CN110569447B

CN110569447B - Network resource recommendation method and device and storage medium

Info

Publication number: CN110569447B
Application number: CN201910867402.4A
Authority: CN
Inventors: 张龙
Original assignee: Tencent Music Entertainment Technology Shenzhen Co Ltd
Current assignee: Tencent Music Entertainment Technology Shenzhen Co Ltd
Priority date: 2019-09-12
Filing date: 2019-09-12
Publication date: 2022-03-15
Anticipated expiration: 2039-09-12
Also published as: CN110569447A

Abstract

The application discloses a recommendation method, a recommendation device and a storage medium of network resources, and belongs to the field of data processing. The method comprises the following steps: acquiring the dominant characteristic and the recessive characteristic of a reference network resource, determining the similarity between each network resource in a first network resource library and the reference network resource through a correlation model based on the dominant characteristic and the recessive characteristic of the reference network resource and the dominant characteristic and the recessive characteristic of each network resource in the first network resource library, and recommending resources for a user based on the similarity between each network resource in the first network resource library and the reference network resource. According to the method and the device, the similarity of any two sample network resources is evaluated through the attribute characteristics, the statistical characteristics and the recessive characteristics of the network resources, and then resource recommendation is performed based on the similarity of the network resources, so that the evaluation of the similarity is more comprehensive and accurate, the accuracy of determining the similarity of the network resources is improved, and the recommendation accuracy is further improved.

Description

Network resource recommendation method and device and storage medium

Technical Field

The present application relates to the field of data processing, and in particular, to a method, an apparatus, and a storage medium for recommending network resources.

Background

With the continuous development of science and technology, more and more network resources can be browsed on the network by a user, and the network resources can be songs, videos, news and the like. In addition, in order to meet the browsing requirements of the user, in the process of browsing the network resources by the user, the resource platform also needs to recommend similar network resources for the user according to the historical browsing records of the user. For example, when a user listens to a song at the song playback platform, the song playback platform may recommend similar songs for the user based on the user's song playback record.

In the related art, the resource platform generally recommends resources for users according to the attribute similarity of network resources. For example, for a reference network resource browsed by a user, the resource platform may obtain attributes such as a title or a label of the reference network resource, determine the similarity between each network resource in the network resource library and the reference network resource according to the similarity between the attribute of each network resource in the network resource library and the attribute of the reference network resource, and recommend a resource for the user based on the similarity between each network resource in the network resource library and the reference network resource. For example, the network resource with the greatest similarity to the reference network resource in the network resource library is recommended to the user.

However, the similarity between the two network resources is evaluated only according to the similarity between the attributes of the two network resources, so that the accuracy of the determined similarity is low, and further the recommendation accuracy of the network resources is low.

Disclosure of Invention

The application provides a recommendation method, a recommendation device and a storage medium for network resources, which can solve the problem that the accuracy of the determined similarity is low and further the recommendation accuracy of the network resources is low due to the fact that the similarity of two network resources is evaluated only according to the similarity between the attributes of the two network resources. The technical scheme is as follows:

in one aspect, a method for recommending network resources is provided, where the method includes:

acquiring dominant features and recessive features of reference network resources, wherein the dominant features comprise attribute features and statistical features, the recessive features refer to vector representation mapped to a recessive space, and the reference network resources are network resources browsed by a user;

determining the similarity between each network resource in the first network resource library and the reference network resource through a correlation model based on the dominant and recessive characteristics of the reference network resource and the dominant and recessive characteristics of each network resource in the first network resource library, wherein the correlation model is used for determining the similarity between any two network resources;

and recommending resources for the user based on the similarity between each network resource in the first network resource library and the reference network resource.

Optionally, the obtaining an implicit characteristic of the reference network resource includes:

mapping the reference network resource to a browsing behavior space to obtain a first word embedded vector of the reference network resource, and determining the first word embedded vector of the reference network resource as a recessive feature of the reference network resource; alternatively, the first and second electrodes may be,

mapping the reference network resource to a browsing behavior space to obtain a first word embedded vector of the reference network resource; mapping the reference network resource to a resource content space to obtain a second word embedded vector of the reference network resource; and determining the first word embedding vector and the second word embedding vector of the reference network resource as the implicit characteristics of the reference network resource.

Optionally, when the reference network resource is a song, the attribute feature includes at least one of a singer, a genre to which the singer belongs, an album to which the singer belongs, a release time, and a region in which the singer is located, and the statistical feature includes at least one of a play amount, a collection amount, and a comment amount.

Optionally, the determining, by using a correlation model, a similarity between each network resource in the first network resource pool and the reference network resource based on the explicit characteristic and the implicit characteristic of the reference network resource and the explicit characteristic and the implicit characteristic of each network resource in the first network resource pool includes:

and for a first network resource in the first network resource library, taking the dominant characteristic and the recessive characteristic of the reference network resource and the dominant characteristic and the recessive characteristic of the first network resource as the input of the association model, and determining the similarity between the first network resource and the reference network resource through the association model, wherein the first network resource is any one of the first network resource library.

comparing the statistical characteristics of the reference network resources with the statistical characteristics of each network resource in the first network resource library to obtain the comparison characteristics between each network resource in the first network resource library and the reference network resources;

for a first network resource in the first network resource library, taking the dominant characteristic and the recessive characteristic of the reference network resource, and the dominant characteristic, the recessive characteristic and the comparison characteristic between the first network resource and the reference network resource as the input of the association model, and determining the similarity between the first network resource and the reference network resource through the association model, wherein the first network resource is any network resource in the first network resource library.

Optionally, the recommending resources for the user based on the similarity between each network resource in the first network resource library and the reference network resource includes:

sequencing the network resources in the first network resource library according to the sequence of similarity between the network resources and the reference network resources from large to small;

and acquiring n network resources ranked at the top from the ranking result, and recommending the n network resources ranked at the top to the user, wherein n is a positive integer.

Optionally, before determining, by using a correlation model, a similarity between each network resource in the first network resource pool and the reference network resource based on the explicit characteristic and the implicit characteristic of the reference network resource and the explicit characteristic and the implicit characteristic of each network resource in the first network resource pool, the method further includes:

obtaining implicit characteristics of a plurality of sample network resources, wherein the sample network resources are network resources browsed by a sample user;

determining k similar network resources of each sample network resource from a second network resource library according to the implicit characteristic of each sample network resource in the plurality of sample network resources, wherein k is an integer greater than 1;

determining a plurality of first positive sample data and a plurality of first negative sample data based on the associated recommendation results of k similar network resources of each sample network resource in the plurality of sample network resources;

each first positive sample data comprises a first sample network resource and a positive similar network resource of the first sample network resource, each first negative sample data comprises a second sample network resource and a negative similar network resource of the second sample network resource, the first sample network resource and the second sample network resource are any one of the plurality of sample network resources, the positive similar network resource refers to a similar network resource which is accepted by a user after being recommended to the user in k similar network resources of the first sample network resource, and the negative similar network resource refers to a similar network resource which is not accepted by the user after being recommended to the user in k similar network resources of the second sample network resource;

determining a plurality of second positive sample data and a plurality of second negative sample data based on the plurality of first positive sample data and the plurality of first negative sample data;

each second positive sample data comprises an explicit characteristic and an implicit characteristic of the first sample network resource, and an explicit characteristic and an implicit characteristic of a positive similar network resource of the first sample network resource, and each second negative sample data comprises an explicit characteristic and an implicit characteristic of the second sample network resource, and an explicit characteristic and an implicit characteristic of a negative similar network resource of the second sample network resource;

and training an association model to be trained based on the plurality of second positive sample data and the plurality of second negative sample data to obtain the association model.

Optionally, the determining k similar network resources of each sample network resource from the second network resource library according to the implicit characteristic of each sample network resource in the plurality of sample network resources includes:

for any sample network resource in the multiple sample network resources, determining the similarity between each network resource in the second network resource library and any sample network resource according to the first word embedding vector of any sample network resource in the browsing behavior space and the first word embedding vector of each network resource in the second network resource library in the browsing behavior space; determining k similar network resources of any sample network resource from the second network resource library based on the similarity of each network resource in the second network resource library to the any sample network resource; alternatively, the first and second electrodes may be,

for any sample network resource in the plurality of sample network resources, determining the similarity between each network resource in the second network resource library and any sample network resource according to a second word embedding vector of the any sample network resource on a resource content space and a second word embedding vector of each network resource in the second network resource library on the resource content space; determining k similar network resources of the any sample network resource from the second network resource library based on the similarity of each network resource in the second network resource library to the any sample network resource.

Optionally, said determining a plurality of second positive sample data and a plurality of second negative sample data based on said plurality of first positive sample data and a plurality of first negative sample data comprises:

for any first positive sample data in the plurality of first positive sample data, respectively extracting explicit characteristics of the first sample network resources and positive similar network resources of the first sample network resources included in the any first positive sample data; determining second positive sample data corresponding to any one first positive sample data based on the dominant characteristic and the recessive characteristic of the first sample network resource and the dominant characteristic and the recessive characteristic of the positive similar network resource of the first sample network resource;

for any first negative sample data in the plurality of first negative sample data, respectively extracting explicit characteristics of the second sample network resources and negative similar network resources of the second sample network resources included in the any first negative sample data; and determining second negative sample data corresponding to any first negative sample data based on the explicit characteristics and the implicit characteristics of the second sample network resources and the explicit characteristics and the implicit characteristics of the negative similar network resources of the second sample network resources.

Optionally, the determining second positive sample data corresponding to any one of the first positive sample data based on the explicit characteristic and the implicit characteristic of the first sample network resource and the explicit characteristic and the implicit characteristic of the positive similar network resource of the first sample network resource includes:

determining the dominant characteristic and the recessive characteristic of the first sample network resource and the dominant characteristic and the recessive characteristic of the positive similar network resource of the first sample network resource as second positive sample data corresponding to any one first positive sample data; or;

comparing the statistical characteristics of the first sample network resource with the statistical characteristics of the positive similar network resource of the first sample network resource to obtain the comparison characteristics between the positive similar network resource of the first sample network resource and the first sample network resource; and determining the dominant characteristic and the recessive characteristic of the first sample network resource, the dominant characteristic and the recessive characteristic of the positive similar network resource of the first sample network resource, and the contrast characteristic between the positive similar network resource of the first sample network resource and the first sample network resource as second positive sample data corresponding to any one of the first positive sample data.

Optionally, the determining second negative sample data corresponding to any one of the first negative sample data based on the explicit characteristic and the implicit characteristic of the second sample network resource and the explicit characteristic and the implicit characteristic of the negative similar network resource of the second sample network resource includes:

determining the dominant characteristic and the recessive characteristic of the second sample network resource and the dominant characteristic and the recessive characteristic of the negative similar network resource of the second sample network resource as second negative sample data corresponding to any one of the first negative sample data; or;

comparing the statistical characteristics of the second sample network resources with the statistical characteristics of the positive similar network resources of the second sample network resources to obtain the comparison characteristics between the negative similar network resources of the second sample network resources and the second sample network resources; and determining the dominant characteristic and the recessive characteristic of the second sample network resource, the dominant characteristic and the recessive characteristic of the negative similar network resource of the second sample network resource, and the contrast characteristic between the negative similar network resource of the second sample network resource and the second sample network resource as second negative sample data corresponding to any one first negative sample data.

In another aspect, an apparatus for recommending network resources is provided, the apparatus including:

the system comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring dominant features and recessive features of reference network resources, the dominant features comprise attribute features and statistical features, the recessive features refer to vector representations mapped to a recessive space, and the reference network resources are network resources browsed by a user;

a first determining module, configured to determine, based on the explicit characteristic and the implicit characteristic of the reference network resource and the explicit characteristic and the implicit characteristic of each network resource in a first network resource library, a similarity between each network resource in the first network resource library and the reference network resource through an association model, where the association model is used to determine a similarity between any two network resources;

and the recommending module is used for recommending resources for the user based on the similarity between each network resource in the first network resource library and the reference network resource.

Optionally, the obtaining module is configured to:

Optionally, the first determining module is configured to:

Optionally, the recommendation module is configured to:

Optionally, the apparatus further comprises:

the second acquisition module is used for acquiring the implicit characteristics of a plurality of sample network resources, wherein the sample network resources are network resources browsed by a sample user;

a second determining module, configured to determine, according to an implicit characteristic of each sample network resource in the multiple sample network resources, k similar network resources of each sample network resource from a second network resource library, where k is an integer greater than 1;

a third determining module, configured to determine, based on the association recommendation result of k similar network resources of each sample network resource in the plurality of sample network resources, a plurality of first positive sample data and a plurality of first negative sample data;

a fourth determining module, configured to determine, based on the plurality of first positive sample data and the plurality of first negative sample data, a plurality of second positive sample data and a plurality of second negative sample data;

each first positive sample data comprises a first sample network resource and a positive similar network resource of the first sample network resource, each first negative sample data comprises a second sample network resource and a negative similar network resource of the second sample network resource, the first sample network resource and the second sample network resource are any one of the plurality of sample network resources, the positive similar network resource refers to a similar network resource which is accepted by a user after being recommended to the user in k similar network resources of the first sample network resource, and the negative similar network resource refers to a similar network resource which is not accepted by the user after being recommended to the user in k similar network resources of the second sample network resource; each second positive sample data comprises an explicit characteristic and an implicit characteristic of the first sample network resource, and an explicit characteristic and an implicit characteristic of a positive similar network resource of the first sample network resource, and each second negative sample data comprises an explicit characteristic and an implicit characteristic of the second sample network resource, and an explicit characteristic and an implicit characteristic of a negative similar network resource of the second sample network resource;

and the training module is used for training the association model to be trained on the basis of the plurality of second positive sample data and the plurality of second negative sample data to obtain the association model.

Optionally, the second determining module includes:

a first determining unit, configured to determine, for any sample network resource in the multiple sample network resources, a similarity between each network resource in the second network resource library and the any sample network resource according to a first word embedding vector of the any sample network resource in a browsing behavior space and a first word embedding vector of each network resource in the second network resource library in the browsing behavior space; determining k similar network resources of any sample network resource from the second network resource library based on the similarity of each network resource in the second network resource library to the any sample network resource;

a second determining unit, configured to determine, for any sample network resource in the multiple sample network resources, a similarity between each network resource in the second network resource library and the any sample network resource according to a second word embedding vector of the any sample network resource in a resource content space and a second word embedding vector of each network resource in the second network resource library in the resource content space; determining k similar network resources of the any sample network resource from the second network resource library based on the similarity of each network resource in the second network resource library to the any sample network resource.

Optionally, the fourth determining module includes:

a first extracting unit, configured to, for any first positive sample data in the plurality of first positive sample data, respectively extract a first sample network resource included in the any first positive sample data and an explicit feature of a positive similar network resource of the first sample network resource;

a third determining unit, configured to determine, based on the explicit feature and the implicit feature of the first sample network resource and the explicit feature and the implicit feature of the positive-phase similar network resource of the first sample network resource, second positive sample data corresponding to any one of the first positive sample data;

a second extracting unit, configured to, for any one of the plurality of first negative sample data, respectively extract dominant features of a second sample network resource included in the any one of the first negative sample data and a negative similar network resource of the second sample network resource;

a fourth determining unit, configured to determine second negative sample data corresponding to any one of the first negative sample data based on the explicit characteristic and the implicit characteristic of the second sample network resource and the explicit characteristic and the implicit characteristic of the negative similar network resource of the second sample network resource.

Optionally, the third determining unit is configured to:

determining the dominant characteristic and the recessive characteristic of the first sample network resource and the dominant characteristic and the recessive characteristic of the positive similar network resource of the first sample network resource as second positive sample data corresponding to any one first positive sample data;

Optionally, the fourth determining unit is configured to:

determining the dominant characteristic and the recessive characteristic of the second sample network resource and the dominant characteristic and the recessive characteristic of the negative similar network resource of the second sample network resource as second negative sample data corresponding to any one of the first negative sample data;

In another aspect, a server is provided, where the terminal includes a processor and a memory, where the memory stores at least one instruction, at least one program, a set of codes, or a set of instructions, and the at least one instruction, the at least one program, the set of codes, or the set of instructions is loaded and executed by the processor to implement the method for recommending network resources.

In another aspect, a computer-readable storage medium is provided, in which at least one instruction, at least one program, a set of codes, or a set of instructions is stored, and the at least one instruction, the at least one program, the set of codes, or the set of instructions is loaded and executed by a processor to implement the method for recommending network resources described above.

In another aspect, a computer program product containing instructions is provided, which when run on a computer, causes the computer to execute the method for recommending network resources described above.

The technical scheme provided by the application can at least bring the following beneficial effects:

in the embodiment of the application, for the reference network resource browsed by a user, the dominant characteristic and the recessive characteristic of the reference network resource can be obtained, the similarity between each network resource in the first network resource library and the reference network resource is determined through a correlation model based on the dominant characteristic and the recessive characteristic of the reference network resource and the dominant characteristic and the recessive characteristic of each network resource in the first network resource library, and then resource recommendation is performed for the user based on the similarity between each network resource in the first network resource library and the reference network resource. The explicit characteristics comprise attribute characteristics and statistical characteristics, and the implicit characteristics refer to vector representation mapped to the implicit space, so that the similarity of any two sample network resources can be evaluated by combining the attribute characteristics, the statistical characteristics and the implicit characteristics of the network resources, and resource recommendation is performed based on the similarity of the network resources, so that the evaluation of the similarity is more comprehensive and accurate, the accuracy of determining the similarity of the network resources is improved, and the recommendation accuracy is further improved.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

FIG. 1 is a schematic illustration of an implementation environment provided by an embodiment of the present application;

FIG. 2 is a flowchart of a method for training a correlation model according to an embodiment of the present disclosure;

fig. 3 is a flowchart of a method for recommending network resources according to an embodiment of the present application;

fig. 4 is a schematic structural diagram of a recommendation apparatus for network resources according to an embodiment of the present application;

fig. 5 is a schematic structural diagram of a server according to an embodiment of the present application.

Detailed Description

To make the objects, technical solutions and advantages of the present application more clear, embodiments of the present application will be described in further detail below with reference to the accompanying drawings.

Before explaining a recommendation method for network resources provided in the embodiment of the present application in detail, an application scenario provided in the embodiment of the present application is introduced.

The method for recommending the network resources is applied to a scene of recommending the similar network resources for the user based on the similarity of the network resources browsed by the user, wherein the network resources can be songs, videos or news and the like. For example, when a user listens to a song through music software, the music software may recommend the song to the user through the method provided by the embodiment of the present application. Or when the user watches videos through the video software, the video software can recommend videos to the user through the method provided by the embodiment of the application. Or, when the user reads news through news software, the news software may perform news recommendation and the like for the user through the method provided by the embodiment of the application. Of course, the recommendation method for network resources provided in the embodiment of the present application may also be applied to other scenarios, and the embodiment of the present application does not limit this.

It should be noted that the terms "first" and "second", etc. in this application are used for distinguishing different objects, and are not used for describing a specific order.

The following describes an implementation environment provided by embodiments of the present application.

Fig. 1 is a schematic diagram of an implementation environment provided by an embodiment of the present application, and as shown in fig. 1, the implementation environment includes a terminal 10 and a server 20, and the terminal 10 and the server 20 may communicate through a wired network or a wireless network. The terminal 10 may be a computer, a mobile phone, or a tablet computer, etc. The terminal 10 is installed with resource software, which is used for providing network resources for users, and specifically may be music software, video software, or news software. The server 20 is a background server of the resource software, and can recommend resources to the user according to the method provided in the embodiment of the present application in the process that the user browses the network resources by using the resource software.

It should be noted that, in the embodiment of the present application, the similarity of the network resources may be determined by using the association model, and then resource recommendation is performed for the user according to the similarity of the network resources. The association model is used for determining the similarity of any two network resources according to the explicit characteristics and the implicit characteristics of the two network resources. Before determining the similarity of the network resources by using the association model, the association model is obtained by performing model training by using sample data. Next, a training process of the association model will be described.

Fig. 2 is a flowchart of a method for training a correlation model according to an embodiment of the present application, where the method may be applied to the server shown in fig. 1, as shown in fig. 2, and the method includes the following steps:

step 201: and acquiring implicit characteristics of a plurality of sample network resources, wherein the sample network resources are network resources browsed by a sample user.

The plurality of sample network resources are network resources selected for training the association model, and are network resources browsed by the sample user, such as resources of songs, videos, or news browsed by the user. The sample user is the user selected for training the association model.

For example, the sample network resource may be a song played or a song collected by the sample user, and may be obtained from a song play list or a song collection list of the sample user. For another example, the sample network resource may be a video that the sample user has viewed or collected, and may be obtained from a viewing record or a collection list of the sample user. For another example, the sample network resource may be news read or collected by the sample user, and may be obtained from a reading record or a favorite list of the sample user.

For each of the plurality of network resources, an implicit characteristic of each network resource may be obtained. Implicit features refer to a vector representation that maps to an implicit space. For example, the vector representation may be a word embedding (embedding) vector, which refers to a low-dimensional vector used to represent the sample network resource in the implicit space.

As one example, the implicit features include at least a word embedding vector in the browsing behavior space. Further, the implicit features may also include word embedding vectors in the browsing behavior space and word embedding vectors in the resource content space.

In some embodiments, the operation of obtaining implicit characteristics of the sample network resource may include the following two implementations:

the first implementation mode comprises the following steps: and mapping the sample network resources into a browsing behavior space to obtain a first word embedding vector of the sample network resources, and determining the first word embedding vector of the sample network resources as the implicit characteristic of the sample network resources.

And mapping the sample network resource to a browsing behavior space to obtain a first word embedding vector of the sample network resource in the browsing behavior space. The first word embedding vector of the sample network resource in the browsing behavior space may indicate the browsing behavior of the user on the sample network resource, such as how many users browse the sample network resource or which users browse the sample network resource. The first word embedding vector may be an N-dimensional vector, illustratively, N is 100.

For example, if the sample network resource is a sample song, the sample song may be mapped to a user playing behavior space to obtain a word embedding vector in the playing behavior space, and/or the sample song may be mapped to a user collection behavior space to obtain a word embedding vector in the collection behavior space. The word embedding vector on the play behavior space may indicate how many users played the sample song, or which users played the sample song. Likewise, the word embedding vector on the collection behavior space may indicate which users collected the sample song, or which users collected the sample song.

By using the first word embedded vector for mapping the sample network resources to the browsing behavior space as the implicit characteristic of the sample network resources, the similarity between the network resources can be conveniently evaluated in subsequent combination with the similarity of the network resources on the aspect of the user browsing behavior.

As one example, a first word embedding vector for a sample network resource may be computed via the item2vec algorithm. For example, if the network resource is a song, a plurality of sample songs may be obtained based on the song play stream and the song collection stream of the sample user, and used as sample data for model training. If the sample song is a song in the play song list of the sample user, a first word embedding vector of the sample song can be obtained through the item2vec algorithm and is recorded as song _ play _ i. If the sample song is a song in the song collection list of the sample user, a first word embedding vector of the sample song can be obtained through the item2vec algorithm and is recorded as song _ favor _ i.

The second implementation mode comprises the following steps: mapping the sample network resources to a browsing behavior space to obtain a first word embedding vector of the sample network resources; mapping the sample network resources into a resource content space to obtain a second word embedding vector of the sample network resources; and determining the first word embedding vector and the second word embedding vector of the sample network resource as the implicit characteristics of the sample network resource.

That is, in the second implementation, the implicit characteristic of the sample web resource includes both the first word embedding vector in the browsing behavior space and the second word embedding vector in the resource content space. The first word embedding vector of the sample network resource on the browsing behavior space can indicate the browsing behavior of the user on the sample network resource, and the second word embedding vector of the sample network resource on the resource content space can indicate the specific content of the sample network resource, so that the similarity between the network resources can be conveniently evaluated by subsequently combining the similarity of the network resources in the two aspects of the browsing behavior of the user and the resource content.

For example, the second word embedding vector may be an N-dimensional vector, that is, the resource content of the sample network resource may be represented by an N-dimensional embedding vector. Illustratively, N is 100.

As an example, mapping the sample network resource into the resource content space, and obtaining the second word embedding vector of the sample network resource may include: and extracting the resource content of the sample network resource, and calculating to obtain a second word embedding vector of the sample network resource through an item2vec algorithm based on the resource content of the sample network resource. Or extracting the resource content of the sample network resource, and determining a second word embedding vector of the sample network resource through a word embedding vector model based on the resource content of the sample network resource. The word embedding vector model is used for determining a second word embedding vector of any network resource based on the resource content of any network resource, and the word embedding vector model can be obtained by training based on the calculation result of the item2vec algorithm.

If the sample network resource is a sample song, the resource content of the sample song can be a song spectrum or a spectrum image formed by the song spectrum; if the sample network resource is a sample video, the resource content of the sample video can be a video image; if the sample network resource is sample news, the resource content of the sample news may be news content.

For example, if the sample network resource is a sample song, a song spectrum of the sample song may be extracted, a spectrum image of the sample song may be determined based on the song spectrum of the sample song, and a second word embedding vector of the sample song may be determined by the word embedding vector model based on the spectrum image of the sample song.

Step 202: and determining k similar network resources of each sample network resource from the second network resource library according to the implicit characteristics of each sample network resource in the plurality of sample network resources.

The k is an integer greater than 1, and a specific value of k may be preset, specifically may be set by a server default, or may be set by a user. For example, k can be 2, 5, or 10, etc.

The second network resource library includes a large number of network resources available for recommendation, and may be a network resource library of resource software, such as a song library of song software or a video library of video software. The similar network resource refers to a network resource similar to the sample network resource, such as a song similar to the sample song, or a video similar to the sample video. In the embodiment of the application, k similar network resources can be determined according to the implicit characteristic of the sample network resource, that is, the determined k similar network resources refer to k network resources in which the implicit characteristic in the second network resource library is similar to the implicit characteristic of the sample network resource.

As an example, determining k similar network resources for each sample network resource from the first network resource pool according to the implicit characteristic of each sample network resource in the plurality of sample network resources may include the following two implementations:

the first implementation mode comprises the following steps: for any sample network resource in the multiple sample network resources, determining the similarity between each network resource in the second network resource library and any sample network resource according to the first word embedding vector of the any sample network resource in the browsing behavior space and the first word embedding vector of each network resource in the second network resource library in the browsing behavior space; and determining k similar network resources of any sample network resource from the second network resource library based on the similarity of each network resource in the second network resource library and any sample network resource.

In some embodiments, the similarity between each network resource in the second network resource pool and the any sample network resource may be calculated according to the first word embedding vector of the any sample network resource in the browsing behavior space and the first word embedding vector of each network resource in the second network resource pool in the browsing behavior space through an ANN (Approximate Nearest Neighbors) algorithm. The ANN algorithm is an algorithm for solving a similarity search problem, and may be, for example, an approximation Nearest neighbor (Annoy) algorithm or a Product Quantization (PQ) algorithm.

In some embodiments, the operation of determining k similar network resources of any sample network resource from the second pool of network resources based on the similarity of each network resource in the second pool of network resources to any sample network resource comprises: based on the similarity between each network resource in the second network resource library and any sample network resource, selecting k network resources in the second network resource library which are ranked in the front as k similar network resources of any sample network resource according to the sequence of similarity from large to small of any network resource.

The second implementation mode comprises the following steps: for any sample network resource in the multiple sample network resources, determining the similarity of each network resource in the second network resource library and any sample network resource according to the second word embedding vector of any sample network resource on the resource content space and the second word embedding vector of each network resource in the second network resource library on the resource content space; and determining k similar network resources of any sample network resource from the second network resource library based on the similarity of each network resource in the second network resource library and any sample network resource.

In some embodiments, the similarity between each network resource in the second network resource pool and any sample network resource may be calculated by an ANN algorithm according to the second word embedding vector of any sample network resource in the resource content space and the second word embedding vector of each network resource in the second network resource pool in the resource content space. For example, the ANN algorithm may be an annoy algorithm or a PQ algorithm, etc.

The third implementation mode comprises the following steps: for any sample network resource in the multiple sample network resources, determining the similarity of each network resource in the second network resource library and any sample network resource according to a first word embedding vector of any sample network resource in a browsing behavior space and a second word embedding vector of any sample network resource in a resource content space, and the first word embedding vector of each network resource in the second network resource library in the browsing behavior space and the second word embedding vector of each network resource in the resource content space; and determining k similar network resources of any sample network resource from the second network resource library based on the similarity of each network resource in the second network resource library and any sample network resource.

In some embodiments, the similarity between each network resource in the second network resource pool and any sample network resource can be calculated by an ANN algorithm according to a first word embedding vector of any sample network resource in the browsing behavior space and a second word embedding vector of any sample network resource in the resource content space, and a first word embedding vector of each network resource in the second network resource pool in the browsing behavior space and a second word embedding vector of each network resource in the resource content space. For example, the ANN algorithm may be an Annoy algorithm or a PQ algorithm.

Step 203: determining a plurality of first positive sample data and a plurality of first negative sample data based on the associated recommendation results of the k similar network resources of each of the plurality of sample network resources.

Each first positive sample data comprises a first sample network resource and a positive similar network resource of the first sample network resource, the first sample network resource is any one of a plurality of sample network resources, and the positive similar network resource refers to a similar network resource which is accepted by a user after being recommended to the user, such as a similar network resource which is clicked by the user after being recommended to the user, in k similar network resources of the first sample network resource.

Each first negative sample data includes a second sample network resource and a negative similar network resource of the second sample network resource, the second sample network resource is any one of the multiple sample network resources, and the negative similar network resource refers to a similar network resource that is not accepted by the user after being recommended to the user, among the k similar network resources of the second sample network resource, such as a similar network resource that is not clicked by the user after being recommended to the user.

After k similar network resources of each sample network resource are obtained through calculation, the k similar network resources can be recommended to the user who browses the sample network resources, and the recommendation mode is called as associated recommendation. The associated recommendation result is used for indicating whether the similar network resources recommended to the user are accepted by the user, for example, whether the similar network resources recommended to the user are clicked by the user.

Then, positive and negative sample data can be determined based on the associated recommendation result, for example, the positive and negative sample data can be determined according to the click data of the associated recommendation. If the recommendation of a similar network resource of a certain sample network resource is accepted by a user, for example, the recommendation is clicked by the user after being exposed to the user, a group of network resource pairs consisting of the sample network resource and the similar network resource can be used as a first positive sample data; if the recommendation of a similar network resource of a certain sample network resource is not accepted by the user, for example, the recommendation is not clicked by the user after being exposed to the user, a group of network resource pairs consisting of the sample network resource and the similar network resource can be used as a first negative sample data.

For example, if the sample network resources are sample songs, k similar songs of each sample song may be online associated and recommended to the user, after the sample songs are recommended to the user, the similar songs clicked by the user are positive similar network resources, the positive similar songs and the corresponding sample songs form a first positive sample data, the similar songs not clicked by the user are negative similar network resources, and the negative similar songs and the corresponding sample songs form a first negative sample data.

For example, the first positive sample data may be represented as < (source _ song, target _ song), 1>, and the first negative positive sample data may be represented as < (source _ song, target _ song), 0 >. Source _ song represents a sample song, target _ song represents a similar song to the sample song, 1 represents a recommended target _ song clicked by the user, and 0 represents not clicked.

It should be noted that this step may be implemented to obtain which similar network resources of the k similar network resources obtained from the second network resource library are more easily browsed by the user, for example, which songs of the k similar songs of any sample song are more easily enjoyed by the sample user.

Step 204: a plurality of second positive sample data and a plurality of second negative sample data are determined based on the plurality of first positive sample data and the plurality of first negative sample data.

Each second positive sample data comprises the dominant characteristic and the recessive characteristic of the first sample network resource, and the dominant characteristic and the recessive characteristic of a positive similar network resource of the first sample network resource, and is determined by one first positive sample data. Each second negative sample data comprises the dominant characteristic and the recessive characteristic of the second sample network resource and the dominant characteristic and the recessive characteristic of a negative similar network resource of the second sample network resource, and is determined by one first negative sample data.

Wherein the explicit characteristics include attribute characteristics and statistical characteristics. The attribute feature is used to indicate the attributes of the sample network resource itself, such as title, label, name, publisher, or time of publication. The statistical characteristics may be obtained by performing statistics on the behavior of the user browsing the sample network resource, and may include browsing amount, collection amount, comment amount, or approval amount of the sample network resource.

For example, when the sample network resource is a sample song, the attribute characteristics of the sample song may include at least one of artist, genre to which the sample song belongs, album to which the sample song belongs, release time, and region in which the sample song is located, and the statistical characteristics of the sample song may include at least one of play amount, collection amount, comment amount, and collection amount.

Further, each second positive sample data may further include a comparison feature between the statistical feature of the first sample network resource and the statistical feature of one positive similar network resource of the first sample network resource, and each second negative sample data may also include a comparison feature between the statistical feature of the second sample network resource and the statistical feature of one negative similar network resource of the second sample network resource.

Specifically, determining the plurality of second positive sample data and the plurality of second negative sample data based on the plurality of first positive sample data and the plurality of first negative sample data may include the following two cases:

in the first case: for any first positive sample data in the plurality of first positive sample data, respectively extracting the dominant characteristic and the recessive characteristic of the first sample network resource and the positive similar network resource of the first sample network resource included in the any first positive sample data; and determining second positive sample data corresponding to any first positive sample data based on the dominant characteristic and the recessive characteristic of the first sample network resource and the dominant characteristic and the recessive characteristic of the positive similar network resource of the first sample network resource.

For example, the dominant characteristic and the recessive characteristic of the first sample network resource, and the dominant characteristic and the recessive characteristic of the positive similar network resource of the first sample network resource may be directly determined as the second positive sample data corresponding to any first positive sample data.

Or, the statistical characteristics of the first sample network resource and the statistical characteristics of the positive similar network resource of the first sample network resource may be compared to obtain the comparison characteristics between the positive similar network resource of the first sample network resource and the first sample network resource; and determining the dominant characteristic and the recessive characteristic of the first sample network resource, the dominant characteristic and the recessive characteristic of the positive similar network resource of the first sample network resource and the contrast characteristic between the positive similar network resource of the first sample network resource and the first sample network resource as second positive sample data corresponding to any first positive sample data.

The comparison characteristic between the positive similar network resource of the first sample network resource and the first sample network resource may be a ratio between statistical characteristics of the two network resources, such as a ratio of browsing amount, collection amount, comment amount, or praise amount.

In the second case: for any first negative sample data in the plurality of first negative sample data, respectively extracting explicit characteristics of a second sample network resource and a negative similar network resource of the second sample network resource included in the any first negative sample data; and determining second negative sample data corresponding to any first negative sample data based on the dominant characteristic and the recessive characteristic of the second sample network resource and the dominant characteristic and the recessive characteristic of the negative similar network resource of the second sample network resource.

For example, the explicit characteristic and the implicit characteristic of the second sample network resource and the explicit characteristic and the implicit characteristic of the negative similar network resource of the second sample network resource may be determined as the second negative sample data corresponding to the any first negative sample data.

Or, the statistical characteristics of the second sample network resources and the statistical characteristics of the positively similar network resources of the second sample network resources may be compared to obtain the comparative characteristics between the negatively similar network resources of the second sample network resources and the second sample network resources; and determining the dominant characteristic and the recessive characteristic of the second sample network resource, the dominant characteristic and the recessive characteristic of the negative similar network resource of the second sample network resource and the comparison characteristic between the negative similar network resource of the second sample network resource and the second sample network resource as second negative sample data corresponding to any first negative sample data.

The comparison characteristic between the negative similar network resource of the second sample network resource and the second sample network resource may be a ratio between statistical characteristics of the two network resources, such as a ratio of browsing amount, collection amount, comment amount, or like amount.

The implementation of the step can obtain more accurate similarity through multi-aspect comparison processing by obtaining the explicit characteristics and the implicit characteristics of the sample network resources, the positive similar network resources and the negative similar network resources, the explicit characteristics and the implicit characteristics of the comparison characteristics of the sample network resources and the positive similar network resources, and the explicit characteristics and the implicit characteristics of the comparison characteristics of the sample network resources and the negative similar network resources, so that the accuracy of recommending the network resources to the user by the correlation model is improved.

Step 205: and training the association model to be trained based on the plurality of second positive sample data and the plurality of second negative sample data to obtain the association model.

The correlation model is a Neural Network model, and may specifically be a CNN (Convolutional Neural Networks), RNN (Recurrent Neural Networks), or LSTM (Long Short-Term Memory Network) model, which is not limited in this embodiment of the present application. The association model to be trained is trained based on a plurality of second positive sample data and a plurality of second negative sample data, so that the association model which can determine the similarity between any two network resources according to the dominant characteristic and the recessive characteristic of any two network resources can be obtained.

The method has the advantages that the dominant characteristic and the recessive characteristic of a plurality of pairs of similar network resources, or the dominant characteristic and the recessive characteristic of a plurality of pairs of similar network resources, and the contrast characteristic of the dominant characteristic and the recessive characteristic of the plurality of pairs of similar network resources are used as training samples for model training, so that the model can be integrated with a plurality of aspects of similar network resources to learn and evaluate the similarity between any two network resources in the training process, the evaluation of the similarity is more comprehensive and accurate, namely, the similarity between any two network resources can be more accurately determined through the correlation model.

In some embodiments, a plurality of second positive sample data and a plurality of second negative sample data may be used as inputs of the association model to be trained, the prediction similarity of each sample data is output through the association model to be trained, the prediction similarity of each sample data is compared with the true similarity, and based on a comparison result, a random gradient descent method is adopted to adjust model parameters in the association model to be trained, so as to obtain the association model.

For example, the association model to be trained may be obtained by training the association model to be trained based on the plurality of second positive sample data and the plurality of second negative sample data with change-loss as a loss function.

After the association model is obtained through training, the association model can be deployed on the line, and ranking or recommendation services of similar network resources with higher quality are provided for users on the line.

In the embodiment of the application, implicit characteristics of a plurality of sample network resources can be obtained, k similar network resources of each sample network resource are determined from a second network resource library according to the implicit characteristics of each sample network resource in the plurality of sample network resources, a plurality of second positive sample data and a plurality of second negative sample data are determined based on the association recommendation result of the k similar network resources of each sample network resource in the plurality of sample network resources, and then an association model to be trained is trained based on the plurality of second positive sample data and the plurality of second negative sample data The method and the device ensure that the trained association model can evaluate the similarity of the network resources more comprehensively and accurately, and further can determine the similarity between any two network resources more accurately.

It should be noted that after the association model is trained, the similarity between the network resources can be determined by using the association model, and then resource recommendation is performed according to the similarity of the network resources. Next, a recommendation process of network resources provided by the embodiment of the present application is described in detail.

Fig. 3 is a flowchart of a method for recommending network resources according to an embodiment of the present application, where the method may be applied to the server shown in fig. 1. Referring to fig. 3, the method includes the following steps.

Step 301: acquiring an explicit characteristic and an implicit characteristic of a reference network resource, wherein the explicit characteristic comprises an attribute characteristic and a statistical characteristic, the implicit characteristic refers to vector representation mapped in an implicit space, and the reference network resource is a network resource browsed by a user.

The reference network resource may be any network resource that the user browses, such as a song that the user has played or collected, a video that the user has watched or collected, news that the user has read or collected, and the like. For example, if the reference network resource is a reference song, the reference song may be a song in a user song playlist or song collection list.

The attribute feature is used to indicate the self attribute of the reference network resource, such as the title, tag, name, publisher or publication time. The statistical characteristics can be obtained by counting the behaviors of the user browsing the reference network resources, and may include browsing amount, collection amount, comment amount, or like amount of the reference network resources.

For example, when the reference network resource is a reference song, the attribute characteristics of the reference song may include at least one of an artist, a genre to which the reference song belongs, an album to which the reference song belongs, a release time, and a region in which the artist is located, and the statistical characteristics of the reference song may include at least one of a play amount, a collection amount, a comment amount, and a collection amount.

As one example, the implicit characteristic of the reference network resource includes at least a word embedding vector of the reference network resource in the browsing behavior space. Further, the implicit characteristic of the reference network resource may further include a word embedding vector of the reference network resource in a browsing behavior space and a word embedding vector of the reference network resource in a resource content space.

In some embodiments, the operation of obtaining the implicit characteristic of the reference network resource may include the following two implementations:

the first implementation mode comprises the following steps: and mapping the reference network resource to a browsing behavior space to obtain a first word embedded vector of the reference network resource, and determining the first word embedded vector of the reference network resource as a recessive characteristic of the reference network resource.

By using the first word embedded vector for mapping the reference network resources to the browsing behavior space as the implicit characteristic of the reference network resources, the similarity between the network resources can be conveniently evaluated in subsequent combination with the similarity of the network resources in the aspect of the user browsing behavior.

The second implementation mode comprises the following steps: mapping the reference network resource to a browsing behavior space to obtain a first word embedded vector of the reference network resource; mapping the reference network resource to a resource content space to obtain a second word embedded vector of the reference network resource; and determining the first word embedding vector and the second word embedding vector of the reference network resource as the implicit characteristics of the reference network resource.

The first word embedded vector of the reference network resource in the browsing behavior space may indicate the browsing behavior of the user on the reference network resource, and the second word embedded vector of the reference network resource in the resource content space may indicate the specific content of the reference network resource. By taking the first word embedding vector of the reference network resource in the browsing behavior space and the second word embedding vector in the resource content space as the implicit characteristics of the reference network resource, the similarity between the network resources can be conveniently evaluated in the subsequent combination of the similarity of the network resources in the two aspects of the user browsing behavior and the resource content.

It should be noted that the implementation manner of mapping the reference network resource to the browsing behavior space to obtain the first word embedded vector of the reference network resource, and mapping the reference network resource to the resource content space to obtain the second word embedded vector of the reference network resource is the same as the processing manner of the sample network resource in the embodiment of fig. 2, and details of the embodiment of the present application are not repeated here.

Step 302: and determining the similarity between each network resource in the first network resource library and the reference network resource through the association model based on the dominant characteristic and the recessive characteristic of the reference network resource and the dominant characteristic and the recessive characteristic of each network resource in the first network resource library.

The first network resource library includes a large number of network resources available for recommendation, and may be a network resource library of resource software, such as a song library of song software or a video library of video software. The first network resource pool may be the same as or different from the second network resource pool used in the model training process of the embodiment in fig. 2, which is not limited in this application.

The association model is used to determine the similarity between any two network resources, and may be obtained by training through the model training method in the embodiment of fig. 2. The similarity of any two network resources is used to indicate the degree of similarity between the any two network resources.

In some embodiments, the operation of determining the similarity between each network resource in the first network resource pool and the reference network resource through the association model based on the explicit characteristics and the implicit characteristics of the reference network resource and the explicit characteristics and the implicit characteristics of each network resource in the first network resource pool may include the following two implementation manners:

the first implementation mode comprises the following steps: for a first network resource in a first network resource library, taking the dominant characteristic and the recessive characteristic of a reference network resource and the dominant characteristic and the recessive characteristic of the first network resource as the input of the association model, and determining the similarity between the first network resource and the reference network resource through the association model, wherein the first network resource is any network resource in the first network resource library.

That is, for any network resource in the first network resource pool, the explicit characteristic and the implicit characteristic of the any network resource, and the explicit characteristic and the implicit characteristic of the reference network resource may be input into the association model, and the input data is processed by the association model, so that the similarity between the any network resource and the reference network resource may be output.

Taking network resources as songs as an example, a background server of song software can extract the dominant characteristic and the recessive characteristic of a reference song played or collected by a user, take the dominant characteristic and the recessive characteristic of any song in a song library and the dominant characteristic and the recessive characteristic of the reference song as the input of an association model, and output the similarity between any song and the reference song through the association model.

The second implementation mode comprises the following steps: comparing the statistical characteristics of the reference network resources with the statistical characteristics of each network resource in the first network resource library to obtain the comparison characteristics between each network resource in the first network resource library and the reference network resources; for a first network resource in a first network resource library, taking the dominant characteristic and the recessive characteristic of a reference network resource, and the dominant characteristic, the recessive characteristic and the comparison characteristic between the first network resource and the reference network resource as the input of the association model, and determining the similarity between the first network resource and the reference network resource through the association model, wherein the first network resource is any one network resource in the first network resource library.

The comparison characteristic between each network resource in the first network resource library and the reference network resource may be a ratio between the statistical characteristic of each network resource in the first network resource library and the statistical characteristic of the reference network resource, for example, the comparison characteristic may be a ratio of browsing amount, a ratio of collection amount, or a ratio of comment amount.

That is, for any network resource in the first network resource pool, the dominant characteristic and the recessive characteristic of the network resource, the dominant characteristic and the recessive characteristic of the reference network resource, and the comparison characteristic of the two network resources may be input into the association model, and the similarity between the network resource and the reference network resource may be output by processing the input data through the association model.

Taking network resources as songs as an example, a background server of song software can extract the dominant characteristic and the recessive characteristic of a reference song played or collected by a user, take the dominant characteristic and the recessive characteristic of any song in a song library, the dominant characteristic and the recessive characteristic of the reference song and the contrast characteristic between any song and the reference song as the input of an association model, and output the similarity between any song and the reference song through the association model.

In the second implementation manner, the similarity of any two network resources is evaluated by combining the comparison characteristics between the two network resources on the basis of the dominant characteristics and the recessive characteristics of the two network resources, so that the evaluation of the similarity of the network resources can be more comprehensive and accurate.

Step 303: and recommending the resources for the user based on the similarity between each network resource in the first network resource library and the reference network resource.

In some embodiments, the resource recommendation for the user based on the similarity between each network resource in the first network resource pool and the reference network resource comprises: sequencing the network resources in the first network resource library according to the sequence of similarity between the network resources and the reference network resources from large to small; and acquiring the n network resources ranked at the top from the ranking result, and recommending the n network resources ranked at the top to the user.

N is a positive integer, and a specific value of n may be set by a default of the server or by a user, which is not limited in the embodiment of the present application. For example, n can be 1, 2, or 5, and the like.

For example, the background server of the song software may sort all the online songs in an order from a large similarity to a small similarity, and select n top-ranked songs from the sorting results to recommend to the user.

In other embodiments, resource recommendation may be performed for the user in other manners based on a similarity between each network resource in the first network resource library and the reference network resource, which is not limited in this embodiment of the present application.

It should be noted that, for a new network resource that is a newly released network resource in the network resource software, since the attribute of the new network resource is known, but the statistical characteristic and the characteristic in the browsing behavior space are unknown, for the new network resource, if the similarity between the new network resource and other network resources is to be calculated, the attribute characteristic of the new network resource may also be obtained, and then the similarity between the new network resource and other network resources is determined through the association model based on the attribute characteristic of the new network resource and the attribute characteristic of the other network resources. The other network resource may be a new network resource, or may be an old network resource that has been released in the network resource software for a certain period of time.

In another embodiment, for a new network resource, if the similarity between the new network resource and other network resources is to be calculated, the attribute characteristics of the new network resource may also be obtained, and then the similarity between the new network resource and other network resources is determined through the association model based on the attribute characteristics of the new network resource and the explicit characteristics and the implicit characteristics of the other network resources. Wherein, the other network resources are old network resources which have been released in the network resource software for a period of time.

For example, for a new song newly released on song software, the attribute characteristics of the new song can be obtained, and then the similarity between the new song and other songs is determined through the association model based on the attribute characteristics of the new song and the attribute characteristics of other songs. Or determining the similarity between the new song and other songs through the association model based on the attribute characteristics of the new song and the explicit characteristics and the implicit characteristics of other songs.

Fig. 4 is a schematic structural diagram of a recommendation apparatus for network resources according to an embodiment of the present application, where the recommendation apparatus for network resources may be implemented by software, hardware, or a combination of the two as part or all of a server, and the server may be the server shown in fig. 1. Referring to fig. 4, the apparatus includes: a first obtaining module 401, a first determining module 402 and a recommending module 403.

A first obtaining module 401, configured to obtain an explicit feature and an implicit feature of a reference network resource, where the explicit feature includes an attribute feature and a statistical feature, the implicit feature refers to a vector representation mapped to an implicit space, and the reference network resource is a network resource browsed by a user;

a first determining module 402, configured to determine, based on the explicit characteristic and the implicit characteristic of the reference network resource and the explicit characteristic and the implicit characteristic of each network resource in the first network resource library, a similarity between each network resource in the first network resource library and the reference network resource through an association model, where the association model is used to determine a similarity between any two network resources;

a recommending module 403, configured to recommend resources for the user based on a similarity between each network resource in the first network resource library and the reference network resource.

Optionally, the obtaining module 401 is configured to:

Optionally, the first determining module 402 is configured to:

for a first network resource in the first network resource library, taking the dominant characteristic and the recessive characteristic of the reference network resource and the dominant characteristic and the recessive characteristic of the first network resource as the input of the association model, and determining the similarity between the first network resource and the reference network resource through the association model, wherein the first network resource is any one of the network resources in the first network resource library.

Optionally, the first determining module 402 is configured to:

comparing the statistical characteristics of the reference network resource with the statistical characteristics of each network resource in the first network resource library to obtain the comparison characteristics between each network resource in the first network resource library and the reference network resource;

for a first network resource in the first network resource library, taking the dominant characteristic and the recessive characteristic of the reference network resource, and the dominant characteristic, the recessive characteristic and the comparison characteristic between the first network resource and the reference network resource as the input of the association model, and determining the similarity between the first network resource and the reference network resource through the association model, wherein the first network resource is any one network resource in the first network resource library.

Optionally, the recommending module 403 is configured to:

sequencing the network resources in the first network resource library according to the sequence of similarity between the network resources and the reference network resource from large to small;

Optionally, the apparatus further comprises:

a second determining module, configured to determine, according to the implicit characteristic of each sample network resource in the multiple sample network resources, k similar network resources of each sample network resource from a second network resource library, where k is an integer greater than 1;

a third determining module, configured to determine, based on the association recommendation result of the k similar network resources of each sample network resource in the plurality of sample network resources, a plurality of first positive sample data and a plurality of first negative sample data;

each first positive sample data comprises a first sample network resource and a positive similar network resource of the first sample network resource, each first negative sample data comprises a second sample network resource and a negative similar network resource of the second sample network resource, the first sample network resource and the second sample network resource are any one of the sample network resources, the positive similar network resource refers to a similar network resource which is accepted by a user after being recommended to the user in k similar network resources of the first sample network resource, and the negative similar network resource refers to a similar network resource which is not accepted by the user after being recommended to the user in k similar network resources of the second sample network resource; each second positive sample data comprises the dominant characteristic and the recessive characteristic of the first sample network resource, and the dominant characteristic and the recessive characteristic of a positively similar network resource of the first sample network resource, and each second negative sample data comprises the dominant characteristic and the recessive characteristic of the second sample network resource, and the dominant characteristic and the recessive characteristic of a negatively similar network resource of the second sample network resource;

Optionally, the second determining module includes:

a first determining unit, configured to determine, for any sample network resource in the multiple sample network resources, a similarity between each network resource in the second network resource library and the any sample network resource according to a first word embedding vector of the any sample network resource in a browsing behavior space and a first word embedding vector of each network resource in the second network resource library in the browsing behavior space; determining k similar network resources of the any sample network resource from the second network resource library based on the similarity of each network resource in the second network resource library to the any sample network resource;

a second determining unit, configured to determine, for any sample network resource in the multiple sample network resources, a similarity between each network resource in the second network resource library and the any sample network resource according to a second word embedding vector of the any sample network resource in a resource content space and a second word embedding vector of each network resource in the second network resource library in the resource content space; and determining k similar network resources of any sample network resource from the second network resource library based on the similarity between each network resource in the second network resource library and any sample network resource.

Optionally, the fourth determining module includes:

a first extracting unit, configured to, for any first positive sample data in the multiple first positive sample data, respectively extract a first sample network resource included in the any first positive sample data and an explicit feature of a positive similar network resource of the first sample network resource;

and a fourth determining unit, configured to determine second negative sample data corresponding to any one of the first negative sample data based on the explicit characteristic and the implicit characteristic of the second sample network resource and the explicit characteristic and the implicit characteristic of the negative similar network resource of the second sample network resource.

Optionally, the third determining unit is configured to:

comparing the statistical characteristics of the first sample network resource with the statistical characteristics of the positive similar network resource of the first sample network resource to obtain the comparison characteristics between the positive similar network resource of the first sample network resource and the first sample network resource; and determining the dominant characteristic and the recessive characteristic of the first sample network resource, the dominant characteristic and the recessive characteristic of the positive similar network resource of the first sample network resource and the contrast characteristic between the positive similar network resource of the first sample network resource and the first sample network resource as second positive sample data corresponding to any one first positive sample data.

Optionally, the fourth determining unit is configured to:

determining the dominant characteristic and the recessive characteristic of the second sample network resource and the dominant characteristic and the recessive characteristic of the negative similar network resource of the second sample network resource as second negative sample data corresponding to any one first negative sample data;

comparing the statistical characteristics of the second sample network resource with the statistical characteristics of the positive similar network resource of the second sample network resource to obtain the comparison characteristics between the negative similar network resource of the second sample network resource and the second sample network resource; and determining the dominant characteristic and the recessive characteristic of the second sample network resource, the dominant characteristic and the recessive characteristic of the negative similar network resource of the second sample network resource and the comparison characteristic between the negative similar network resource of the second sample network resource and the second sample network resource as second negative sample data corresponding to any one first negative sample data.

In the embodiment of the application, for the reference network resource browsed by the user, the dominant characteristic and the recessive characteristic of the reference network resource can be obtained, the similarity between each network resource in the first network resource library and the reference network resource is determined through the association model based on the dominant characteristic and the recessive characteristic of the reference network resource and the dominant characteristic and the recessive characteristic of each network resource in the first network resource library, and then resource recommendation is performed for the user based on the similarity between each network resource in the first network resource library and the reference network resource. The explicit characteristics comprise attribute characteristics and statistical characteristics, and the implicit characteristics refer to vector representation mapped to the implicit space, so that the similarity of any two sample network resources can be evaluated by combining the attribute characteristics, the statistical characteristics and the implicit characteristics of the network resources, and resource recommendation is performed based on the similarity of the network resources, so that the evaluation of the similarity is more comprehensive and accurate, the accuracy of determining the similarity of the network resources is improved, and the recommendation accuracy is further improved.

It should be noted that: in the recommendation apparatus for network resources according to the foregoing embodiment, only the division of the functional modules is illustrated when recommending network resources, and in practical applications, the function allocation may be completed by different functional modules according to needs, that is, the internal structure of the apparatus is divided into different functional modules to complete all or part of the functions described above. In addition, the recommendation apparatus for network resources and the recommendation method for network resources provided in the foregoing embodiments belong to the same concept, and specific implementation processes thereof are described in detail in the method embodiments and are not described herein again.

Fig. 5 is a schematic structural diagram of a server 500 according to an embodiment of the present application, where the server 500 may generate a relatively large difference due to different configurations or performances, and may include one or more processors (CPUs) 501 and one or more memories 502, where the memory 502 stores at least one instruction, and the at least one instruction is loaded and executed by the processors 501 to implement the recommendation method for network resources provided by the above-mentioned method embodiments. Of course, the server 500 may also have components such as a wired or wireless network interface, a keyboard, and an input/output interface, so as to perform input and output, and the server 500 may also include other components for implementing the functions of the device, which is not described herein again.

In some embodiments, there is also provided a computer readable storage medium having stored therein at least one instruction, at least one program, set of codes, or set of instructions, which is loaded and executed by a processor to implement the method of pushing network resources in the above embodiments. For example, the computer readable storage medium may be a ROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.

It is noted that the computer-readable storage medium referred to herein may be a non-volatile storage medium, in other words, a non-transitory storage medium.

It should be understood that all or part of the steps for implementing the above embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. The computer instructions may be stored in the computer-readable storage medium described above.

That is, in some embodiments, there is also provided a computer program product containing instructions which, when run on a computer, cause the computer to perform the method for recommending network resources described above.

The above-mentioned embodiments are provided not to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for recommending network resources, the method comprising:

obtaining implicit characteristics of a plurality of sample network resources, wherein the sample network resources are network resources browsed by a sample user, and the implicit characteristics refer to vector representation mapped to an implicit space;

each second positive sample data comprises an explicit characteristic and an implicit characteristic of the first sample network resource, and an explicit characteristic and an implicit characteristic of a positive similar network resource of the first sample network resource, and each second negative sample data comprises an explicit characteristic and an implicit characteristic of the second sample network resource, and an explicit characteristic and an implicit characteristic of a negative similar network resource of the second sample network resource; the explicit features comprise attribute features and statistical features;

training an association model to be trained based on the plurality of second positive sample data and the plurality of second negative sample data to obtain an association model;

acquiring an explicit characteristic and a implicit characteristic of a reference network resource, wherein the reference network resource is a network resource browsed by a user;

determining the similarity between each network resource in the first network resource library and the reference network resource through the association model based on the dominant and recessive characteristics of the reference network resource and the dominant and recessive characteristics of each network resource in the first network resource library, wherein the association model is used for determining the similarity between any two network resources;

2. The method of claim 1, wherein the obtaining the implicit characteristic of the reference network resource comprises:

3. The method of claim 1, wherein when the reference network resource is a song, the attribute feature comprises at least one of artist, genre of genre, album of genre, release time, and location of artist, and the statistical feature comprises at least one of play amount, collection amount, and comment amount.

4. The method of claim 1, wherein determining the similarity between each network resource in the first network resource pool and the reference network resource through the association model based on the explicit characteristics and the implicit characteristics of the reference network resource and the explicit characteristics and the implicit characteristics of each network resource in the first network resource pool comprises:

5. The method of claim 1, wherein determining the similarity between each network resource in the first network resource pool and the reference network resource through the association model based on the explicit characteristics and the implicit characteristics of the reference network resource and the explicit characteristics and the implicit characteristics of each network resource in the first network resource pool comprises:

6. The method of claim 1, wherein the making resource recommendations for the user based on the similarity between each network resource in the first network resource pool and the reference network resource comprises:

7. The method of claim 1, wherein determining k similar network resources for each sample network resource from a second pool of network resources based on implicit characteristics of each sample network resource in the plurality of sample network resources comprises:

8. The method according to claim 1, wherein said determining a plurality of second positive sample data and a plurality of second negative sample data based on said plurality of first positive sample data and a plurality of first negative sample data comprises:

9. The method of claim 8, wherein determining second positive sample data corresponding to any of the first positive sample data based on the explicit and implicit characteristics of the first sample network resources and the explicit and implicit characteristics of the positively similar network resources of the first sample network resources comprises:

10. The method according to claim 8, wherein said determining second negative sample data corresponding to any of the first negative sample data based on the explicit and implicit characteristics of the second sample network resource and the explicit and implicit characteristics of the negative similar network resource of the second sample network resource comprises:

comparing the statistical characteristics of the second sample network resources with the statistical characteristics of the negative similar network resources of the second sample network resources to obtain comparison characteristics between the negative similar network resources of the second sample network resources and the second sample network resources; and determining the dominant characteristic and the recessive characteristic of the second sample network resource, the dominant characteristic and the recessive characteristic of the negative similar network resource of the second sample network resource, and the contrast characteristic between the negative similar network resource of the second sample network resource and the second sample network resource as second negative sample data corresponding to any one first negative sample data.

11. An apparatus for recommending network resources, the apparatus comprising:

the recommendation module is used for recommending resources for the user based on the similarity between each network resource in the first network resource library and the reference network resource;

the device further comprises:

12. The apparatus of claim 11, wherein the obtaining module is configured to:

13. The apparatus of claim 11, wherein when the reference network resource is a song, the attribute feature comprises at least one of artist, genre of genre, album of genre, release time, and location of artist, and the statistical feature comprises at least one of play amount, collection amount, and comment amount.

14. The apparatus of claim 11, wherein the first determining module is configured to:

15. The apparatus of claim 11, wherein the first determining module is configured to:

16. The apparatus of claim 11, wherein the recommendation module is configured to:

17. The apparatus of claim 11, wherein the second determining module comprises:

18. The apparatus of claim 11, wherein the fourth determining module comprises:

19. The apparatus of claim 18, wherein the third determining unit is configured to:

20. The apparatus of claim 18, wherein the fourth determining unit is configured to:

21. A server, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a set of codes, or a set of instructions, and the at least one instruction, the at least one program, the set of codes, or the set of instructions is loaded and executed by the processor to implement a method for recommending network resources according to any of claims 1-10.

22. A computer readable storage medium having stored therein at least one instruction, at least one program, a set of codes, or a set of instructions, which is loaded and executed by a processor to implement a method of recommending network resources according to any of claims 1 to 10.