherein the initial prediction model comprises an initial common prediction network and a plurality of initial scene prediction networks, one initial scene prediction network corresponding to one sample scene information;
the determining, according to each sample feature set, a target sample recommendation index of the sample object corresponding to each sample feature set for the sample recommendation resource corresponding to the sample object includes:
generating a first sample recommendation index of a target sample object for recommending resources to a target sample according to the target feature set in the initial public prediction network; the target feature set is any one of the sample feature sets, the target sample object is a sample object corresponding to the target feature set, and the target sample recommendation resource is a sample recommendation resource corresponding to the target sample object;
in a target scene prediction network, determining a second sample recommendation index of the target sample object for the target sample recommendation resource according to the sample object attribute features in the target feature set and the sample resource attribute features in the target feature set; the target scene prediction network is an initial scene prediction network corresponding to target sample scene information in the plurality of initial scene prediction networks, and the target sample scene information is sample scene information corresponding to the target sample recommended resources;
and determining a target sample recommendation index of the target sample object for the target sample recommended resource according to the first sample recommendation index and the second sample recommendation index.
7. The method of claim 5, wherein the initial prediction model comprises an initial public prediction network and a plurality of initial scene prediction networks; the step of correcting the model parameters of the initial prediction model according to the recommended indexes of each target sample to obtain the target prediction model comprises the following steps:
correcting the network parameters of the initial public prediction network according to the recommended index of each target sample to obtain a public prediction network;
correcting network parameters of the plurality of initial scene prediction networks according to the recommendation index of each target sample and the sample scene information of each sample recommendation resource to obtain a plurality of scene prediction networks;
determining the target prediction model from the common prediction network and the plurality of scene prediction networks.
8. The method of claim 7, wherein an initial scene prediction network corresponds to a sample scene information; correcting the network parameters of the multiple initial scene prediction networks according to the recommendation index of each target sample and the sample scene information of each sample recommendation resource to obtain multiple scene prediction networks, wherein the method comprises the following steps:
according to the sample scene information of each sample recommended resource, dividing each target sample recommended index into a plurality of initial scene prediction networks;
and correcting the network parameters of the divided initial scene prediction networks according to the recommended index of each target sample to obtain the plurality of scene prediction networks.
9. An electronic device comprising a processor and a memory, wherein the memory is configured to store a computer program comprising program instructions, and wherein the processor is configured to invoke the program instructions to perform the method of any of claims 1-8.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program comprising program instructions that, when executed by a processor, cause the processor to carry out the method according to any one of claims 1-8.
CN202210436960.7A 2022-04-24 2022-04-24 Data recommendation method, electronic device and storage medium Pending CN115033777A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210436960.7A CN115033777A (en) 2022-04-24 2022-04-24 Data recommendation method, electronic device and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210436960.7A CN115033777A (en) 2022-04-24 2022-04-24 Data recommendation method, electronic device and storage medium

Publications (1)

Publication Number Publication Date
CN115033777A true CN115033777A (en) 2022-09-09

Family

ID=83118692

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210436960.7A Pending CN115033777A (en) 2022-04-24 2022-04-24 Data recommendation method, electronic device and storage medium

Country Status (1)

Country Link
CN (1) CN115033777A (en)

Similar Documents

Publication Publication Date Title
US10958748B2 (en) Resource push method and apparatus
US20200242450A1 (en) User behavior prediction method and apparatus, and behavior prediction model training method and apparatus
WO2022022152A1 (en) Video clip positioning method and apparatus, and computer device and storage medium
CN112883257B (en) Behavior sequence data processing method and device, electronic equipment and storage medium
CN113873330B (en) Video recommendation method and device, computer equipment and storage medium
CN111159563A (en) Method, device and equipment for determining user interest point information and storage medium
CN113515690A (en) Training method of content recall model, content recall method, device and equipment
CN105426392B (en) Collaborative filtering recommendation method and system
CN113918738B (en) Multimedia resource recommendation method and device, electronic equipment and storage medium
CN115455280A (en) Recommendation list determining method and server
CN113836390B (en) Resource recommendation method, device, computer equipment and storage medium
CN115017362A (en) Data processing method, electronic device and storage medium
CN115618121B (en) Personalized information recommendation method, device, equipment and storage medium
CN113836388A (en) Information recommendation method and device, server and storage medium
CN115033777A (en) Data recommendation method, electronic device and storage medium
CN115129945A (en) Graph structure contrast learning method, equipment and computer storage medium
KR101663359B1 (en) Method and apparatus for providing updated news contents
CN112651764A (en) Target user identification method, device, equipment and storage medium
CN111507141B (en) Picture identification method, service interface display method, system and equipment
CN116777529A (en) Object recommendation method, device, equipment, storage medium and program product
CN117056561A (en) Video recommendation method, device, equipment, storage medium and program product
CN116112710A (en) Information recommendation method, device and server
CN117493666A (en) Article cold start recommendation method and article cold start recommendation model training method
CN117112880A (en) Information recommendation and multi-target recommendation model training method and device and computer equipment
CN114299571A (en) Model training method and device based on balance sample, electronic equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination