CN113569136A - Video recommendation method and device, electronic equipment and storage medium - Google Patents

Video recommendation method and device, electronic equipment and storage medium Download PDF

Info

Publication number
CN113569136A
CN113569136A CN202110752995.7A CN202110752995A CN113569136A CN 113569136 A CN113569136 A CN 113569136A CN 202110752995 A CN202110752995 A CN 202110752995A CN 113569136 A CN113569136 A CN 113569136A
Authority
CN
China
Prior art keywords
video
account
preset operation
encoding
layer
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110752995.7A
Other languages
Chinese (zh)
Other versions
CN113569136B (en
Inventor
常超
宋金波
胡国勇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Dajia Internet Information Technology Co Ltd
Original Assignee
Beijing Dajia Internet Information Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Dajia Internet Information Technology Co Ltd filed Critical Beijing Dajia Internet Information Technology Co Ltd
Priority to CN202110752995.7A priority Critical patent/CN113569136B/en
Publication of CN113569136A publication Critical patent/CN113569136A/en
Application granted granted Critical
Publication of CN113569136B publication Critical patent/CN113569136B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
    • G06F16/71—Indexing; Data structures therefor; Storage structures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
    • G06F16/78—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/783—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Abstract

The disclosure relates to a video recommendation method, a video recommendation device, an electronic device and a storage medium, wherein the method comprises the following steps: acquiring a topological graph corresponding to each preset operation; for the topological graph corresponding to each preset operation, respectively executing a first operation on each account and respectively executing a second operation on each video by using a graph neural network based on the topological graph corresponding to the preset operation; determining the total code of each account based on the total code of each account corresponding to the preset operation, and determining the total code of each video based on the total code of each video corresponding to the preset operation; and determining the recommended video corresponding to each account based on the total code of each account and the total code of each video. The method comprises the steps of corresponding to each preset operation in a plurality of preset operations, considering the correlation between a plurality of accounts and a plurality of videos and the preset operation, determining a recommended video corresponding to each account by using a graph neural network based on a topological graph corresponding to each preset operation, and improving the accuracy of video recommendation.

Description

Video recommendation method and device, electronic equipment and storage medium
Technical Field
The present disclosure relates to the field of videos, and in particular, to a video recommendation method and apparatus, an electronic device, and a storage medium.
Background
In the related art, when video recommendation is performed on one account, a video in which the account has performed an operation is determined, a video related to the video in which the account has performed the operation is searched for from other videos, and the video related to the video in which the account has performed the operation is taken as a recommended video corresponding to the account. Only the relevance between the video operated by the account and other videos is considered, and factors which can influence the accuracy of video recommendation, such as the relevance between the operation performed by the account and other videos and the relevance between the account and other accounts operated by the video operated by the user, are not considered, so that the accuracy of video recommendation is low.
Disclosure of Invention
In order to overcome the problems in the related art, the present disclosure provides a video recommendation method, apparatus, electronic device and storage medium, so as to at least solve the problem of low accuracy of video recommendation in the related art. The technical scheme of the disclosure is as follows:
according to a first aspect of the embodiments of the present disclosure, there is provided a video recommendation method, including:
acquiring a topological graph corresponding to each preset operation, wherein the topological graph corresponding to the preset operation indicates the relation between a plurality of accounts and a plurality of videos, and the relation is related to the preset operation;
for the topological graph corresponding to each preset operation, respectively executing a first operation on each account and respectively executing a second operation on each video by using a graph neural network based on the topological graph corresponding to the preset operation, wherein the first operation comprises the following steps: determining a total code of the account corresponding to the preset operation based on codes of the account corresponding to the preset operation and output by each coding layer of the graph neural network, wherein the codes of the account corresponding to the preset operation and output by the coding layers are determined based on the codes of the associated videos of the account corresponding to the preset operation and output by the previous coding layer; the second operation includes: determining a total encoding of the video corresponding to the preset operation based on the encoding of the video corresponding to the preset operation output by each encoding layer, wherein the encoding of the video corresponding to the preset operation output by the encoding layer is determined based on the encoding of the associated account of the video corresponding to the preset operation output by the previous layer of the encoding layer;
for each account, determining the total code of the account based on the total code of the account corresponding to the preset operation, and for each video, determining the total code of the video based on the total code of the video corresponding to the preset operation;
and determining the recommended video corresponding to each account based on the total code of each account and the total code of each video.
According to a second aspect of the embodiments of the present disclosure, there is provided a video recommendation apparatus including:
the system comprises an acquisition module, a processing module and a display module, wherein the acquisition module is configured to acquire a topological graph corresponding to each preset operation, and the topological graph corresponding to the preset operation indicates the relation between a plurality of accounts and a plurality of videos and relevant to the preset operation;
the encoding module is configured to, for a topological graph corresponding to each preset operation, respectively perform a first operation on each account and respectively perform a second operation on each video by using a graph neural network based on the topological graph corresponding to the preset operation, where the first operation includes: determining a total code of the account corresponding to the preset operation based on codes of the account corresponding to the preset operation and output by each coding layer of the graph neural network, wherein the codes of the account corresponding to the preset operation and output by the coding layers are determined based on the codes of the associated videos of the account corresponding to the preset operation and output by the previous coding layer; the second operation includes: determining a total encoding of the video corresponding to the preset operation based on the encoding of the video corresponding to the preset operation output by each encoding layer, wherein the encoding of the video corresponding to the preset operation output by the encoding layer is determined based on the encoding of the associated account of the video corresponding to the preset operation output by the previous layer of the encoding layer;
a total code determining module configured to determine, for each account, a total code of the account based on the total codes of the accounts corresponding to the preset operations, and determine, for each video, a total code of the video based on the total codes of the videos corresponding to the preset operations;
and the recommended video determining module is configured to determine the recommended video corresponding to each account based on the total code of each account and the total code of each video.
The technical scheme provided by the embodiment of the disclosure can have the following beneficial effects:
when video recommendation is performed, for each preset operation in a plurality of preset operations, the correlation between a plurality of accounts and a plurality of videos and the preset operation is considered, the corresponding topological graph corresponding to the corresponding preset operation is used for indicating the correlation between the plurality of accounts and the plurality of videos and the preset operation, the total code of each account and the total code of each video are determined based on the topological graph corresponding to each preset operation by using a graph neural network, the recommended video corresponding to each account is determined, and the accuracy of video recommendation is improved.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and together with the description, serve to explain the principles of the disclosure.
FIG. 1 is a flow diagram illustrating one embodiment of a video recommendation method in accordance with an exemplary embodiment;
FIG. 2 is a schematic diagram of a topology map corresponding to the set operation;
fig. 3 is a block diagram illustrating a configuration of a video recommendation apparatus according to an exemplary embodiment;
fig. 4 is a block diagram illustrating a server according to an example embodiment.
Detailed Description
In order to make the technical solutions of the present disclosure better understood by those of ordinary skill in the art, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.
It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the above-described drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the disclosure described herein are capable of operation in sequences other than those illustrated or otherwise described herein. The implementations described in the exemplary embodiments below are not intended to represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.
FIG. 1 is a flow diagram illustrating one embodiment of a video recommendation method in accordance with an example embodiment. The method comprises the following steps:
step 101, obtaining a topological graph corresponding to each preset operation.
For example, the plurality of preset operations include: the method comprises the steps of effective watching operation, praise operation and watching completion operation, wherein the effective watching operation is that the watching duration is longer than a duration threshold, the praise operation is that a control for indicating favorite videos is clicked, the watching completion operation is that the videos are watched completely, and a topological graph corresponding to the effective watching operation, a topological graph corresponding to the praise operation and a topological graph corresponding to the watching completion operation can be obtained.
In this disclosure, the operation topology corresponding to the preset operation may include: account nodes corresponding to each account in the plurality of accounts and video nodes corresponding to each video in the plurality of videos.
And indicating the relation between the plurality of accounts and the plurality of videos related to the preset operation by the topological graph corresponding to the preset operation.
For each account in the plurality of accounts, in a topological graph corresponding to a preset operation, an account node corresponding to the account is connected with at least one video node, the account node corresponding to the account is connected with a video node corresponding to a video to indicate that the account performs the preset operation on the video, and the video is a related video of the account corresponding to the preset operation. For each video in the plurality of videos, in a topological graph corresponding to a preset operation, a video node corresponding to the video is connected with at least one account node, the video node corresponding to the video is connected with an account node corresponding to an account, the preset operation is performed on the video by the account, and the account is an associated account of the video corresponding to the preset operation.
Please refer to fig. 2, which illustrates a schematic diagram of a topology diagram corresponding to a preset operation.
In fig. 2, a topological diagram corresponding to three preset operations is exemplarily shown. The active view operation may be referred to as Effect-view, the like operation may be referred to as Favorite, and the view complete operation may be referred to as Long-view.
In the topological graph corresponding to Effective-view, an account node corresponding to an account u1 is simultaneously connected with a video node corresponding to a video i1, a video node corresponding to a video i2 and a video node corresponding to a video i4, and shows that account u1 carries out Effective-view on i1, i2 and i4, and i1, i2 and i4 are all related videos corresponding to the Effective-view and of an account u 1.
In the topological graph corresponding to Effective-view, a video node corresponding to a video i1 is simultaneously connected with an account node corresponding to an account u1, an account node corresponding to an account u4 and an account node corresponding to an account u5, and u1, u4 and u5 are all related accounts corresponding to Effective-view of the video i 1.
In the topological graph corresponding to the Favorite, an account node corresponding to an account u1 is simultaneously connected with a video node corresponding to a video i1, a video node corresponding to a video i2 and a video node corresponding to a video i3, and shows that the account u1 performs Favorite on i1, i2 and i3, and i1, i2 and i3 are all associated videos corresponding to the Favorite and of the account u 1.
In the topological graph corresponding to the Favorite, a video node corresponding to the video i1 is simultaneously connected with an account node corresponding to the account u1, an account node corresponding to the account u4 and an account node corresponding to the account u6, and u1, u4 and u6 are all associated accounts corresponding to the Favorite and of the video i 1.
In the topological graph corresponding to the Long-view, an account node corresponding to an account u1 is simultaneously connected with a video node corresponding to a video i1, a video node corresponding to a video i3 and a video node corresponding to a video i4, and shows that the account u1 performs the Long-view on i1, i3 and i4, and i1, i3 and i4 are all associated videos corresponding to the Long-view and corresponding to an account u 1.
In the topological graph corresponding to Long-view, in the topological graph corresponding to Favorite, the video node corresponding to video i1 is simultaneously connected with the account node corresponding to account u1, the account node corresponding to account u4 and the account node corresponding to account u6, and u1, u4 and u6 are all related accounts corresponding to Long-view of video i 1.
And 102, for the topological graph corresponding to each preset operation, respectively executing a first operation on each account and respectively executing a second operation on each video by using a graph neural network based on the topological graph corresponding to the preset operation.
In the present disclosure, a Graph Neural Network (GNN) may include: the account initial coding and the video initial coding generate a layer and a plurality of coding layers.
The account primary encoding and video primary encoding generation layers may perform an encoding (embedding) operation on each of the accounts, resulting in a primary encoding for each of the plurality of accounts. The account initial encoding and video initial encoding generation layers may perform an encoding operation on each of the plurality of videos, respectively, resulting in an initial encoding of each of the plurality of videos.
The account coding and video coding generates layer output, the coding of each account is the initial coding of each account, the account coding and video coding generates layer output, the initial coding of each video is the initial coding of each video.
In the present disclosure, for the 1 st coding layer in the graph neural network, the previous layer of the 1 st coding layer is the account coding and video coding generation layer.
For each other coding layer except the 1 st coding layer in the plurality of coding layers in the graph neural network, the previous layer of the other coding layers is the previous coding layer of the other coding layers.
In the disclosure, for the topological graph corresponding to each preset operation, a graph neural network is used for respectively executing a first operation on each account in a plurality of accounts and respectively executing a second operation on each video in a plurality of videos based on the topological graph corresponding to the preset operation.
For example, the plurality of preset operations include: effective watching operation, praise operation and watching completion operation.
And respectively executing a first operation on each account by using the graph neural network based on the topological graph corresponding to the effective viewing operation to obtain the total code of each account, which corresponds to the effective viewing operation. And respectively executing a second operation on each video by using the graph neural network based on the topological graph corresponding to the effective viewing operation to obtain the total code of each video, which corresponds to the effective viewing operation.
And respectively executing a first operation on each account by using the graph neural network based on the topological graph corresponding to the praise operation to obtain a total code corresponding to the praise operation of each account. And respectively executing a second operation on each video by using the graph neural network based on the topological graph corresponding to the praise operation to obtain the total code of each video, which corresponds to the praise operation.
And respectively executing a first operation on each account by using the graph neural network based on the topological graph corresponding to the viewing completion operation to obtain the total code of each account corresponding to the viewing completion operation. And respectively executing a second operation on each video by using the graph neural network based on the topological graph corresponding to the watching completion operation to obtain the total code of each video, which corresponds to the watching completion operation.
In the disclosure, for each account of a plurality of accounts, performing a first operation on the account based on a topological graph corresponding to a preset operation by using a graph neural network includes: and determining a total code of the account corresponding to the preset operation based on the codes of the account corresponding to the preset operation output by each coding layer.
In the disclosure, when a first operation is performed on an account based on a topological graph corresponding to a preset operation by using a graph neural network, for each coding layer, a code corresponding to the preset operation of the account output by the coding layer may be determined based on a code of an associated video of the account, which is output by a layer preceding the coding layer and corresponds to the preset operation.
For example, when a first operation is performed on an account based on a topological graph corresponding to a preset operation by using a graph neural network, for each coding layer, the codes of the associated videos, which are output by the previous layer of the coding layer and correspond to the preset operation, of the account may be added, and the obtained result is divided by the square root of the number of all the associated videos, which correspond to the preset operation, of the account, so as to obtain the code, which is output by the coding layer and corresponds to the preset operation, of the account.
The following examples illustrate a process of performing a first operation on one account based on a topological graph corresponding to one preset operation by using a graph neural network, and a process of performing a first operation on any account based on a topological graph corresponding to any preset operation by using a graph neural network, which is the same as the process:
when a first operation is performed on an account u by using the graph neural network based on a topological graph corresponding to a preset operation b, the graph neural network can determine associated videos of the account u, each corresponding to b, based on the topological graph corresponding to b.
When the first operation is executed on the account u based on the topological graph corresponding to b by using the graph neural network, for each coding layer, the coding corresponding to b of the account u output by the coding layer can be determined based on the coding of the associated video corresponding to b of the account u output by the previous layer of the coding layer.
After determining the code corresponding to b for account u output by each encoding layer, a total code corresponding to b for account u is determined based on the code corresponding to b for account u output by each encoding layer. The codes corresponding to b of the account u output by each coding layer may be added, and the obtained result is divided by the number of coding layers to obtain the total code corresponding to b of the account u.
That is, the total code for account u corresponding to b may be determined using the following formula
Figure BDA0003145755840000061
Figure BDA0003145755840000071
Figure BDA0003145755840000072
And representing the code of the account u output by the kth coding layer and corresponding to the preset operation b, wherein K represents the number of the coding layers.
In some embodiments, when a first operation is performed on an account based on a topological graph corresponding to a preset operation by using a graph neural network, for each coding layer, the coding output by the coding layer and corresponding to the preset operation of the account is determined by the following steps: for each associated video of the account corresponding to the preset operation, determining the weight of the associated video corresponding to the preset operation based on the association degree of the code of the associated video corresponding to the preset operation output by the previous layer of the coding layer and the code of each other associated video of the account corresponding to the preset operation output by the previous layer of the coding layer; calculating a weighted sum of the codes of the associated videos, which are output by a previous layer of the coding layer and correspond to the preset operation, of the account by using the weight of each associated video corresponding to the preset operation; and obtaining the code of the account corresponding to the preset operation, which is output by the coding layer, based on the weighted sum of the codes of each associated video corresponding to the preset operation, which are output by the previous layer of the coding layer, and the number of all associated videos corresponding to the preset operation, which are output by the account.
The other associated videos of an account corresponding to a preset operation are not specific to the associated video of the account corresponding to the preset operation. The other associated videos of an account corresponding to the preset operation are relative to a specific associated video of the account corresponding to the preset operation.
For example, when the weight of the associated video i corresponding to b of the account u is calculated, for the associated video i corresponding to b of the account u, the associated videos except the video i of the account u are the associated videos of the account u and other associated videos corresponding to b, and the number of the associated videos of the account u and other associated videos corresponding to b is one or more.
For the account u and the preset operation b, when calculating the encoding of the corresponding b of the account u output by the k-th encoding layer, for each associated video of the account u corresponding to the b, determining the weight of the associated video corresponding to the b based on the association degree of the encoding of the associated video corresponding to the b output by the previous layer of the k-th encoding layer and the encoding of each other associated video of the account u corresponding to the b output by the previous layer of the k-th encoding layer.
For example, for the account u and the preset operation b, when the encoding corresponding to b of the account u output by the k-th encoding layer is calculated, if the number of other associated videos corresponding to b of the account u is one for the associated video i corresponding to b of the account u, the association degree between the encoding corresponding to b of the account u output by the previous layer of the k-th encoding layer and the encoding corresponding to b of the account u output by the previous layer of the k-th encoding layer is determined as the weight of the associated video i corresponding to b of the account u.
If the number of the other related videos corresponding to b of the account u is multiple for the related video i corresponding to b of the account u, the association degree of the coding of the related video i corresponding to b of the account u output by the previous layer of the k-th coding layer and the coding of each other related video corresponding to b of the account u output by the previous layer of the k-th coding layer is respectively determined, and multiple association degrees are obtained. If there are two, one of the minimum value, the maximum value, and the average value of the two association degrees may be determined as the weight of the associated video i corresponding to b for the account u. If the number is greater than two, a median of the number of degrees of association may be determined as the weight of the associated video i of account u corresponding to b.
In the disclosure, for an account and a preset operation, when determining a code of the account output by a coding layer and corresponding to the preset operation, the weight of each associated video corresponding to the preset operation of the account is used to represent the association degree between the code of each associated video corresponding to the preset operation of the account and the code of the account output by the coding layer, the association degree between the code of each associated video corresponding to the preset operation of the account and the code of the account output by the coding layer is considered, the association degree between the code of each associated video corresponding to the preset operation of the account and the code of the account output by the coding layer and corresponding to the preset operation is considered, and the accuracy of the determined code of the account output by the coding layer and corresponding to the preset operation is improved.
In the present disclosure, a function f (·) may be used to calculate a degree of association between an encoding of an associated video output by a previous layer of a k-th encoding layer, where one of the accounts u corresponds to a preset operation, and an encoding of an associated video output by the previous layer of the k-th encoding layer, where another of the accounts u corresponds to the preset operation. f (.) may be a ReLU function.
f () is related to the encoding of the associated video of the account u corresponding to the preset operation output by the previous layer of the k-th encoding layer and the encoding of the associated video of the account u corresponding to the preset operation output by the previous layer of the k-th encoding layer.
In calculating the degree of association of the encoding of the associated video i corresponding to b of the account u with the encoding of one other associated video j corresponding to b of the account u, f (·) can be expressed as the portion to the right of the equal sign in the following formula:
Figure BDA0003145755840000091
Figure BDA0003145755840000092
representing the encoding of the associated video i corresponding to the preset operation b of the account u output by the layer preceding the kth encoding layer,
Figure BDA0003145755840000093
representing the encoding of other associated video j of the account u corresponding to the preset operation b, output by the layer preceding the kth encoding layer.
In some embodiments, when performing a first operation on an account based on a topological graph corresponding to a preset operation by using a graph neural network, for an encoding layer, determining a weight of an associated video of the account corresponding to the preset operation based on an association degree between an encoding of the associated video of the account corresponding to the preset operation output by a previous layer of the encoding layer and an encoding of each other associated video of the account corresponding to the preset operation output by the previous layer of the encoding layer comprises: determining the average value of the association degrees of the encoding of the associated video corresponding to the preset operation of the account output by the previous layer of the encoding layer and the encoding of the other associated video corresponding to the preset operation of the account output by the previous layer of the encoding layer as the weight of the associated video corresponding to the preset operation of the account.
When determining the weight of the associated video i corresponding to the preset operation b of the account u output by one coding layer, the association degree of the associated video i corresponding to b of the account u output by the previous layer of the coding layer and each other associated video i corresponding to b of the account u output by the previous layer of the coding layer may be added, and the obtained result is divided by the number of all associated videos corresponding to b of the account u, and the weight of the associated video i corresponding to b of the account u.
That is, the weight of account u corresponding to associated video i of b can be calculated using the following formula
Figure BDA0003145755840000094
Figure BDA0003145755840000095
NuA set of associated videos corresponding to b representing account u, | Nu | represents the number of all associated videos corresponding to b for account u.
In the disclosure, for an account and a preset operation, when a weight of an associated video of the account corresponding to the preset operation is determined based on a correlation degree of an associated video of the account output by a previous layer of an encoding layer, the associated video of the account corresponding to the preset operation being output by the previous layer of the encoding layer, and each other associated video of the account corresponding to the preset operation being output by the previous layer of the encoding layer, an average value of the correlation degree of the associated video of the account corresponding to the preset operation, the associated video of the account output by the previous layer of the encoding layer, the associated video of the account corresponding to the preset operation, and each other associated video of the account corresponding to the preset operation being output by the previous layer of the encoding layer is calculated. The average value can reflect the relevance of the total of the encoding of the video associated with the preset operation of the account and the encoding of the other video associated with the preset operation of the account, which are output by the previous layer of the encoding layer, relatively accurately, and the average value is determined as the weight of the video associated with the preset operation of the account, so that the accuracy of the determined weight of the video associated with the preset operation of the account is improved.
In the disclosure, for an account u and a preset operation b, when the encoding corresponding to b of the account u output by the k-th encoding layer is calculated, a weighted sum of the encoding corresponding to b of the account u output by the previous layer of the k-th encoding layer is calculated by using the weight of the associated video corresponding to b of each account u, and the encoding corresponding to b of the account u output by the k-th encoding layer is obtained based on the weighted sum of the encoding corresponding to b of the associated video of each account u output by the previous layer of the k-th encoding layer and the number of all the associated videos corresponding to b of the account u.
The coding of account u corresponding to b output by the kth coding layer may be obtained by dividing a weighted sum of the codings of account u each corresponding to b associated video output by the previous layer of the kth coding layer by the square root of the number of all associated videos of account u corresponding to b.
The following formula can be adopted to obtain the code of the k coding layer output corresponding to b of the account u
Figure BDA0003145755840000101
Figure BDA0003145755840000102
Figure BDA0003145755840000103
The weight of the associated video i corresponding to b representing account u,
Figure BDA0003145755840000104
representing the encoding of the associated video i corresponding to the preset operation b of the account u output by the layer preceding the kth encoding layer.
In the disclosure, for each of the plurality of videos, performing, by using the graph neural network, a second operation on the video based on the topological graph corresponding to the preset operation includes: and determining the total code of the video corresponding to the preset operation based on the codes of the video corresponding to the preset operation output by each coding layer.
In the disclosure, when a second encoding operation is performed on a video based on a topological graph corresponding to a preset operation by using a graph neural network, for each encoding layer, an encoding output by the encoding layer and corresponding to the preset operation may be determined based on an encoding of an associated account of the video output by a layer preceding the encoding layer and corresponding to the preset operation.
For example, when a second encoding operation is performed on a video based on a topological graph corresponding to a preset operation by using a graph neural network, for each encoding layer, the encodings of each associated account of the video, which are output by a layer before the encoding layer and correspond to the preset operation, may be added, and the obtained result is divided by the square root of the number of all associated accounts of the video, which correspond to the preset operation, to obtain the encoding of the video, which is output by the encoding layer and corresponds to the preset operation.
The following examples illustrate the process of performing the second operation on one video based on the topological graph corresponding to one preset operation by using the graph neural network, and the process of performing the second operation on any one video based on the topological graph corresponding to any one preset operation by using the graph neural network is the same as the process:
when a second operation is performed on a video i based on a topological graph corresponding to a preset operation b by using the graph neural network, the graph neural network can determine associated accounts of the video i, each corresponding to b, based on the topological graph corresponding to b, and for each associated account of the video i, each associated video corresponding to b can be determined by the graph neural network.
When the second operation is performed on the video i based on the topological graph corresponding to b by using the graph neural network, for each coding layer, the coding corresponding to b of the video i output by the coding layer can be determined based on the coding of the associated account corresponding to b of the video i output by the previous layer of the coding layer.
After determining the encoding of video i corresponding to b output by each encoding layer, a total encoding of video i corresponding to b may be determined based on the encoding of video i corresponding to b output by each encoding layer.
In the present disclosure, the following formula may be employed to determine the overall encoding of video i corresponding to b
Figure BDA0003145755840000111
Figure BDA0003145755840000112
Figure BDA0003145755840000121
Represents the encoding of the video i output by the kth encoding layer corresponding to the preset operation b, and K represents the number of encoding layers.
In some embodiments, when performing the second operation on a video based on a topological graph corresponding to a preset operation by using the graph neural network, for each coding layer, the coding output by the coding layer and corresponding to the preset operation is determined by the following steps: for each associated account of the video corresponding to the preset operation, determining the weight of the associated account corresponding to the preset operation based on the association degree of the code of the associated account corresponding to the preset operation output by the previous layer of the coding layer and the code of each other associated account of the video corresponding to the preset operation output by the previous layer of the coding layer; calculating a weighted sum of codes, which are output by a previous layer of the coding layer and correspond to each associated account of the preset operation, by using the weight of each associated account corresponding to the preset operation; and obtaining the code of the video, which is output by the coding layer and corresponds to the preset operation, based on the weighted sum of the codes, which are output by the previous layer of the coding layer and correspond to each associated account of the preset operation, of the video and the number of all associated accounts of the video, which correspond to the preset operation.
The other associated accounts of a video corresponding to the preset operation are relative to a specific associated account of the video corresponding to the preset operation.
For example, when the weight of the associated account u corresponding to b of the video i is calculated, the accounts other than u in all the associated accounts corresponding to b of the video i are the other associated accounts corresponding to b of the account u, and the number of the other associated accounts corresponding to b of the account u is one or more.
In the disclosure, for a video i and a preset operation b, when calculating an encoding corresponding to b of the video i output by a k-th encoding layer, for each associated account corresponding to b of the video i, determining a weight of the associated account corresponding to b based on a degree of association between an encoding of the associated account corresponding to b output by a previous layer of the k encoding layers and an encoding of each other associated account corresponding to b output by a previous layer of the k encoding layers.
For example, for a video i and a preset operation b, when calculating the encoding corresponding to b of the video i output by the k-th encoding layer, if the number of other associated accounts corresponding to b of the video i is one for the associated account u corresponding to b of the video i, the association degree between the encoding corresponding to b of the video i output by the previous layer of the k-th encoding layer and the encoding corresponding to b of the other associated video is determined as the weight of the associated account u corresponding to b of the account u.
If the number of the associated accounts u corresponding to b of the video i is multiple, the association degrees of the codes of the associated accounts u corresponding to b of the video i and the codes of each other associated account u corresponding to b of the video i output by the previous layer of the k-th coding layer and the codes of the associated accounts u corresponding to b of the video i output by the previous layer of the k-th coding layer are respectively determined, and the multiple association degrees are obtained. If there are two, one of the minimum value, the maximum value, and the average value of the two determined association degrees may be determined as the weight of the associated account u corresponding to b of the video i. If the number is more than two, the median of the determined multiple association degrees can be determined as the weight of the associated account u of the video i corresponding to the b.
When determining the encoding of video i corresponding to b output to the k-th encoding layer, the weighted sum of the encodings of video i corresponding to b output from the previous layer of the k-th encoding layer may be divided by the square root of the number of all associated accounts of video i corresponding to b to obtain the encoding of video i corresponding to b output from the k-th encoding layer.
In the disclosure, for a video and a preset operation, when determining an encoding of the video output by an encoding layer and corresponding to the preset operation, the weight of each associated account corresponding to the preset operation of the video is used to represent the association degree of the encoding of each associated account corresponding to the preset operation of the video output by a previous layer of the encoding layer and the encoding of the video output by the encoding layer and corresponding to the preset operation, the association degree of each associated account corresponding to the preset operation of the video output by the previous layer of the encoding layer and the encoding of the video output by the encoding layer and corresponding to the preset operation is considered, and the accuracy of the determined encoding of the video output by the encoding layer and corresponding to the preset operation is improved.
In some embodiments, when determining the encoding of the video output by an encoding layer and corresponding to a preset operation by using a topological graph corresponding to the preset operation through a graph neural network, determining the weight of the associated account of the video corresponding to the preset operation based on the association degree between the encoding of the associated account of the video output by the previous layer of the encoding layer and each other encoding of the associated account of the video output by the previous layer of the encoding layer and corresponding to the preset operation includes: and determining the average value of the association degrees of the codes of the video, which are output by the previous layer of the coding layer and correspond to the associated account of the preset operation, and the codes of the other associated accounts of the video, which are output by the previous layer of the coding layer and correspond to the preset operation, as the weight of the associated account of the video, which corresponds to the preset operation.
When the weight of the associated account u corresponding to b of the video i output by one coding layer is determined, the association degree of the code of the associated account u corresponding to b of the video i output by the previous layer of the coding layer and the code of each other associated account corresponding to b of the video i output by the previous layer of the coding layer are added, and the obtained result is divided by the number of all associated accounts corresponding to b of the video i to obtain the weight of the associated account u corresponding to b of the video i.
For a video and a preset operation, when a weight of an associated account of the video corresponding to the preset operation is determined based on a correlation degree of an encoding of an associated account of the video corresponding to the preset operation and an encoding of each other associated account of the video corresponding to the preset operation, which are output by a previous layer of an encoding layer, the average of the correlation degrees of the encoding of the associated account of the video corresponding to the preset operation and the encoding of the other associated account of the video corresponding to the preset operation, which are output by the previous layer of the encoding layer, can be calculated. The average value can relatively accurately represent the overall association degree of the code output by the previous layer of the coding layer and the code of one associated account of the video corresponding to the preset operation and the codes output by the previous layer of the coding layer and the other associated accounts of the video corresponding to the preset operation, the average value is determined as the weight of one associated account of the video corresponding to the preset operation, and the accuracy of the determined weight of one associated account of the video corresponding to the preset operation is improved.
Step 103, for each account, determining the total code of the account based on the total code of the account corresponding to the preset operation, and for each video, determining the total code of the video based on the total code of the video corresponding to the preset operation.
For example, the plurality of preset operations include: effective watching operation, praise operation and watching completion operation.
And respectively executing a first operation on each account and a second operation on each video by using the graph neural network based on the topological graph corresponding to the effective viewing operation to obtain the total code corresponding to the effective viewing operation of each account and the total code corresponding to the effective viewing operation of each video.
And respectively executing a first operation on each account and a second operation on each video by using the graph neural network based on the topological graph corresponding to the praise operation to obtain the total code corresponding to the praise operation of each account and the total code corresponding to the praise operation of each video.
And respectively executing a first operation on each account and a second operation on each video by using the graph neural network based on the topological graph corresponding to the viewing completion operation to obtain the total code corresponding to the viewing completion operation of each account and the total code corresponding to the viewing completion operation of each video.
For each account, the total code of the account corresponding to the preset operation comprises: the total code of the account corresponding to the valid viewing operation, the total code of the account corresponding to the approval operation, and the total code of the account corresponding to the viewing completion operation are determined based on the total codes of the account corresponding to the preset operations, for example, the total code of the account corresponding to the valid viewing operation, the total code of the account corresponding to the approval operation, and the total code of the account corresponding to the viewing completion operation may be added to obtain the total code of the account.
For each video, the total encoding of the video corresponding to the preset operation comprises: the total encoding of the video corresponding to the valid viewing operation, the total encoding of the video corresponding to the approval operation, and the total encoding of the video corresponding to the viewing completion operation are determined based on the total encoding of the account corresponding to the preset operation, for example, the total encoding of the video corresponding to the valid viewing operation, the total encoding of the video corresponding to the approval operation, and the total encoding of the video corresponding to the viewing completion operation may be added to obtain the total encoding of the video.
In some embodiments, for each account, determining the total code of the account based on the total codes of the accounts corresponding to the preset operations comprises: and calculating the weighted sum of the total codes of the account, which correspond to the preset operations, by using the weight corresponding to each preset operation, and determining the weighted sum of the total codes of the account, which correspond to the preset operations, as the total code of the account. The weight corresponding to each preset operation may be hyper-parameters (hyper-parameters).
For example, the plurality of preset operations include: effective view operation Effective-view, like operation Favorite, view completion operation Long-view. For each account, the total code of the account corresponding to the preset operation comprises: the total code of the account corresponding to Effect-view, the total code of the account corresponding to Favorite, the total code of the account corresponding to Long-view.
The total code for account u may be determined using the following formula
Figure BDA0003145755840000151
Figure BDA0003145755840000152
Figure BDA0003145755840000153
A total code corresponding to Effective-view representing account u,
Figure BDA0003145755840000154
a total code corresponding to Favorite representing account u,
Figure BDA0003145755840000155
the total code corresponding to Long-view representing account u.
βEVRepresents the weight, β, corresponding to Effective-viewLRepresents the weight, β, corresponding to FavoriteLVAnd represents the weight corresponding to the Long-view.
In the disclosure, for an account, when the total code of the account is determined based on the total codes of the account, which correspond to the preset operations, the weight of each preset operation is used to represent the degree of association between each total code of the account, which corresponds to the preset operation, and the total code of the account, and the degree of association between each total code of the account, which corresponds to the preset operation, and the total code of the account is considered, so that the accuracy of the determined total code of the account is improved.
In some embodiments, for each video, determining the overall encoding of the video based on the overall encoding of the video that each corresponds to a preset operation comprises: and calculating the weighted sum of the total codes of the video, which correspond to the preset operations, by using the weight corresponding to each preset operation, and determining the weighted sum of the total codes of the video, which correspond to the preset operations, as the total code of the video.
For example, the plurality of preset operations include: effective view operation Effective-view, like operation Favorite, view completion operation Long-view. For each video, the total encoding of the video corresponding to the preset operation comprises: a total encoding of the video corresponding to Effect-view, a total encoding of the video corresponding to Favorite, a total encoding of the video corresponding to Long-view.
The following formula may be used to determine the overall encoding of video i
Figure BDA0003145755840000165
Figure BDA0003145755840000161
Figure BDA0003145755840000162
The overall code corresponding to Effective-view representing video i,
Figure BDA0003145755840000163
represents the total encoding of video i corresponding to Favorite,
Figure BDA0003145755840000164
representing the overall coding of video i, corresponding to Long-view.
In the disclosure, for a video, when the total coding of the video is determined based on the total coding of the video corresponding to the preset operations, the degree of association between each total coding of the video corresponding to the preset operations and the total coding of the video is represented by the weight of each preset operation, and the determined accuracy of the total coding of the video is improved by considering the degree of association between each total coding of the video corresponding to the preset operations and the total coding of the video.
And step 104, determining a recommended video corresponding to each account based on the total code of each account and the total code of each video.
In the present disclosure, for each account, a distance, for example, a euclidean distance, between the total code of the account and the total code of each video may be calculated, all videos are sorted according to the distance from small to large, and the top k videos are determined. For example, the top k videos may be determined using an Approximate Nearest Neighbor (ANN) search engine. The recommended video corresponding to the account may be determined as the first k videos or videos of the first k videos except for the associated video of the account corresponding to the preset operation.
According to the video recommendation method, when video recommendation is performed, for each preset operation in a plurality of preset operations, the relation between a plurality of accounts and a plurality of videos and the relation between the plurality of accounts and the plurality of videos and the preset operation is considered, the topological graph corresponding to the corresponding preset operation is used for indicating the relation between the plurality of accounts and the plurality of videos and the relation between the plurality of accounts and the preset operation and the relation between the plurality of accounts and the plurality of videos and the relation between the plurality of accounts and the relation between the plurality of videos and the preset operation, the total code of each account and the total code of each video are determined by using the graph neural network based on the topological graph corresponding to each preset operation, the recommended video corresponding to each account is determined, and the accuracy of video recommendation is improved.
In this disclosure, before obtaining the topological graph corresponding to each preset operation, the graph neural network may be repeatedly trained in the following manner:
in the process of training the neural network of the graph once, a topological graph which is used for training and corresponds to each preset operation is obtained. And acquiring the total code of each account for training and the total code of each video for training by using the graph neural network based on the topological graph corresponding to each preset operation for training.
The process of acquiring the total code of each account and the total code of each video for training by using the graph neural network based on the topological graph corresponding to each preset operation for training is the same as the process of acquiring the total code of each account and the total code of each video.
And in the process of training the neural network of the graph once, loss calculation operation is respectively executed on each preset operation, and the loss corresponding to each preset operation is obtained. Performing a penalty calculation operation on a predetermined operation includes: for each account, calculating a matching score between a total code of the account corresponding to the preset operation and a total code of an associated video of the account corresponding to the preset operation and corresponding to the preset operation, and calculating a matching score between the total code of the account corresponding to the preset operation and a total code of an unassociated video of the account corresponding to the preset operation and corresponding to the preset operation; and calculating the loss corresponding to the preset operation according to all the calculated matching scores.
After the loss corresponding to each preset operation is obtained, the total loss is calculated based on the loss corresponding to each preset operation, then, back propagation is carried out based on the total loss, and the parameters of the graph neural network are updated.
In this disclosure, for an account and a preset operation, if the account does not perform the preset operation on a video but at least one of the accounts other than the account performs the preset operation on the video, the video is a non-associated video of the account corresponding to the preset operation.
The following illustrates a process of calculating a loss corresponding to one preset operation, and the same process of calculating a loss corresponding to any one preset operation is as follows:
the matching score between the total code for account u, corresponding to b, and the total code for video i, corresponding to b, can be calculated using the following formula
Figure BDA0003145755840000171
Figure BDA0003145755840000172
Figure BDA0003145755840000181
The total code corresponding to b representing account u,
Figure BDA0003145755840000182
representing the overall coding of video i, corresponding to b.
In the present disclosure, can be prepared byThe following formula calculates the loss corresponding to the preset operation b
Figure BDA0003145755840000183
Figure BDA0003145755840000184
Wherein the content of the first and second substances,
Figure BDA0003145755840000185
representing the association of the account with the video observable in the topological graph corresponding to the preset operation b for training,
Figure BDA0003145755840000186
represents the association of the account with the video, not observed in the topological graph corresponding to the preset operation b used for training, sigma represents the soft-plus function,
Figure BDA0003145755840000187
a match score between the total code corresponding to b representing account u and the total code corresponding to b for video i,
Figure BDA0003145755840000188
a match score, S, between the total encoding of account u corresponding to b and the total encoding of video j corresponding to bbL represents the total number of the non-associated videos corresponding to b participating in the calculation of the loss corresponding to the preset operation b, video i is one associated video corresponding to b of the account u, video j is one non-associated video randomly selected from all the non-associated videos corresponding to b of the account u, and video i is one associated video corresponding to b of the account u and all the non-associated videos corresponding to b of the account u are observed through the topological graph corresponding to the preset operation b for training.
In the process of training the neural network once, after calculating the loss corresponding to each preset operation, the following method can be adoptedThe total loss L is calculated by a formulaMulti-behavior：
Figure BDA0003145755840000189
Figure BDA00031457558400001810
Representing the hyper-parameter set and η representing the coefficient control l2 regular.
Fig. 3 is a block diagram illustrating a configuration of a video recommendation apparatus according to an exemplary embodiment. Referring to fig. 3, the video recommendation apparatus includes: the video recommendation system comprises an acquisition module 301, an encoding module 302, a total encoding determination module 303 and a recommended video determination module 304.
The obtaining module 301 is configured to obtain a topological graph corresponding to each preset operation, where the topological graph corresponding to the preset operation indicates a relationship, related to the preset operation, between multiple accounts and multiple videos;
the encoding module 302 is configured to, for a topological graph corresponding to each preset operation, perform, by using a graph neural network, a first operation on each account and a second operation on each video respectively based on the topological graph corresponding to the preset operation, where the first operation includes: determining a total code of the account corresponding to the preset operation based on codes of the account corresponding to the preset operation and output by each coding layer of the graph neural network, wherein the codes of the account corresponding to the preset operation and output by the coding layers are determined based on the codes of the associated videos of the account corresponding to the preset operation and output by the previous coding layer; the second operation includes: determining a total encoding of the video corresponding to the preset operation based on the encoding of the video corresponding to the preset operation output by each encoding layer, wherein the encoding of the video corresponding to the preset operation output by the encoding layer is determined based on the encoding of the associated account of the video corresponding to the preset operation output by the previous layer of the encoding layer;
the total coding determination module 303 is configured to determine, for each account, a total coding of the account based on the total coding of the account corresponding to the preset operation, and determine, for each video, a total coding of the video based on the total coding of the video corresponding to the preset operation;
the recommended video determination module 304 is configured to determine a recommended video corresponding to each account based on the total encoding of each account and the total encoding of each video.
In some embodiments, the encoding module 302 includes:
an account code determining sub-module configured to determine, for each associated video of the account corresponding to the preset operation, a weight of the associated video corresponding to the preset operation based on a degree of association between an encoding of the associated video corresponding to the preset operation output by a previous layer of the encoding layer and an encoding of each other associated video of the account corresponding to the preset operation output by the previous layer of the encoding layer; calculating a weighted sum of the encoding of the associated video corresponding to the preset operation and output by the previous layer of the encoding layer by utilizing the weight of the associated video corresponding to the preset operation and output by each account; and obtaining the coding of the account corresponding to the preset operation, which is output by a coding layer, based on the weighted sum and the number of all the associated videos of the account corresponding to the preset operation.
In some embodiments, the account code determining submodule is further configured to determine an average of the association degrees of the coding of the associated video corresponding to the preset operation output by the previous layer of the coding layer and the coding of the other associated video corresponding to the preset operation output by the previous layer of the coding layer as the weight of the associated video corresponding to the preset operation.
In some embodiments, the encoding module 302 includes:
a video encoding determination sub-module configured to determine, for each associated account of the video corresponding to the preset operation, a weight of the associated account corresponding to the preset operation based on a degree of association between an encoding of the associated account corresponding to the preset operation output by a previous layer of the encoding layer and an encoding of each other associated account of the video corresponding to the preset operation output by the previous layer of the encoding layer; calculating a weighted sum of codes, which are output by a layer before the coding layer and correspond to each associated account of the preset operation, by using the weight of each associated account of the videos corresponding to the preset operation; and obtaining the code of the video corresponding to the preset operation, which is output by a coding layer, based on the weighted sum and the number of all the associated accounts of the video corresponding to the preset operation.
In some embodiments, the video encoding determination submodule is further configured to determine, as the weight of the associated account corresponding to the preset operation, an average of degrees of association between the encoding of the associated account corresponding to the preset operation output by the previous layer of the encoding layer and the encoding of the other associated accounts of the video corresponding to the preset operation output by the previous layer of the encoding layer.
In some embodiments, the overall encoding determination module 303 includes:
the account total code determining submodule is configured to calculate a weighted sum of total codes of the accounts, which correspond to the preset operations, by using the weight corresponding to each preset operation, and determine the weighted sum of the total codes of the accounts, which correspond to the preset operations, as the total code of the account.
In some embodiments, the overall encoding determination module 303 includes:
and the video total coding determining sub-module is configured to calculate a weighted sum of total codes of the video, which respectively correspond to the preset operations, by using the weight corresponding to each preset operation, and determine the weighted sum of the total codes of the video, which respectively correspond to the preset operations, as the total codes of the video.
With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.
Fig. 4 is a block diagram illustrating a structure of an electronic device according to an example embodiment. Referring to fig. 4, the electronic device includes a processing component 422, which further includes one or more processors, and memory resources, represented by memory 432, for storing instructions, such as application programs, that are executable by the processing component 422. The application programs stored in memory 432 may include one or more modules that each correspond to a set of instructions. Further, the processing component 422 is configured to execute instructions to perform the above-described methods.
The electronic device may also include a power component 426 configured to perform power management of the electronic device, a wired or wireless network interface 450 configured to connect the electronic device to a network, and an input/output (I/O) interface 458. The electronic device may operate based on an operating system stored in memory 432, such as Windows Server, MacOS XTM, UnixTM, LinuxTM, FreeBSDTM, or the like.
In an exemplary embodiment, a storage medium comprising instructions, such as a memory comprising instructions, executable by an electronic device to perform the video recommendation method is also provided. Alternatively, the computer readable storage medium may be a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.
In an exemplary embodiment, the present application further provides a computer program product comprising computer readable code which, when run on an electronic device, causes the electronic device to perform a video recommendation method.
Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the invention and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.
It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the invention is limited only by the appended claims.

Claims (10)

1. A method for video recommendation, the method comprising:
acquiring a topological graph corresponding to each preset operation, wherein the topological graph corresponding to the preset operation indicates the relation between a plurality of accounts and a plurality of videos, and the relation is related to the preset operation;
for the topological graph corresponding to each preset operation, respectively executing a first operation on each account and respectively executing a second operation on each video by using a graph neural network based on the topological graph corresponding to the preset operation, wherein the first operation comprises the following steps: determining a total code of the account corresponding to the preset operation based on codes of the account corresponding to the preset operation and output by each coding layer of the graph neural network, wherein the codes of the account corresponding to the preset operation and output by the coding layers are determined based on the codes of the associated videos of the account corresponding to the preset operation and output by the previous coding layer; the second operation includes: determining a total encoding of the video corresponding to the preset operation based on the encoding of the video corresponding to the preset operation output by each encoding layer, wherein the encoding of the video corresponding to the preset operation output by an encoding layer is determined based on the encoding of an associated account of the video corresponding to the preset operation output by a previous layer of the encoding layer;
for each account, determining the total code of the account based on the total code of the account corresponding to the preset operation, and for each video, determining the total code of the video based on the total code of the video corresponding to the preset operation;
and determining the recommended video corresponding to each account based on the total code of each account and the total code of each video.
2. The method of claim 1, wherein the encoding of the account corresponding to the preset operation output by the encoding layer is determined by:
for each associated video of the account corresponding to the preset operation, determining the weight of the associated video corresponding to the preset operation based on the association degree of the encoding of the associated video corresponding to the preset operation output by the previous layer of the encoding layer and the encoding of each other associated video of the account corresponding to the preset operation output by the previous layer of the encoding layer;
calculating a weighted sum of the encoding of the associated video corresponding to the preset operation and output by the previous layer of the encoding layer by utilizing the weight of the associated video corresponding to the preset operation and output by each account;
and obtaining the coding of the account corresponding to the preset operation, which is output by a coding layer, based on the weighted sum and the number of all the associated videos of the account corresponding to the preset operation.
3. The method of claim 2, wherein determining the weight of the associated video corresponding to the preset operation based on the association degree between the encoding of the associated video corresponding to the preset operation output by the previous layer of the encoding layer and the encoding of each other associated video of the account corresponding to the preset operation output by the previous layer of the encoding layer comprises:
determining an average of the relevance degrees of the coding of the associated video corresponding to the preset operation output by the previous layer of the coding layers and the coding of other associated videos of the account corresponding to the preset operation output by the previous layer of the coding layers as the weight of the associated video corresponding to the preset operation.
4. The method of claim 1, wherein the encoding of the video output by the encoding layer corresponding to the preset operation is determined by:
for each associated account of the video corresponding to the preset operation, determining the weight of the associated account corresponding to the preset operation based on the association degree of the code of the associated account corresponding to the preset operation output by the previous layer of the coding layer and the code of each other associated account of the video corresponding to the preset operation output by the previous layer of the coding layer;
calculating a weighted sum of codes, which are output by a layer before the coding layer and correspond to each associated account of the preset operation, by using the weight of each associated account of the videos corresponding to the preset operation;
and obtaining the code of the video corresponding to the preset operation, which is output by a coding layer, based on the weighted sum and the number of all the associated accounts of the video corresponding to the preset operation.
5. The method of claim 4, wherein determining the weight of the associated account corresponding to the preset operation based on the association degree between the code output by the previous layer of the coding layer and the code output by the previous layer of the coding layer for each other associated account of the video corresponding to the preset operation comprises:
determining an average of the association degrees of the codes of the associated accounts corresponding to the preset operation output by the previous layer of the coding layer and the codes of the other associated accounts corresponding to the preset operation output by the previous layer of the coding layer as the weight of the associated account corresponding to the preset operation.
6. The method of claim 1, wherein determining the total code of the account based on the total code of the account each corresponding to a preset operation comprises:
and calculating the weighted sum of the total codes of the accounts corresponding to the preset operations by using the weight corresponding to each preset operation, and determining the weighted sum of the total codes of the accounts corresponding to the preset operations as the total code of the account.
7. The method of claim 1, wherein determining the overall encoding of the video based on the overall encoding of the video each corresponding to a preset operation comprises:
and calculating a weighted sum of total codes of the video, which correspond to the preset operations, by using the weight corresponding to each preset operation, and determining the weighted sum of the total codes of the video, which correspond to the preset operations, as the total codes of the video.
8. A video recommendation apparatus, characterized in that the apparatus comprises:
the system comprises an acquisition module, a processing module and a display module, wherein the acquisition module is configured to acquire a topological graph corresponding to each preset operation, and the topological graph corresponding to the preset operation indicates the relation between a plurality of accounts and a plurality of videos and relevant to the preset operation;
the encoding module is configured to, for a topological graph corresponding to each preset operation, respectively perform a first operation on each account and respectively perform a second operation on each video by using a graph neural network based on the topological graph corresponding to the preset operation, where the first operation includes: determining a total code of the account corresponding to the preset operation based on codes of the account corresponding to the preset operation and output by each coding layer of the graph neural network, wherein the codes of the account corresponding to the preset operation and output by the coding layers are determined based on the codes of the associated videos of the account corresponding to the preset operation and output by the previous coding layer; the second operation includes: determining a total encoding of the video corresponding to the preset operation based on the encoding of the video corresponding to the preset operation output by each encoding layer, wherein the encoding of the video corresponding to the preset operation output by the encoding layer is determined based on the encoding of the associated account of the video corresponding to the preset operation output by the previous layer of the encoding layer;
a total code determining module configured to determine, for each account, a total code of the account based on the total codes of the accounts corresponding to the preset operations, and determine, for each video, a total code of the video based on the total codes of the videos corresponding to the preset operations;
and the recommended video determining module is configured to determine the recommended video corresponding to each account based on the total code of each account and the total code of each video.
9. An electronic device, comprising:
a processor;
a memory for storing the processor-executable instructions;
wherein the processor is configured to execute the instructions to implement the method of any one of claims 1 to 7.
10. A storage medium having instructions that, when executed by a processor of an electronic device, enable the electronic device to perform the method of any of claims 1-7.
CN202110752995.7A 2021-07-02 2021-07-02 Video recommendation method and device, electronic equipment and storage medium Active CN113569136B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110752995.7A CN113569136B (en) 2021-07-02 2021-07-02 Video recommendation method and device, electronic equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110752995.7A CN113569136B (en) 2021-07-02 2021-07-02 Video recommendation method and device, electronic equipment and storage medium

Publications (2)

Publication Number Publication Date
CN113569136A true CN113569136A (en) 2021-10-29
CN113569136B CN113569136B (en) 2024-03-05

Family

ID=78163562

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110752995.7A Active CN113569136B (en) 2021-07-02 2021-07-02 Video recommendation method and device, electronic equipment and storage medium

Country Status (1)

Country Link
CN (1) CN113569136B (en)

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106126524A (en) * 2016-06-07 2016-11-16 腾讯科技（深圳）有限公司 Information-pushing method and device
CN108694223A (en) * 2018-03-26 2018-10-23 北京奇艺世纪科技有限公司 The construction method and device in a kind of user's portrait library
CN110557659A (en) * 2019-08-08 2019-12-10 北京达佳互联信息技术有限公司 Video recommendation method and device, server and storage medium
CN111198956A (en) * 2019-12-24 2020-05-26 北京达佳互联信息技术有限公司 Multimedia resource interaction method and device, electronic equipment and storage medium
CN111225282A (en) * 2019-12-30 2020-06-02 重庆特斯联智慧科技股份有限公司 Video information recommendation method and system based on big data and readable storage medium
CN111368210A (en) * 2020-05-27 2020-07-03 腾讯科技（深圳）有限公司 Information recommendation method and device based on artificial intelligence and electronic equipment
CN111708948A (en) * 2020-06-19 2020-09-25 北京达佳互联信息技术有限公司 Content item recommendation method, device, server and computer readable storage medium
CN111859136A (en) * 2020-07-23 2020-10-30 深圳前海微众银行股份有限公司 Personalized recommendation method, device, equipment and readable storage medium
CN111932386A (en) * 2020-09-09 2020-11-13 腾讯科技（深圳）有限公司 User account determining method and device, information pushing method and device, and electronic equipment
CN112214636A (en) * 2020-09-21 2021-01-12 华为技术有限公司 Audio file recommendation method and device, electronic equipment and readable storage medium
CN112749296A (en) * 2019-10-31 2021-05-04 北京达佳互联信息技术有限公司 Video recommendation method and device, server and storage medium
CN112925924A (en) * 2019-12-05 2021-06-08 北京达佳互联信息技术有限公司 Multimedia file recommendation method and device, electronic equipment and storage medium

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106126524A (en) * 2016-06-07 2016-11-16 腾讯科技（深圳）有限公司 Information-pushing method and device
CN108694223A (en) * 2018-03-26 2018-10-23 北京奇艺世纪科技有限公司 The construction method and device in a kind of user's portrait library
CN110557659A (en) * 2019-08-08 2019-12-10 北京达佳互联信息技术有限公司 Video recommendation method and device, server and storage medium
CN112749296A (en) * 2019-10-31 2021-05-04 北京达佳互联信息技术有限公司 Video recommendation method and device, server and storage medium
CN112925924A (en) * 2019-12-05 2021-06-08 北京达佳互联信息技术有限公司 Multimedia file recommendation method and device, electronic equipment and storage medium
CN111198956A (en) * 2019-12-24 2020-05-26 北京达佳互联信息技术有限公司 Multimedia resource interaction method and device, electronic equipment and storage medium
CN111225282A (en) * 2019-12-30 2020-06-02 重庆特斯联智慧科技股份有限公司 Video information recommendation method and system based on big data and readable storage medium
CN111368210A (en) * 2020-05-27 2020-07-03 腾讯科技（深圳）有限公司 Information recommendation method and device based on artificial intelligence and electronic equipment
CN111708948A (en) * 2020-06-19 2020-09-25 北京达佳互联信息技术有限公司 Content item recommendation method, device, server and computer readable storage medium
CN111859136A (en) * 2020-07-23 2020-10-30 深圳前海微众银行股份有限公司 Personalized recommendation method, device, equipment and readable storage medium
CN111932386A (en) * 2020-09-09 2020-11-13 腾讯科技（深圳）有限公司 User account determining method and device, information pushing method and device, and electronic equipment
CN112214636A (en) * 2020-09-21 2021-01-12 华为技术有限公司 Audio file recommendation method and device, electronic equipment and readable storage medium

Also Published As

Publication number Publication date
CN113569136B (en) 2024-03-05

Similar Documents

Publication Publication Date Title
KR20180011221A (en) Select representative video frames for videos
CN110348362B (en) Label generation method, video processing method, device, electronic equipment and storage medium
US9092725B2 (en) Classifying attribute data intervals
CN110347872B (en) Video cover image extraction method and device, storage medium and electronic equipment
CN111382255A (en) Method, apparatus, device and medium for question and answer processing
CN112733043B (en) Comment recommendation method and device
CN115471771A (en) Video time sequence action positioning method based on semantic level time sequence correlation modeling
CN115062709A (en) Model optimization method, device, equipment, storage medium and program product
CN113033507B (en) Scene recognition method and device, computer equipment and storage medium
US7895206B2 (en) Search query categrization into verticals
CN112148994B (en) Information push effect evaluation method and device, electronic equipment and storage medium
CN113392321A (en) Information recommendation method and device, electronic equipment and storage medium
CN113886697A (en) Clustering algorithm based activity recommendation method, device, equipment and storage medium
CN113569136B (en) Video recommendation method and device, electronic equipment and storage medium
CN110598127A (en) Group recommendation method and device
CN111126617B (en) Method, device and equipment for selecting fusion model weight parameters
CN114357242A (en) Training evaluation method and device based on recall model, equipment and storage medium
CN110502715B (en) Click probability prediction method and device
CN112801076A (en) Electronic commerce video highlight detection method and system based on self-attention mechanism
CN113724023B (en) Media resource pushing method and device, electronic equipment and storage medium
CN117058489B (en) Training method, device, equipment and storage medium of multi-label recognition model
CN113705873B (en) Construction method of film and television work score prediction model and score prediction method
CN117194772B (en) Content pushing method and device based on user tag
CN113473124B (en) Information acquisition method, device, electronic equipment and storage medium
CN117351279A (en) Self-distillation realization method for space-time distillation fusion

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant