WO2023115974A1

WO2023115974A1 - Multimedia resource recommendation method and apparatus and object representation network generation method and apparatus

Info

Publication number: WO2023115974A1
Application number: PCT/CN2022/111130
Authority: WO
Inventors: 文剑烽; 穆冠宇
Original assignee: 北京达佳互联信息技术有限公司
Priority date: 2021-12-22
Filing date: 2022-08-09
Publication date: 2023-06-29
Also published as: CN114491093A; CN114491093B

Abstract

A multimedia resource recommendation method and apparatus and an object representation network generation method and apparatus. The multimedia resource recommendation method comprises: in response to a multimedia resource obtaining request of a target object, obtaining a target object attribute, a target number of pieces of historical operation sequence information, a target number of pieces of target resource category information, and resource feature information of multimedia resources to be recommended (S201); inputting the target object attribute, the target number of pieces of historical operation sequence information, and the target resource category information into an object representation network for object representation to obtain a target number of pieces of first object feature information corresponding to the target number of pieces of target resource category information (S203); determining, according to the target number of pieces of first object feature information and the resource feature information, a target multimedia resource from the multimedia resources to be recommended (S205); and recommending a resource to the target object on the basis of the target multimedia resource (S207). The method can improve recommendation accuracy and a recommendation effect, reduce system resource waste, and improve system performance.

Description

Method and device for multimedia resource recommendation and object representation network generation

Related Application Cross Reference

This application is based on a Chinese patent application with application number 202111579243.1 and a filing date of December 22, 2021, and claims the priority of this Chinese patent application. The entire content of this Chinese patent application is hereby incorporated by reference into this application.

technical field

The present disclosure relates to the technical field of artificial intelligence, and in particular to a method and device for recommending multimedia resources.

Background technique

With the development of Internet technology, a large number of network platforms are also constantly upgraded. In addition to publishing some graphic information, users can also share daily short videos and other multimedia resources at any time. How to accurately push these multimedia resources is a huge challenge. Challenges encountered in recommender systems.

In related technologies, in the process of multimedia resource recommendation, the object feature information of the recommended object and the multimedia resource to be recommended are often extracted in combination with the object tower corresponding to the recommended object and the resource tower corresponding to the multimedia resource to be recommended in the two-tower model. The resource characteristic information of the recommended multimedia resource, and then, calculate the dot product of the resource characteristic information of the multimedia resource to be recommended and the object characteristic information to obtain the estimated probability that the multimedia resource to be recommended is recommended to the recommended object, and combined with the estimated probability, the multimedia resource recommendation.

Contents of the invention

The disclosure provides a method and device for recommending multimedia resources, a method and device for generating an object representation network, electronic equipment, a computer-readable storage medium, and a computer program product. According to the first aspect of the embodiments of the present disclosure, a method for recommending multimedia resources is provided, including:

In response to the multimedia resource acquisition request of the target object, acquire the target object attributes of the target object, the target number of historical operation sequence information, the target number of target resource category information and the resource feature information of the multimedia resources to be recommended, the target The number of historical operation sequence information is the operation association information corresponding to the multimedia resources belonging to the target number of target resource category information among the historical multimedia resources recommended to the target object within the historical time period;

inputting the target object attributes, the target number of historical operation sequence information and the target number of target resource category information into an object characterization network for object characterization, and obtaining the target quantity corresponding to the target number of target resource category information A first object feature information;

Determine a target multimedia resource from the multimedia resources to be recommended according to the target number of first object feature information and the resource feature information;

Based on the target multimedia resource, resource recommendation is performed to the target object.

In some embodiments, the object representation network includes: object feature extraction network, feature intersection processing network, splicing network and feature fusion network;

Said inputting said target object attributes, said target number of historical operation sequence information and said target number of target resource category information into an object characterization network for object characterization, and obtaining said target number of target resource category information corresponding to The target number of first object feature information includes:

Based on the object feature extraction network, perform feature extraction processing on the target object attributes, the target number of historical operation sequence information, and the target number of target resource category information, and obtain the target number of sequence feature information and all The target number of category object feature information, the target number of category object feature information is the feature information corresponding to the target object attributes and the target number of target resource category information;

Based on the feature intersection processing network, perform feature intersection processing on the target number of sequence feature information and the target number of category object feature information to obtain the target number of cross feature information;

Perform splicing processing on the target number of cross feature information and the target number of category object feature information based on the splicing network to obtain the target number of spliced feature information;

The target number of spliced feature information is respectively fused based on the feature fusion network to obtain the target number of first object feature information.

In some embodiments, the object feature extraction network includes a first feature extraction network and a second feature extraction network;

Based on the object feature extraction network, perform feature extraction processing on the target object attributes, the target number of historical operation sequence information and the target number of target resource category information to obtain the target number of sequence feature information And the target number category object feature information includes:

Inputting the target object attributes and the target number of target resource category information into the first feature extraction network for feature extraction processing to obtain the target number of category object feature information;

Inputting the target number of historical operation sequence information into the second feature extraction network for feature extraction processing to obtain the target number of sequence feature information.

In some embodiments, the determining the target multimedia resource from the multimedia resources to be recommended according to the target number of first object feature information and the resource feature information includes:

determining resource category information of the multimedia resource to be recommended;

Using resource category information included in the target number of target resource category information to be recommended multimedia resources as primary multimedia resources;

Determine the resource recommendation index of the primary multimedia resource according to the resource feature information of the primary multimedia resource and the first object feature information corresponding to the resource category information of the primary multimedia resource;

Based on the resource recommendation index, the target multimedia resource is determined from the primary multimedia resources.

In some embodiments, the object representation network comprises a base object representation network; the method further comprising:

inputting the attributes of the target object into the basic object characterization network for object characterization to obtain second object feature information;

The determining the target multimedia resource from the multimedia resources to be recommended according to the target number of first object feature information and the resource feature information includes:

The target multimedia resource is determined from the multimedia resources to be recommended according to the target number of first object feature information, the second object feature information, and the resource feature information.

In some embodiments, the multimedia resource to be recommended includes a plurality of multimedia resources; the target number of first object feature information, the second object feature information, and the resource feature information is obtained from the to-be-recommended In recommending multimedia resources, determining the target multimedia resources includes:

determining resource category information of the plurality of multimedia resources;

In response to the resource type information of any multimedia resource being included in the target number of target resource type information, according to the resource feature information of the first multimedia resource and the resource type information of the first multimedia resource corresponding to the first An object feature information to determine the resource recommendation index of the first multimedia resource; the first multimedia resource is the resource category information of the multimedia resource to be recommended included in the target number of target resource category information multimedia resources;

In response to resource type information of any multimedia resource not included in the target number of target resource type information, determine the second multimedia resource according to the resource feature information of the second multimedia resource and the second object feature information The resource recommendation indicator of the body resource; the second multimedia resource is a multimedia resource whose resource category information in the multimedia resource to be recommended is not included in the target number of target resource category information;

Based on the resource recommendation index of the first multimedia resource and the resource recommendation index of the second multimedia resource, the target multimedia resource is determined from the multimedia resources to be recommended.

In some embodiments, obtaining the target number of historical operation sequence information includes:

Determine the preset number of resource category information to which the historical multimedia resource belongs and the resource quantity corresponding to the preset number of resource category information;

Determining the target number of target resource category information from a preset number of resource category information based on the resource quantity;

The target number of historical operation sequence information is generated based on the operation association information of the multimedia resources corresponding to the target number of target resource category information among the historical multimedia resources.

In some embodiments, obtaining resource characteristic information of the multimedia resource to be recommended includes:

Obtain the target resource attribute of the multimedia resource to be recommended;

Inputting the attribute of the target resource into a resource characterization network for resource characterization to obtain resource feature information of the multimedia resource to be recommended.

In some embodiments, the response to the multimedia resource acquisition request of the target object includes:

In response to receiving the multimedia resource acquisition request, triggering in parallel a recall instruction corresponding to the multimedia resource to be recommended and an object characterization instruction corresponding to the target object;

Wherein, the recall instruction is used to instruct to recall the multimedia resource to be recommended, and the object characterization instruction is used to instruct to execute object characterization processing corresponding to the target object.

According to a second aspect of an embodiment of the present disclosure, a method for generating an object representation network is provided, including:

Obtain the sample object attributes of the sample object, the target number of sample operation sequence information, the target number of sample resource category information, and the sample resource feature information of the sample multimedia resource, and the target number of sample operation sequence information is recommended within the sample time period Among the multimedia resources of the sample object, the operation association information corresponding to the multimedia resources belonging to the category information of the target number of sample resources;

Input the sample object attributes, the target number of sample operation sequence information and the target number of sample resource category information into the object representation network to be trained for object representation, and obtain the target number of sample resource category information corresponding to Target number of sample object feature information;

Determine a sample resource recommendation index according to the target number of sample object feature information and the sample resource feature information;

Based on the resource recommendation index, train the object representation network to be trained to obtain an object representation network.

In some embodiments, obtaining the sample resource characteristic information of the sample multimedia resource includes:

Obtain the sample resource attribute of the sample multimedia resource;

inputting the attributes of the sample resources into the resource characterization network to be trained for resource characterization, and obtaining feature information of the sample resources;

The step of training the object representation network to be trained based on the resource recommendation index, and obtaining the object representation network includes:

Based on the resource recommendation index, train the object representation network to be trained and the resource representation network to be trained to obtain the object representation network and resource representation network.

According to a third aspect of the embodiments of the present disclosure, a device for recommending multimedia resources is provided, including:

The data acquisition module is configured to, in response to the multimedia resource acquisition request of the target object, acquire the target object attributes of the target object, the target number of historical operation sequence information, the target number of target resource category information and the multimedia resources to be recommended Resource feature information, the target number of historical operation sequence information is the operation association information corresponding to the multimedia resources belonging to the target number of target resource category information among the historical multimedia resources recommended to the target object within the historical time period;

The first object characterization module is configured to input the target object attributes, the target number of historical operation sequence information and the target number of target resource category information into the object characterization network for object characterization, and obtain the target number of targets The target number of first object feature information corresponding to the resource category information;

The target multimedia resource determination module is configured to determine a target multimedia resource from the multimedia resources to be recommended according to the target number of first object feature information and the resource feature information;

The resource recommendation module is configured to recommend resources to the target object based on the target multimedia resources.

The first object characterization module includes:

The feature extraction processing unit is configured to perform feature extraction processing on the target object attributes, the target number of historical operation sequence information and the target number of target resource category information based on the object feature extraction network, to obtain the A target number of sequence feature information and the target number of category object feature information, where the target number of category object feature information is feature information corresponding to the target object attributes and the target number of target resource category information;

The feature intersection processing unit is configured to perform feature intersection processing on the target number of sequence feature information and the target number of category object feature information based on the feature intersection processing network to obtain the target number of intersection feature information;

The splicing processing unit is configured to splice the target number of cross feature information and the target number of category object feature information based on the splicing network to obtain the target number of spliced feature information;

The fusion processing unit is configured to respectively perform fusion processing on the target number of spliced feature information based on the feature fusion network to obtain the target number of first object feature information.

The feature extraction processing unit includes:

The first feature extraction processing subunit is configured to input the target object attributes and the target number of target resource category information into the first feature extraction network to perform feature extraction processing, and obtain the target number of category object features information;

The second feature extraction processing subunit is configured to input the target amount of historical operation sequence information into the second feature extraction network for feature extraction processing, and obtain the target amount of sequence feature information.

In some embodiments, the target multimedia resource determination module includes:

The first resource type information determining unit is configured to determine the resource type information of the multimedia resource to be recommended;

The primary multimedia resource determination unit is configured to include the resource category information in the target number of target resource category information to be recommended multimedia resources as the primary multimedia resource;

The first resource recommendation index determining unit is configured to determine the resource of the primary multimedia resource according to the resource feature information of the primary multimedia resource and the first object feature information corresponding to the resource category information of the primary multimedia resource. Recommended indicators;

The first target multimedia resource determining unit is configured to determine the target multimedia resource from the primary multimedia resources based on the resource recommendation index.

In some embodiments, the object representation network comprises a base object representation network; the apparatus further comprises:

The second object representation network is configured to input the attributes of the target object into the basic object representation network for object representation to obtain second object feature information;

The target multimedia resource determining module is further configured to: determine the Target multimedia resource.

In some embodiments, the multimedia resources to be recommended include multiple multimedia resources; the target multimedia resource determination module includes:

a second resource type information determining unit configured to determine resource type information of the plurality of multimedia resources;

The second resource recommendation index determination unit is configured to respond to resource category information of any multimedia resource being included in the target number of target resource category information, according to the resource feature information of the first multimedia resource and the first The first object feature information corresponding to the resource category information of the multimedia resource determines the resource recommendation indicator of the first multimedia resource; the first multimedia resource is the resource category information of the multimedia resource to be recommended. Multimedia resources with target resource category information in said target number;

The third resource recommendation index determining unit is configured to respond to resource category information of any multimedia resource not included in the target number of target resource category information, according to the resource characteristic information of the second multimedia resource and the first multimedia resource. Two object characteristic information, to determine the resource recommendation index of the second multimedia resource; the second multimedia resource is the resource category information of the multimedia resource to be recommended that is not included in the target number of target resource category information Multimedia resources in ;

The second target multimedia resource determining unit is configured to determine the multimedia resource to be recommended based on the resource recommendation index of the first multimedia resource and the resource recommendation index of the second multimedia resource. Target multimedia resource.

In some embodiments, the data acquisition module includes:

A resource data acquisition unit configured to determine a preset number of resource category information to which the historical multimedia resource belongs and a resource quantity corresponding to the preset number of resource category information;

The target resource category information determining unit is configured to determine the target number of target resource category information from a preset number of resource category information based on the resource quantity;

The historical operation sequence information generation unit is configured to generate the target amount of historical operation sequence information based on the operation association information of the multimedia resources corresponding to the target amount of target resource category information among the historical multimedia resources.

In some embodiments, the data acquisition module includes:

a target resource attribute acquisition unit configured to acquire target resource attributes of multimedia resources to be recommended;

The resource characterization unit is configured to input the attribute of the target resource into a resource characterization network for resource characterization, and obtain resource characteristic information of the multimedia resource to be recommended.

In some embodiments, the data acquisition module is further configured to: in response to receiving the multimedia resource acquisition request, trigger a recall instruction corresponding to the multimedia resource to be recommended and an object characterization instruction corresponding to the target object in parallel;

According to a fourth aspect of an embodiment of the present disclosure, an apparatus for generating an object representation network is provided, including:

The sample data acquisition module is configured to acquire the sample object attributes of the sample object, the target number of sample operation sequence information, the target number of sample resource category information and the sample resource characteristic information of the sample multimedia resource, and the target number of sample operations The sequence information is the operation related information corresponding to the multimedia resources that belong to the target number of sample resource category information among the multimedia resources recommended to the sample object within the sample time period;

The third object characterization module is configured to input the sample object attributes, the target number of sample operation sequence information and the target number of sample resource category information into the object characterization network to be trained for object characterization, and obtain the target number The target number of sample object feature information corresponding to the sample resource category information;

The sample resource recommendation index determination module is configured to determine the sample resource recommendation index according to the target number of sample object feature information and the sample resource feature information;

The network training module is configured to train the object representation network to be trained based on the resource recommendation index to obtain an object representation network.

In some embodiments, the sample data acquisition module includes:

a sample resource attribute acquiring unit configured to acquire a sample resource attribute of a sample multimedia resource;

The resource characterization unit is configured to input the sample resource attributes into the resource characterization network to be trained for resource characterization, and obtain the sample resource feature information;

The network training module is further configured to: train the object representation network to be trained and the resource representation network to be trained based on the resource recommendation index to obtain the object representation network and resource representation network.

According to a fifth aspect of the embodiments of the present disclosure, there is provided an electronic device, including: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement The method as described in any one of the first aspect or the second aspect above.

According to the sixth aspect of the embodiments of the present disclosure, there is provided a computer-readable storage medium. When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device can execute the first method of the embodiments of the present disclosure. Aspect, or the method described in any one of the second aspect.

According to a seventh aspect of the embodiments of the present disclosure, there is provided a computer program product containing instructions, which, when run on a computer, causes the computer to execute the first aspect of the embodiments of the present disclosure, or any one of the second aspects method.

In the technical solution provided by the embodiments of the present disclosure, after receiving the multimedia resource acquisition request of the target object, combining the target object attributes of the target object and the historical multimedia resources recommended to the target object in the historical time period, which belong to the target number The historical operation sequence information corresponding to the multimedia resource of the target resource category information is used to represent the object, which can better improve the accuracy of the obtained object feature information on the target object's interests and preferences, and because the historical operation sequence information is combined with the number of targets The target resource category information of object preference is extracted from the historical operation association information of the object, which can reduce the granularity of object features to the resource category, and simplify the number of operations of object representation processing after introducing historical operation sequence information to the number of categories, which can effectively guarantee The processing efficiency of object characterization processing; and in combination with the resource feature information of the multimedia resources to be recommended and the target number of first object feature information that can represent the object's behavior preferences, determine the target multimedia resources for recommendation processing from the multimedia resources to be recommended , can better capture the object's interest preference, improve the recommendation accuracy and recommendation effect in the recommendation system, reduce the waste of system resources caused by invalid multimedia resource recommendation, and improve system performance.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure.

Description of drawings

The accompanying drawings here are incorporated into the specification and constitute a part of the specification, show embodiments consistent with the disclosure, and are used together with the description to explain the principle of the disclosure, and do not constitute an improper limitation of the disclosure.

Fig. 1 is a schematic diagram showing an application environment according to some embodiments;

Fig. 2 is a flowchart of a method for recommending multimedia resources according to some embodiments;

Fig. 3 is a flowchart of a method for recommending multimedia resources according to some embodiments;

Fig. 4 is a flowchart of a method for recommending multimedia resources according to some embodiments;

Fig. 5 is a flowchart of a method for recommending multimedia resources according to some embodiments;

Fig. 6 is a flowchart of a method for recommending multimedia resources according to some embodiments;

Fig. 7 is a schematic diagram of an object representation network and a resource representation network provided according to some embodiments;

Fig. 8 is a flowchart of a method for generating an object representation network according to some embodiments;

Fig. 9 is a schematic diagram of a recommendation system provided according to some embodiments;

Fig. 10 is a block diagram of an apparatus for recommending multimedia resources according to some embodiments;

Fig. 11 is a block diagram of an apparatus for generating an object representation network according to some embodiments;

Fig. 12 is a block diagram of an electronic device according to some embodiments;

Fig. 13 is a block diagram of another electronic device according to some embodiments.

Detailed ways

In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

It should be noted that the terms "first" and "second" in the specification and claims of the present disclosure and the above drawings are used to distinguish similar objects, but not necessarily used to describe a specific sequence or sequence. It is to be understood that the data so used are interchangeable under appropriate circumstances such that the embodiments of the disclosure described herein can be practiced in sequences other than those illustrated or described herein. The implementations described in the following exemplary examples do not represent all implementations consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with aspects of the present disclosure as recited in the appended claims.

It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this disclosure are authorized by the user. Or information and data fully authorized by the parties.

Please refer to FIG. 1 . FIG. 1 is a schematic diagram of an application environment according to some embodiments. As shown in FIG. 1 , the application environment may include a server 100 and a terminal 200 .

In some embodiments, the server 100 can be used to train object representation networks and resource representation networks. For example, the server 100 can be an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, and can also provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud Cloud servers for basic cloud computing services such as communications, middleware services, domain name services, security services, CDN (Content Delivery Network, content distribution network), and big data and artificial intelligence platforms.

In some embodiments, the terminal 200 may be used to provide a multimedia resource recommendation service for any user. For example, multimedia resource recommendation can be performed in combination with the object representation network and resource representation network trained by the server 100 . The terminal 200 may include, but not limited to, smart phones, desktop computers, tablet computers, notebook computers, smart speakers, digital assistants, augmented reality (augmented reality, AR)/virtual reality (virtual reality, VR) devices, smart wearable devices, etc. The electronic device may also be software running on the above-mentioned electronic device, such as an application program. In some embodiments, the operating system running on the electronic device may include but not limited to Android system, IOS system, linux, windows and so on.

In addition, it should be noted that what is shown in FIG. 1 is only an application environment provided by the present disclosure, and in actual application, other application environments may also be included, for example, more terminals may be included.

In the embodiment of the present disclosure, the server 100 and the terminal 200 may be connected directly or indirectly through wired or wireless communication, which is not limited in the present disclosure.

In some technical solutions, the object tower and the resource tower can only use the information on the object side and the multimedia resource side respectively, resulting in the relatively static information of the recommended object and the multimedia resource to be recommended being considered separately in the recommendation process, which cannot be captured well. The target interest preference of the recommendation system leads to poor recommendation accuracy and effect in the recommendation system, and invalid multimedia resource recommendation, which also causes problems such as waste of system resources and degradation of system performance in the recommendation system.

Therefore, the present disclosure provides a method and device for recommending multimedia resources, a method and device for generating an object representation network, electronic equipment, a computer-readable storage medium, and a computer program product, so as to at least solve problems that cannot be well captured in related technologies The target interest preference, the recommendation accuracy and effect in the recommendation system are poor, and the invalid multimedia resource recommendation also causes the system resource waste and system performance degradation of the recommendation system.

Fig. 2 is a flowchart of a method for recommending multimedia resources according to some embodiments. As shown in Fig. 2, the method for recommending multimedia resources is used in electronic devices such as terminals and servers, and includes the following steps S201, S203, S205 and S207 .

In step S201, in response to the multimedia resource acquisition request of the target object, acquire the target object attributes of the target object, the target number of historical operation sequence information, the target number of target resource category information and the resource feature information of the multimedia resources to be recommended.

In some embodiments, the above-mentioned target object may be a recommendation object of multimedia resources to be recommended; for example, the target object may be any user account in the recommendation system, and the target object attribute of the target object may be an attribute that can characterize the interest preference of the target object information. In some embodiments, the attributes of the target object may include but not limited to user gender, age, education, region, and information about multimedia resources recommended to the target object, and contextual information when recommending multimedia resources to the target object (such as recommended time, Information that characterizes the interests and preferences of the target object, such as the network information used, the recommended background such as address location), and the feedback information of the target object on the recommended multimedia resources (such as playing time, whether to like, share, etc.).

In some embodiments, the above-mentioned target number of historical operation sequence information may be operation association information corresponding to multimedia resources belonging to the target number of target resource category information among the historical multimedia resources recommended to the target object within the historical time period. In some embodiments, the historical time period can be set in combination with the actual application, and the operation-related information can be the information in the process of recommending historical multimedia resources to the target object. For example, the operation-associated information may include the resource identifier of the multimedia resource, the identifier information of the publisher of the multimedia resource, the browsing duration, the corresponding display duration of the multimedia resource, resource category information of the multimedia resource, and the like.

In some embodiments, the resource category information may characterize the subject content of the multimedia resource. In practical applications, the recommendable multimedia resources in the recommender system may be clustered in advance, so as to determine various resource category information corresponding to the recommendable multimedia resources in the recommender system. The classification of multimedia resources can be configured in combination with actual applications. For example, resource category information of multimedia resources may include sports, gourmet, travel and so on. The target quantity can be set according to the actual application. In some embodiments, the target number of target resource category information may be category information of at least one multimedia resource preferred by the target object.

In some embodiments, the multimedia resource to be recommended may be a multimedia resource in the recommendation system. For example, multimedia resources may include static resources such as text and images, and may also include dynamic resources such as short videos. In some embodiments, the multimedia resource to be recommended may be a multimedia resource that can be recommended to the target object recalled from the recommendation system in combination with preset recall rules.

In some embodiments, as shown in FIG. 3 , the following steps S301, S303 and S305 are used to obtain the above target number of historical operation sequence information:

In step S301, determine the preset number of resource category information to which the historical multimedia resource belongs and the resource quantity corresponding to the preset number of resource category information;

In step S303, based on the number of resources, determine a target number of target resource category information from a preset number of resource category information;

In step S305, a target number of historical operation sequence information is generated based on the operation-related information of multimedia resources corresponding to the target number of target resource type information among the historical multimedia resources.

In some embodiments, the historical multimedia resources recommended to the target object in the historical time period may often include multiple multimedia resources. The resource category information to which each multimedia resource belongs can be determined. The preset number may be the number of resource categories of the multimedia resources in the historical multimedia resources.

In some embodiments, group statistics can be performed according to the resource category information to which each multimedia resource belongs, and the previous target resource category information with the largest number of resources is selected, and the target resource category information of the target quantity is selected to correspond to the operation of the multimedia resource The associated information is used as the above-mentioned target number of historical operation sequence information. Any target resource type information corresponds to the operation association information of the multimedia resource, and historical operation sequence information corresponding to the target resource type information can be generated.

In the above-mentioned embodiment, combined with the quantity statistics of various historical multimedia resources recommended to the target object within the historical time period, the category information of the target number of multimedia resources preferred by the target object can be screened out, and then the target number of target objects preferred by the target object can be selected. The resource category information corresponds to the operation association information of the multimedia resources to generate the historical operation sequence information of the target object, which can effectively improve the accuracy of the filtered historical operation sequence information on the target object's interest preference, and then improve the subsequent recommendation effect.

In some embodiments, the resource characteristic information of the multimedia resource to be recommended is acquired in the following ways:

The target resource attributes are input into the resource characterization network for resource characterization, and the resource feature information of the multimedia resources to be recommended is obtained.

In some embodiments, the attribute of the target resource may be information used to describe the multimedia resource. Taking the multimedia resource as video as an example, the resource feature information may include publisher information, resource identifier, release date, and video frame image of the multimedia resource to be recommended. , audio information, playing duration, title information, etc. can describe the information of the multimedia resource to be recommended. Correspondingly, the resource characteristic information may be a characteristic representation of the attribute of the target resource.

In some embodiments, the resource characterization network may be a pre-trained machine learning network capable of characterizing resource features. The specific structure of the resource representation network can be configured in combination with practical applications.

In the above embodiments, resource characterization is performed in combination with the resource characterization network, which can better extract feature information of multimedia resources to be recommended, improve the accuracy of representation of multimedia resources to be recommended, and further improve the accuracy and effect of recommendation in the recommendation system.

In step S203, input the target object attributes, the target number of historical operation sequence information and the target number of target resource category information into the object representation network for object representation, and obtain the target number of first object features corresponding to the target number of target resource category information information;

In some embodiments, the object representation network may be a pre-trained machine learning network capable of characterizing object features. The specific structure of the object representation network can be configured in combination with practical applications.

In some embodiments, the above-mentioned object representation network may include: an object feature extraction network, a feature cross processing network, a splicing network, and a feature fusion network; as shown in FIG. The target number of target resource category information is input into the object characterization network for object characterization, and the target number of target resource category information corresponding to the target number of first object characteristic information may include the following steps S401, S403, S405 and S407:

In step S401, based on the object feature extraction network, perform feature extraction processing on the target object attributes, the target number of historical operation sequence information and the target number of target resource category information, and obtain the target number of sequence feature information and the target number of category object features information;

In some embodiments, the target number of category object feature information may be feature information of target object attributes corresponding to target number of target resource category information.

In some embodiments, the above-mentioned object feature extraction network may include a first feature extraction network and a second feature extraction network; correspondingly, the above-mentioned object-based feature extraction network is used for target object attributes, target quantity historical operation sequence information and target quantity The target resource category information is subjected to feature extraction processing, and the target quantity sequence feature information and the target quantity category object feature information can include:

input the target object attributes and the target number of target resource category information into the first feature extraction network for feature extraction processing, and obtain the target number of category object feature information;

The target number of historical operation sequence information is input into the second feature extraction network for feature extraction processing, and the target number of sequence feature information is obtained.

In some embodiments, the specific network structures of the first feature extraction network and the second feature extraction network can be set according to actual application requirements.

In some embodiments, each historical operation sequence information can be sequentially input into the second feature extraction network for feature extraction processing to obtain the sequence feature information corresponding to the historical operation sequence information; at the same time, the target resource corresponding to the historical operation sequence information The category information and target object attributes are input into the first feature extraction network for feature extraction processing to obtain category object feature information.

In the above embodiment, combined with the first feature extraction network and the second feature extraction network, respectively extracting category object feature information and sequence feature information, different feature extraction networks can be used to extract features of different information in a more targeted manner, thereby improving The accuracy and effectiveness of the extracted category object feature information and sequence feature information.

In step S403, based on the feature intersection processing network, perform feature intersection processing on the target number of sequence feature information and the target number of category object feature information to obtain the target number of cross feature information.

In some embodiments, the feature intersection processing network can be used to combine the feature information of the category object to represent the sequence feature information of the target object into a high-dimensional dense feature information.

In some embodiments, the specific network structure of the feature intersection processing network can be set in combination with practical applications, such as multi-head attention network, deep interest network, long-term and short-term interest network, and gated recurrent network.

In some embodiments, each sequence feature information and the category object feature information corresponding to the sequence feature information may be input into the feature intersection processing network for feature intersection processing to obtain the intersection feature information corresponding to the sequence feature information.

In step S405, the target number of intersection feature information and the target number of category object feature information are spliced based on the splicing network to obtain the target number of spliced feature information.

In some embodiments, each intersection feature information and corresponding category object feature information may be input into the splicing network for splicing processing to obtain corresponding splicing feature information.

In some embodiments, the specific network structure of the mosaic network can be set in combination with practical applications.

In step S407, the target number of spliced feature information is respectively fused based on the feature fusion network to obtain the target number of first object feature information.

In some embodiments, each concatenated feature information may be input into a feature fusion network for fusion processing to obtain corresponding first object feature information.

In some embodiments, the specific network structure of the feature fusion network can be set in combination with practical applications.

In the above embodiment, after combining the object feature extraction network to extract the sequence feature information and category object feature information, combining the feature cross processing network to represent the sequence feature information of the target object into a high-dimensional dense feature information, which can greatly improve the accuracy of the target object. The representation accuracy of the operation sequence can better capture the object's interest preference, improve the recommendation accuracy and recommendation effect in the recommendation system, reduce the waste of system resources caused by invalid multimedia resource recommendation, and improve system performance.

In step S205, according to the target number of first object feature information and resource feature information, from the multimedia resources to be recommended, determine the target multimedia resource;

In some embodiments, as shown in FIG. 5, determining the target multimedia resource from among the multimedia resources to be recommended according to the target number of first object feature information and resource feature information may include the following steps S501, S503, S505, and S507:

In step S501, determine the resource category information of the multimedia resource to be recommended;

In step S503, multimedia resources to be recommended whose resource category information is included in the target number of target resource category information are used as primary multimedia resources;

In step S505, according to the resource feature information of the primary multimedia resource and the first object feature information corresponding to the resource category information of the primary multimedia resource, determine the resource recommendation index of the primary multimedia resource;

In step S507, the target multimedia resource is determined from the primary multimedia resources based on the resource recommendation index.

In some embodiments, the multimedia resource to be recommended may include a plurality of multimedia resources. Correspondingly, the resource category information may be combined with the resource category information of the multimedia resource to be recommended, and the resource category information belongs to the target number of multimedia resources with target resource category information preferred by the target object. , as the primary multimedia resource, and filter out non-object interest-preferred multimedia resources. Next, the resource recommendation index of the primary multimedia resource is determined in combination with the resource feature information of the primary multimedia resource and the first object feature information corresponding to the resource category information of the primary multimedia resource. The resource recommendation index of any multimedia resource can represent the probability that the multimedia resource is recommended to the target object. The higher the probability, the higher the preference degree of the target object for multimedia resources can be represented.

In some embodiments, the resource feature information and the corresponding first object feature information may be subjected to dot product processing to obtain the corresponding resource recommendation index.

In some embodiments, the primary multimedia resource may include a plurality of multimedia resources to be recommended; based on the above-mentioned resource recommendation index, determining the target multimedia resource from the primary multimedia resource may include combining the resource recommendation index with the primary multimedia resource. The plurality of multimedia resources are sorted in descending order, and the first preset number of multimedia resources before sorting are selected as target multimedia resources. The first preset number can be set according to actual application, for example, the first preset number can be greater than or equal to one.

In some embodiments, based on the resource recommendation index, determining the target multimedia resource from the primary multimedia resources may include: taking a multimedia resource whose resource recommendation index is greater than or equal to a first preset index threshold among the primary multimedia resources as the target multimedia resource . The first preset index threshold can be set in combination with actual applications.

In the above embodiment, combined with the resource category information of the multimedia resources to be recommended, the multimedia resources corresponding to the target resource category information that is not preferred by the object are filtered out, and the resource feature information of the filtered primary multimedia resources is combined with the primary multimedia resource. The first object feature information corresponding to the resource category information is used to generate a resource recommendation index, which can more accurately reflect the preferences of the target object for multimedia resources, and then accurately screen out the target multimedia resources that the target object prefers based on the resource recommendation index. Improve the accuracy of capturing object interests and preferences.

In some embodiments, the above-mentioned object representation network includes a basic object representation network; correspondingly, the above-mentioned method may further include:

Inputting the attributes of the target object into the basic object representation network for object representation to obtain second object feature information;

In some embodiments, the base object representation network may include a first feature extraction network, a stitching network, and a feature fusion network. Correspondingly, the above-mentioned inputting the attributes of the target object into the basic object characterization network for object characterization and obtaining the second object feature information may include: inputting the target object attributes into the first feature extraction network for feature extraction processing to obtain initial object feature information; The feature information is input into the splicing network for splicing processing to obtain the object splicing feature information (that is, the feature information corresponding to different attributes in the target object attribute is spliced); then, the object splicing feature information is input into the feature fusion network for fusion processing to obtain the second object characteristic information.

Correspondingly, determining the target multimedia resource from the multimedia resources to be recommended according to the target number of first object feature information and resource feature information may include:

The target multimedia resource is determined from the multimedia resources to be recommended according to the target number of first object feature information, second object feature information, and resource feature information.

In the above-mentioned embodiment, the characterization of the target object is carried out in combination with the attributes of the target object. On the basis of the first object feature information that can characterize the object's behavior preferences, the second object feature information that characterizes the object's personalized long-tail interest can be introduced, and then It can more comprehensively and accurately characterize the interest preferences of objects, improve the recommendation accuracy and recommendation effect in the recommendation system, reduce the waste of system resources caused by invalid multimedia resource recommendations, and improve system performance.

In some embodiments, the multimedia resources to be recommended include a plurality of multimedia resources; correspondingly, as shown in FIG. Among the multimedia resources, determining the target multimedia resource may include the following steps S601, S603, S605 and S607:

In step S601, determine resource category information of a plurality of multimedia resources;

In step S603, in response to the resource type information of any multimedia resource being included in the target number of target resource type information, according to the resource characteristic information of the first multimedia resource and the resource type information of the first multimedia resource corresponding The first object characteristic information determines the resource recommendation index of the first multimedia resource.

In some embodiments, the above-mentioned first multimedia resource may be a multimedia resource whose resource category information contains more than a target number of target resource category information among the multimedia resources to be recommended; the resource feature information and the corresponding first object feature information may be combined Dot product processing to obtain the corresponding resource recommendation indicators.

In step S605, in response to the resource type information of any multimedia resource not included in the target number of target resource type information, determine the second multimedia resource according to the resource feature information and the second object feature information of the second multimedia resource. The resource recommendation index of the entity resource.

In some embodiments, the above-mentioned second multimedia resource may be a multimedia resource whose resource type information is not included in the target number of target resource type information among the multimedia resources to be recommended. The resource feature information and the second object feature information may be subjected to dot product processing to obtain the resource recommendation index of the second multimedia resource.

In step S607, based on the resource recommendation index of the first multimedia resource and the resource recommendation index of the second multimedia resource, the target multimedia resource is determined from the multimedia resources to be recommended.

In some embodiments, both the first multimedia resource and the second multimedia resource may include at least one multimedia resource. Based on the resource recommendation index of the first multimedia resource and the resource recommendation index of the second multimedia resource, determining the target multimedia resource from the multimedia resources to be recommended may include: combining the resource recommendation index with the first multimedia resource and The second multimedia resources are sorted in descending order, and a second preset number of multimedia resources before sorting are selected as target multimedia resources. The second preset number can be set in combination with actual applications, for example, the second preset number can be greater than or equal to one.

In some embodiments, based on the resource recommendation index of the first multimedia resource and the resource recommendation index of the second multimedia resource, determining the target multimedia resource from the multimedia resources to be recommended may include: Among the resource and the second multimedia resource, the multimedia resource whose resource recommendation index is greater than or equal to the second preset index threshold is used as the target multimedia resource. The second preset index threshold can be set in combination with actual applications.

In the above embodiment, in combination with the resource type information of the multimedia resources to be recommended, it is determined that the resource type information belongs to the first multimedia resource of the target resource type information preferred by the object, and the resource type information does not belong to the target resource type information of the object preference. The second multimedia resource; and for the first multimedia resource, combine the resource characteristic information of the first multimedia resource and the first object characteristic information corresponding to the resource category information of the first multimedia resource to generate a resource recommendation index; for the second multimedia resource, combined with the resource characteristic information of the second multimedia resource and the second object characteristic information, a resource recommendation index is generated, which can more comprehensively reflect the preferences of the target object for the multimedia resource, and then can be based on The resource recommendation index accurately screens out more comprehensive target multimedia resources that are more in line with the preferences of the target audience.

In some embodiments, as shown in FIG. 7 , FIG. 7 is a schematic diagram of an object representation network and a resource representation network provided according to some embodiments. In conjunction with Figure 7, the object representation network may include an object feature extraction network, a feature cross processing network, a splicing network, and a feature fusion network, wherein the object feature extraction network in the object representation network may include a first feature extraction network and a second feature extraction network . The resource representation network can include resource feature extraction network, splicing network and feature fusion network. A target number of historical operation sequence information, a target number of target resource category information, and target object attributes can be input into the object representation network for object representation processing to obtain the target number of first object feature information; and input the target object attributes into the object The characterization network performs object characterization processing to obtain the second object feature information; in addition, the target resource attributes of the multimedia resources to be recommended can be input into the resource characterization network for resource characterization to obtain resource feature information; then, based on the target number of first object feature information , the second object characteristic information and the resource characteristic information, and determine the resource recommendation index of the multimedia resource to be recommended.

In step S207, resource recommendation is performed to the target object based on the target multimedia resource.

In some embodiments, based on the target multimedia resource, recommending the resource to the target object may include recommending the target multimedia resource to the target object.

In some embodiments, the recommendation system can generally be divided into four stages: recall, pre-sorting (rough sorting), fine sorting, and rearrangement. The target multimedia resource can be used as the output of the pre-sorting stage; correspondingly, the target multimedia resource can be transmitted to the corresponding processing module of the refinement stage, so as to further screen the multimedia resource.

From the technical solutions provided by the above embodiments, it can be seen that in this disclosure, after receiving the multimedia resource acquisition request of the target object, among the historical multimedia resources recommended to the target object within the historical time period in combination with the target object attribute of the target object, the The historical operation sequence information corresponding to the multimedia resources of the target resource category information is used to represent the object, which can better improve the accuracy of the obtained object feature information on the target object's interest and preference, and because the historical operation sequence information is combined with the target The target resource category information preferred by a number of objects is extracted from the object historical operation association information, which can reduce the object feature granularity to the resource category, and simplify the number of operations of the object representation processing after introducing the historical operation sequence information to the number of categories, which can Effectively ensure the processing efficiency of object representation processing; and combine the resource feature information of the multimedia resources to be recommended and the target number of first object feature information that can represent the object's behavior preferences, and determine the target for recommendation processing from the multimedia resources to be recommended Multimedia resources can better capture the object's interest preferences, improve the recommendation accuracy and recommendation effect in the recommendation system, reduce the waste of system resources caused by invalid multimedia resource recommendations, and improve system performance.

Fig. 8 is a flowchart of a method for generating an object representation network according to some embodiments. As shown in Fig. 8, the object representation network is used in electronic devices such as terminals and servers, and includes the following steps S801, S803, S805 and S807.

In step S801, sample object attributes of sample objects, target number of sample operation sequence information, target number of sample resource category information and sample resource feature information of sample multimedia resources are acquired.

In step S803, input the sample object attributes, the target number of sample operation sequence information and the target number of sample resource category information into the object representation network to be trained for object representation, and obtain the target number of sample objects corresponding to the target number of sample resource category information characteristic information;

In step S805, determine the sample resource recommendation index according to the target number of sample object feature information and sample resource feature information;

In step S807, based on the resource recommendation index, the object representation network to be trained is trained to obtain the object representation network.

In some embodiments, the sample object may be a user account that has performed preset operations on multimedia resources in the recommendation system. The target number of sample operation sequence information is the operation association information corresponding to the multimedia resources belonging to the target number of sample resource category information among the multimedia resources recommended to the sample object within the sample time period; the target number of sample resource category information can be sample Category information of at least one multimedia resource preferred by the object.

In some embodiments, for detailed steps of the above steps S801 to S805, please refer to the detailed details of the above steps S201 to S205, which will not be repeated here.

In some embodiments, the above-mentioned training of the object representation network based on the resource recommendation index, and obtaining the object representation network may include generating first loss information according to the resource recommendation index; updating network parameters of the object representation network to be trained based on the first loss information, Based on the updated object representation network to be trained, repeat the above object representation, determine the sample resource recommendation index, generate the first loss information, and update the training iteration steps of the network parameters until the first preset convergence condition is met, and the first preset convergence condition will be satisfied. The object representation network to be trained corresponding to the convergence condition is set as the object representation network.

In some embodiments, the resource recommendation index may include at least one recommendation index corresponding to a task, which may be different according to actual application requirements. For example, the resource recommendation index may include the duration of the target object's browsing of multimedia resources. Correspondingly, the The longer the browsing time, the higher the probability that the multimedia resource will be recommended to the target object; the resource recommendation index may include whether the target object will perform preset operations on the multimedia resource (such as click, like, etc.). Correspondingly, in the process of generating the first loss information according to the resource recommendation index, the loss information corresponding to at least one task may be determined in combination with business requirements and a preset loss function.

In some embodiments, satisfying the above-mentioned first preset convergence condition may mean that the number of training iteration operations reaches a preset number of training times. In some embodiments, satisfying the first preset convergence condition may also be that the first loss information is less than a specified threshold. In the embodiment of this specification, the preset training times and the designated threshold can be preset in combination with the training speed and accuracy of the network in practical applications.

In some embodiments, the following methods are used to obtain the sample resource characteristic information of the sample multimedia resource:

Obtain the sample resource attribute of the sample multimedia resource;

Input the attributes of the sample resources into the resource characterization network to be trained for resource characterization, and obtain the feature information of the sample resources;

Correspondingly, based on the above-mentioned resource recommendation index, the object representation network to be trained is trained, and the obtained object representation network may include:

Based on the resource recommendation index, the object representation network to be trained and the resource representation network to be trained are trained to obtain the object representation network and the resource representation network.

In some embodiments, the above-mentioned training of the object representation network and the resource representation network to be trained based on the resource recommendation index, and obtaining the object representation network and the resource representation network may include generating the second loss information according to the resource recommendation index; based on the second loss information Update the network parameters of the object representation network to be trained and the resource representation network to be trained, repeat the above object representation, resource representation, determine the sample resource recommendation index, and generate the second loss information based on the updated object representation network and resource representation network to be trained , and the training iteration step of updating the network parameters until the second preset convergence condition is met, and the corresponding object representation network to be trained when the second preset convergence condition is met is used as the object representation network, and the second preset convergence condition When the corresponding resource representation network to be trained is used as the resource representation network.

In some embodiments, satisfying the above-mentioned second preset convergence condition may mean that the number of training iteration operations reaches a preset number of training times. In some embodiments, satisfying the preset convergence condition may also be that the second loss information is less than a specified threshold. In the embodiment of this specification, the preset number of training times and the specified threshold can be preset in combination with the training speed and accuracy of the network in practical applications.

In the above-mentioned embodiment, the combined training of the object representation network and the resource representation network can better improve the respective feature representation accuracy of the trained object representation network and resource representation network, and further improve the accuracy and accuracy of recommendation in the recommendation system. Recommended effect.

In the embodiment of this specification, during the training process of the object representation network, the sample object attributes of the sample object and the sample operation sequence information corresponding to the target number of sample resource category information are used to perform sample object characterization, which can better improve the sample object. The accuracy of the characteristic information on the interest preference of the target object, and since the sample operation sequence information is extracted from the object historical operation association information by the sample resource category information of the target number of sample object preferences, the object feature granularity in the training process can be reduced to The resource category is approximated, and the number of operations of the object representation processing after the introduction of the sample operation sequence information is simplified to the number of categories, which can effectively improve the processing efficiency of the object representation processing during the training process, and then improve the representation accuracy of the trained object representation network. On the one hand, the training efficiency is greatly improved and the system performance is improved.

In some embodiments, as shown in FIG. 9 , FIG. 9 is a schematic diagram of a recommendation system provided according to some embodiments. Referring to FIG. 9 , the recommendation system may include a recall module, a rough sorting module, a resource category determination module, a historical operation sequence storage module, an object representation server, a resource recommendation prediction module, a resource representation server and a network storage server. In order to improve the recommendation efficiency, the object representation process can be performed while the recall module recalls the multimedia resources to be recommended. The resource category determination module can combine the object historical operation association information stored in the historical operation sequence to determine the target number of target resource category information that the target object likes; the object representation server combines the target number of target resource category information corresponding to the historical operation sequence information and The object representation network stored in the network storage server performs object representation to obtain object feature information; the recall module can transfer the recalled multimedia resources and object feature information to the rough sorting module; then, the rough sorting module can call the resource prediction module to recommend resources For index prediction, the resource characterization server can continuously update the resource feature information of the multimedia resources in the recommendation system according to the preset frequency, combined with the resource characterization network stored in the network storage server; correspondingly, the resource prediction module can obtain the multimedia resources to be recommended from the resource characterization server resource feature information of the resource; in some embodiments, in order to improve system performance, the resource characterization server can transmit the resource feature information of the multimedia resource to be recommended to the resource recommendation prediction module through a message queue; further, the resource recommendation prediction module can combine The resource feature information and object feature information of the multimedia resources to be recommended are used to determine the resource recommendation index, and combined with the resource recommendation index, the target multimedia resource is determined.

Correspondingly, the above-mentioned response to the multimedia resource acquisition request of the target object includes:

In the case of receiving a multimedia resource acquisition request, triggering in parallel a recall instruction corresponding to the multimedia resource to be recommended and an object characterization instruction corresponding to the target object;

In a specific embodiment, the above-mentioned recall instruction may be used to instruct to recall multimedia resources to be recommended, and the above-mentioned object characterization instruction may be used to instruct to execute object characterization processing corresponding to the target object.

In the above embodiment, when a multimedia resource acquisition request is received, the recall operation of media resources to be recommended and the operation of object representation processing are performed in parallel, which can greatly reduce the time consumption of recommendation processing and improve the efficiency of recommendation processing.

Fig. 10 is a block diagram of an apparatus for recommending multimedia resources according to some embodiments. 10, the device includes a data acquisition module 1010, a first object representation module 1020, a target multimedia resource determination module 1030, and a resource recommendation module 1040:

The data acquisition module 1010 is configured to, in response to the multimedia resource acquisition request of the target object, acquire the target object attribute of the target object, the target number of historical operation sequence information, the target number of target resource category information and the resource characteristic information of the multimedia resource to be recommended , the target number of historical operation sequence information is the operation association information corresponding to the multimedia resources belonging to the target number of target resource category information among the historical multimedia resources recommended to the target object within the historical time period;

The first object characterization module 1020 is configured to input the target object attribute, the target number of historical operation sequence information and the target number of target resource category information into the object characterization network for object characterization, and obtain the target number corresponding to the target number of target resource category information A first object feature information;

The target multimedia resource determining module 1030 is configured to determine the target multimedia resource from the multimedia resources to be recommended according to the target number of first object characteristic information and resource characteristic information;

The resource recommendation module 1040 is configured to recommend resources to target objects based on the target multimedia resources.

In some embodiments, the object representation network includes: an object feature extraction network, a feature intersection processing network, a splicing network, and a feature fusion network;

The first object characterization module 1020 includes:

The feature extraction processing unit is configured to perform feature extraction processing on target object attributes, target quantity historical operation sequence information and target quantity target resource category information based on the object feature extraction network, to obtain target quantity sequence feature information and target quantity Category object feature information, the target number of category object feature information is the feature information corresponding to the target object attributes and the target number of target resource category information;

The feature cross processing unit is configured to perform feature cross processing on the target number of sequence feature information and the target number of category object feature information based on the feature cross processing network to obtain the target number of cross feature information;

The feature extraction processing unit includes:

The first feature extraction processing subunit is configured to input the target object attributes and the target number of target resource category information into the first feature extraction network to perform feature extraction processing, and obtain the target number of category object feature information;

The second feature extraction processing subunit is configured to input a target number of historical operation sequence information into the second feature extraction network for feature extraction processing, and obtain a target number of sequence feature information.

In some embodiments, the target multimedia resource determination module 1030 includes:

The first resource category information determining unit is configured to determine resource category information of multimedia resources to be recommended;

The first resource recommendation index determination unit is configured to determine the resource recommendation index of the primary multimedia resource according to the resource feature information of the primary multimedia resource and the first object feature information corresponding to the resource category information of the primary multimedia resource;

In some embodiments, the object representation network includes a base object representation network; the apparatus further includes:

The target multimedia resource determining module 1030 is further configured to: determine the target multimedia resource from the multimedia resources to be recommended according to the target number of first object characteristic information, second object characteristic information and resource characteristic information.

In some embodiments, the multimedia resources to be recommended include multiple multimedia resources; the target multimedia resource determination module 1030 includes:

a second resource type information determining unit configured to determine resource type information of a plurality of multimedia resources;

The second resource recommendation index determination unit is configured to respond to resource category information of any multimedia resource being included in target resource category information of a target quantity, according to the resource characteristic information of the first multimedia resource and the first multimedia resource The resource category information corresponding to the first object characteristic information, determine the resource recommendation index of the first multimedia resource; resource;

The third resource recommendation indicator determination unit is configured to respond to resource category information of any multimedia resource not included in the target number of target resource category information, according to the resource characteristic information of the second multimedia resource and the second object characteristic information , determining the resource recommendation index of the second multimedia resource; the second multimedia resource is a multimedia resource whose resource category information in the multimedia resource to be recommended is not included in the target number of target resource category information;

The second target multimedia resource determining unit is configured to determine the target multimedia resource from the multimedia resources to be recommended based on the resource recommendation index of the first multimedia resource and the resource recommendation index of the second multimedia resource.

In some embodiments, the data acquisition module 1010 includes:

The resource data acquisition unit is configured to determine the preset number of resource category information to which the historical multimedia resource belongs and the resource quantity corresponding to the preset number of resource category information;

The target resource category information determining unit is configured to determine a target number of target resource category information from a preset number of resource category information based on the resource quantity;

The historical operation sequence information generation unit is configured to generate a target number of historical operation sequence information based on the operation association information of multimedia resources corresponding to the target number of target resource category information among the historical multimedia resources.

In some embodiments, the data acquisition module 1010 includes:

The resource characterization unit is configured to input the attribute of the target resource into the resource characterization network for resource characterization, and obtain resource feature information of the multimedia resource to be recommended.

In an optional embodiment, the data acquisition module 1010 is further configured to: in response to receiving a multimedia resource acquisition request, trigger in parallel a recall instruction corresponding to the multimedia resource to be recommended and an object characterization instruction corresponding to the target object;

Wherein, the recall instruction is used to instruct to recall multimedia resources to be recommended, and the object characterization instruction is used to instruct to execute object characterization processing corresponding to the target object.

Fig. 11 is a block diagram of an apparatus for generating an object representation network according to some embodiments. Referring to FIG. 11 , the device includes a sample data acquisition module 1110, a third object characterization module 1120, a sample resource recommendation index determination module 1130, and a network training module 1140:

The sample data acquiring module 1110 is configured to acquire the sample object attribute of the sample object, the target number of sample operation sequence information, the target number of sample resource category information and the sample resource characteristic information of the sample multimedia resource, the target number of sample operation sequence information is Among the multimedia resources recommended to the sample object within the sample time period, the operation related information corresponding to the multimedia resources belonging to the target number of sample resource category information;

The third object characterization module 1120 is configured to input the sample object attributes, the target number of sample operation sequence information and the target number of sample resource category information into the object characterization network to be trained for object characterization, and obtain the target number of sample resource category information corresponding to Target number of sample object feature information;

The sample resource recommendation index determination module 1130 is configured to determine the sample resource recommendation index according to the target number of sample object feature information and sample resource feature information;

The network training module 1140 is configured to train the object representation network to be trained based on the resource recommendation index to obtain the object representation network.

In some embodiments, sample data acquisition module 1110 includes:

The resource characterization unit is configured to input the attributes of the sample resources into the resource characterization network to be trained for resource characterization, and obtain the characteristic information of the sample resources;

The network training module is further configured to: train the object representation network to be trained and the resource representation network to be trained based on the resource recommendation index to obtain the object representation network and the resource representation network.

Regarding the apparatus in the foregoing embodiments, the specific manner in which each module executes operations has been described in detail in the embodiments related to the method, and will not be described in detail here.

Fig. 12 is a block diagram of an electronic device for recommending multimedia resources or generating an object representation network according to some embodiments. The electronic device may be a terminal, and its internal structure may be as shown in Fig. 12 . The electronic device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Wherein, the processor of the electronic device is used to provide calculation and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for recommending multimedia resources or a method for generating an object representation network is realized. The display screen of the electronic device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic device may be a touch layer covered on the display screen, or a button, a trackball or a touch pad provided on the housing of the electronic device , and can also be an external keyboard, touchpad, or mouse.

Fig. 13 is a block diagram of an electronic device for recommending multimedia resources or generating an object representation network according to some embodiments. The electronic device may be a server, and its internal structure may be as shown in Fig. 13 . The electronic device includes a processor, memory and network interface connected by a system bus. Wherein, the processor of the electronic device is used to provide calculation and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for recommending multimedia resources or a method for generating an object representation network is implemented.

Those skilled in the art can understand that the structure shown in FIG. 12 or FIG. 13 is only a block diagram of a partial structure related to the disclosed solution, and does not constitute a limitation on the electronic device to which the disclosed solution is applied. Specifically The electronic device may include more or fewer components than shown in the figures, or combine certain components, or have a different arrangement of components.

In an exemplary embodiment, there is also provided an electronic device, including: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions, so as to realize the implementation of the present disclosure. In this example, a method for recommending multimedia resources or a method for generating object representation networks.

In an exemplary embodiment, a computer-readable storage medium is also provided. When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device can execute the multimedia resource recommendation method or the A Generative Approach to Object Representation Networks.

In an exemplary embodiment, there is also provided a computer program product containing instructions, which, when run on a computer, cause the computer to execute the method for recommending multimedia resources or the method for generating an object representation network in the embodiments of the present disclosure.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be realized by instructing related hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium , when the computer program is executed, it may include the procedures of the embodiments of the above-mentioned methods. Wherein, any references to memory, storage, database or other media used in the various embodiments provided in the present application may include non-volatile and/or volatile memory. Nonvolatile memory can include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Chain Synchlink DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Other embodiments of the present disclosure will be readily apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present disclosure, and these modifications, uses or adaptations follow the general principles of the present disclosure and include common knowledge or conventional technical means in the technical field not disclosed in the present disclosure . The specification and examples are to be considered exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It should be understood that the present disclosure is not limited to the precise constructions which have been described above and shown in the drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

A method for recommending multimedia resources, comprising:

In response to the multimedia resource acquisition request of the target object, acquire the target object attributes of the target object, the target number of historical operation sequence information, the target number of target resource category information and the resource feature information of the multimedia resources to be recommended, the target The number of historical operation sequence information is the operation association information corresponding to the multimedia resources belonging to the target number of target resource category information among the historical multimedia resources recommended to the target object within the historical time period;

inputting the target object attributes, the target number of historical operation sequence information and the target number of target resource category information into an object characterization network for object characterization, and obtaining the target quantity corresponding to the target number of target resource category information A first object feature information;

Determine a target multimedia resource from the multimedia resources to be recommended according to the target number of first object feature information and the resource feature information;

Based on the target multimedia resource, resource recommendation is performed to the target object.
The method for recommending multimedia resources according to claim 1, wherein the object representation network comprises: an object feature extraction network, a feature cross processing network, a splicing network, and a feature fusion network;

Said inputting said target object attributes, said target number of historical operation sequence information and said target number of target resource category information into an object characterization network for object characterization, and obtaining said target number of target resource category information corresponding to The target number of first object feature information includes:

Based on the object feature extraction network, perform feature extraction processing on the target object attributes, the target number of historical operation sequence information, and the target number of target resource category information, and obtain the target number of sequence feature information and all The target number of category object feature information, the target number of category object feature information is the feature information corresponding to the target object attributes and the target number of target resource category information;

Based on the feature intersection processing network, perform feature intersection processing on the target number of sequence feature information and the target number of category object feature information to obtain the target number of cross feature information;

Perform splicing processing on the target number of cross feature information and the target number of category object feature information based on the splicing network to obtain the target number of spliced feature information;

The target number of spliced feature information is respectively fused based on the feature fusion network to obtain the target number of first object feature information.
The method for recommending multimedia resources according to claim 2, wherein the object feature extraction network comprises a first feature extraction network and a second feature extraction network;

Based on the object feature extraction network, perform feature extraction processing on the target object attributes, the target number of historical operation sequence information and the target number of target resource category information to obtain the target number of sequence feature information And the target number category object feature information includes:

Inputting the target object attributes and the target number of target resource category information into the first feature extraction network for feature extraction processing to obtain the target number of category object feature information;

Inputting the target number of historical operation sequence information into the second feature extraction network for feature extraction processing to obtain the target number of sequence feature information.
The method for recommending multimedia resources according to any one of claims 1 to 3, wherein, according to the target number of first object feature information and the resource feature information, the target multimedia resource is determined from the multimedia resources to be recommended. Resources include:

determining resource category information of the multimedia resource to be recommended;

Using resource category information included in the target number of target resource category information to be recommended multimedia resources as primary multimedia resources;

Determine the resource recommendation index of the primary multimedia resource according to the resource feature information of the primary multimedia resource and the first object feature information corresponding to the resource category information of the primary multimedia resource;

Based on the resource recommendation index, the target multimedia resource is determined from the primary multimedia resources.
The method for recommending multimedia resources according to any one of claims 1 to 3, wherein the object representation network comprises a basic object representation network; the method further comprises:

inputting the attributes of the target object into the basic object characterization network for object characterization to obtain second object feature information;

The determining the target multimedia resource from the multimedia resources to be recommended according to the target number of first object feature information and the resource feature information includes:

The target multimedia resource is determined from the multimedia resources to be recommended according to the target number of first object feature information, the second object feature information, and the resource feature information.
The method for recommending multimedia resources according to claim 5, wherein, the multimedia resources to be recommended include a plurality of multimedia resources; the first object characteristic information, the second object characteristic information and the The resource feature information, from the multimedia resources to be recommended, determining the target multimedia resource includes:

determining resource category information of the plurality of multimedia resources;

In response to the resource type information of any multimedia resource being included in the target number of target resource type information, according to the resource feature information of the first multimedia resource and the resource type information of the first multimedia resource corresponding to the first An object feature information to determine the resource recommendation index of the first multimedia resource; the first multimedia resource is the resource category information of the multimedia resource to be recommended included in the target number of target resource category information multimedia resources;

In response to resource type information of any multimedia resource not being included in the target number of target resource type information, determining the second multiple A resource recommendation index of a media resource; the second multimedia resource is a multimedia resource whose resource category information in the multimedia resource to be recommended is not included in the target number of target resource category information;

Based on the resource recommendation index of the first multimedia resource and the resource recommendation index of the second multimedia resource, the target multimedia resource is determined from the multimedia resources to be recommended.
The method for recommending multimedia resources according to any one of claims 1 to 3, wherein obtaining the target number of historical operation sequence information includes:

Determine the preset number of resource category information to which the historical multimedia resource belongs and the resource quantity corresponding to the preset number of resource category information;

Determining the target number of target resource category information from a preset number of resource category information based on the resource quantity;

The target number of historical operation sequence information is generated based on the operation association information of the multimedia resources corresponding to the target number of target resource category information among the historical multimedia resources.
The method for recommending multimedia resources according to any one of claims 1 to 3, wherein obtaining resource feature information of the multimedia resources to be recommended comprises:

Obtain the target resource attribute of the multimedia resource to be recommended;

Inputting the attribute of the target resource into a resource characterization network for resource characterization to obtain resource feature information of the multimedia resource to be recommended.
The multimedia resource recommendation method according to any one of claims 1 to 3, wherein said response to the multimedia resource acquisition request of the target object comprises:

In response to receiving the multimedia resource acquisition request, triggering in parallel a recall instruction corresponding to the multimedia resource to be recommended and an object characterization instruction corresponding to the target object;

Wherein, the recall instruction is used to instruct to recall the multimedia resource to be recommended, and the object characterization instruction is used to instruct to execute object characterization processing corresponding to the target object.
A method for generating an object representation network, comprising:

Obtain the sample object attributes of the sample object, the target number of sample operation sequence information, the target number of sample resource category information, and the sample resource feature information of the sample multimedia resource, and the target number of sample operation sequence information is recommended within the sample time period Among the multimedia resources of the sample object, the operation association information corresponding to the multimedia resources belonging to the category information of the target number of sample resources;

Input the sample object attributes, the target number of sample operation sequence information and the target number of sample resource category information into the object representation network to be trained for object representation, and obtain the target number of sample resource category information corresponding to Target number of sample object feature information;

Determine a sample resource recommendation index according to the target number of sample object feature information and the sample resource feature information;

Based on the resource recommendation index, train the object representation network to be trained to obtain an object representation network.
According to the method for generating an object representation network according to claim 10, obtaining the sample resource characteristic information of the sample multimedia resource comprises:

Obtain the sample resource attribute of the sample multimedia resource;

inputting the attributes of the sample resources into the resource characterization network to be trained for resource characterization, and obtaining feature information of the sample resources;

The step of training the object representation network to be trained based on the resource recommendation index, and obtaining the object representation network includes:

Based on the resource recommendation index, train the object representation network to be trained and the resource representation network to be trained to obtain the object representation network and resource representation network.
A device for recommending multimedia resources, comprising:

The data acquisition module is configured to, in response to the multimedia resource acquisition request of the target object, acquire the target object attributes of the target object, the target number of historical operation sequence information, the target number of target resource category information and the multimedia resources to be recommended Resource feature information, the target number of historical operation sequence information is the operation association information corresponding to the multimedia resources belonging to the target number of target resource category information among the historical multimedia resources recommended to the target object within the historical time period;

The first object characterization module is configured to input the target object attributes, the target number of historical operation sequence information and the target number of target resource category information into the object characterization network for object characterization, and obtain the target number of targets The target number of first object feature information corresponding to the resource category information;

The target multimedia resource determination module is configured to determine a target multimedia resource from the multimedia resources to be recommended according to the target number of first object feature information and the resource feature information;

The resource recommendation module is configured to recommend resources to the target object based on the target multimedia resources.
The multimedia resource recommendation device according to claim 12, wherein the object representation network comprises: an object feature extraction network, a feature intersection processing network, a splicing network, and a feature fusion network;

The first object characterization module includes:

The feature extraction processing unit is configured to perform feature extraction processing on the target object attributes, the target number of historical operation sequence information and the target number of target resource category information based on the object feature extraction network, to obtain the A target number of sequence feature information and the target number of category object feature information, where the target number of category object feature information is feature information corresponding to the target object attributes and the target number of target resource category information;

The feature intersection processing unit is configured to perform feature intersection processing on the target number of sequence feature information and the target number of category object feature information based on the feature intersection processing network to obtain the target number of intersection feature information;

The splicing processing unit is configured to splice the target number of cross feature information and the target number of category object feature information based on the splicing network to obtain the target number of spliced feature information;

The fusion processing unit is configured to respectively perform fusion processing on the target number of spliced feature information based on the feature fusion network to obtain the target number of first object feature information.
The multimedia resource recommendation device according to claim 13, wherein the object feature extraction network comprises a first feature extraction network and a second feature extraction network;

The feature extraction processing unit includes:

The first feature extraction processing subunit is configured to input the target object attributes and the target number of target resource category information into the first feature extraction network to perform feature extraction processing, and obtain the target number of category object features information;

The second feature extraction processing subunit is configured to input the target amount of historical operation sequence information into the second feature extraction network for feature extraction processing, and obtain the target amount of sequence feature information.
The device for recommending multimedia resources according to any one of claims 12 to 14, wherein the target multimedia resource determination module includes:

The first resource type information determining unit is configured to determine the resource type information of the multimedia resource to be recommended;

The primary multimedia resource determination unit is configured to include the resource category information in the target number of target resource category information to be recommended multimedia resources as the primary multimedia resource;

The first resource recommendation index determining unit is configured to determine the resource of the primary multimedia resource according to the resource feature information of the primary multimedia resource and the first object feature information corresponding to the resource category information of the primary multimedia resource. Recommended indicators;

The first target multimedia resource determining unit is configured to determine the target multimedia resource from the primary multimedia resources based on the resource recommendation index.
The device for recommending multimedia resources according to any one of claims 12 to 14, wherein the object representation network comprises a basic object representation network; the device further comprises:

The second object representation network is configured to input the attributes of the target object into the basic object representation network for object representation to obtain second object feature information;

The target multimedia resource determining module is further configured to: determine the Target multimedia resource.
The device for recommending multimedia resources according to claim 16, wherein the multimedia resources to be recommended include a plurality of multimedia resources; the target multimedia resource determination module includes:

a second resource type information determining unit configured to determine resource type information of the plurality of multimedia resources;

The second resource recommendation index determining unit is configured to respond to resource category information of any multimedia resource being included in the target number of target resource category information, according to the resource feature information of the first multimedia resource and the first The first object feature information corresponding to the resource category information of the multimedia resource determines the resource recommendation indicator of the first multimedia resource; the first multimedia resource is the resource category information of the multimedia resource to be recommended. Multimedia resources with target resource category information in said target number;

The third resource recommendation index determining unit is configured to respond to resource category information of any multimedia resource not included in the target number of target resource category information, according to the resource characteristic information of the second multimedia resource and the first multimedia resource. Two object characteristic information, to determine the resource recommendation index of the second multimedia resource; the second multimedia resource is the resource category information of the multimedia resource to be recommended that is not included in the target number of target resource category information Multimedia resources in ;

The second target multimedia resource determining unit is configured to determine the multimedia resource to be recommended based on the resource recommendation index of the first multimedia resource and the resource recommendation index of the second multimedia resource. Target multimedia resource.
The device for recommending multimedia resources according to any one of claims 12 to 14, wherein the data acquisition module includes:

A resource data acquisition unit configured to determine a preset number of resource category information to which the historical multimedia resource belongs and a resource quantity corresponding to the preset number of resource category information;

The target resource category information determining unit is configured to determine the target number of target resource category information from a preset number of resource category information based on the resource quantity;

The historical operation sequence information generation unit is configured to generate the target amount of historical operation sequence information based on the operation association information of the multimedia resources corresponding to the target amount of target resource category information among the historical multimedia resources.
The device for recommending multimedia resources according to any one of claims 12 to 14, wherein the data acquisition module includes:

a target resource attribute acquisition unit configured to acquire target resource attributes of multimedia resources to be recommended;

The resource characterization unit is configured to input the attribute of the target resource into a resource characterization network for resource characterization, and obtain resource characteristic information of the multimedia resource to be recommended.
The device for recommending multimedia resources according to any one of claims 12 to 14, wherein the data acquisition module is further configured to: trigger a recall corresponding to the multimedia resources to be recommended in parallel in response to receiving the multimedia resource acquisition request an object representation instruction corresponding to the instruction and the target object;

Wherein, the recall instruction is used to instruct to recall the multimedia resource to be recommended, and the object characterization instruction is used to instruct to execute object characterization processing corresponding to the target object.
A device for generating an object representation network, comprising:

The sample data acquisition module is configured to acquire the sample object attributes of the sample object, the target number of sample operation sequence information, the target number of sample resource category information and the sample resource characteristic information of the sample multimedia resource, and the target number of sample operations The sequence information is the operation related information corresponding to the multimedia resources that belong to the target number of sample resource category information among the multimedia resources recommended to the sample object within the sample time period;

The third object characterization module is configured to input the sample object attributes, the target number of sample operation sequence information and the target number of sample resource category information into the object characterization network to be trained for object characterization, and obtain the target number The target number of sample object feature information corresponding to the sample resource category information;

The sample resource recommendation index determination module is configured to determine the sample resource recommendation index according to the target number of sample object feature information and the sample resource feature information;

The network training module is configured to train the object representation network to be trained based on the resource recommendation index to obtain an object representation network.
The device for generating an object representation network according to claim 21, wherein the sample data acquisition module includes:

a sample resource attribute acquiring unit configured to acquire a sample resource attribute of a sample multimedia resource;

The resource characterization unit is configured to input the sample resource attributes into the resource characterization network to be trained for resource characterization, and obtain the sample resource feature information;

The network training module is also configured to: based on the resource recommendation index, train the object representation network to be trained and the resource representation network to be trained to obtain the object representation network and resource representation network.
An electronic device comprising:

processor;

memory for storing said processor-executable instructions;

Wherein, the processor is configured to execute the instructions to implement the method for recommending multimedia resources according to any one of claims 1 to 9 or the method for generating an object representation network according to any one of claims 10 to 11.
A computer-readable storage medium, when the instructions in the storage medium are executed by the processor of the electronic device, the electronic device can execute the method for recommending multimedia resources according to any one of claims 1 to 9 or The method for generating the object representation network described in any one of 10 to 11.
A computer program product, comprising computer instructions, wherein, when the computer instructions are executed by a processor, the multimedia resource recommendation method described in any one of claims 1 to 9 or the object representation described in any one of claims 10 to 11 is implemented How the network is generated.