CN111491202B - Video publishing method, device, equipment and storage medium - Google Patents

Video publishing method, device, equipment and storage medium Download PDF

Info

Publication number
CN111491202B
CN111491202B CN201910087567.XA CN201910087567A CN111491202B CN 111491202 B CN111491202 B CN 111491202B CN 201910087567 A CN201910087567 A CN 201910087567A CN 111491202 B CN111491202 B CN 111491202B
Authority
CN
China
Prior art keywords
video
attraction
training
candidate
cover
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.)
Active
Application number
CN201910087567.XA
Other languages
Chinese (zh)
Other versions
CN111491202A (en
Inventor
张树业
王俊东
张壮辉
梁德澎
岑洪杰
梁柱锦
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.)
Bigo Technology Pte Ltd
Original Assignee
Guangzhou Baiguoyuan 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 Guangzhou Baiguoyuan Information Technology Co Ltd filed Critical Guangzhou Baiguoyuan Information Technology Co Ltd
Priority to CN201910087567.XA priority Critical patent/CN111491202B/en
Priority to PCT/CN2020/072191 priority patent/WO2020156171A1/en
Publication of CN111491202A publication Critical patent/CN111491202A/en
Application granted granted Critical
Publication of CN111491202B publication Critical patent/CN111491202B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
    • H04N21/44—Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to encoded video stream scene graphs
    • H04N21/44008—Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to encoded video stream scene graphs involving operations for analysing video streams, e.g. detecting features or characteristics in the video stream
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/40—Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
    • G06F16/43—Querying
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25—Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/258—Client or end-user data management, e.g. managing client capabilities, user preferences or demographics, processing of multiple end-users preferences to derive collaborative data
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
    • H04N21/44—Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to encoded video stream scene graphs
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H04N21/4662—Learning process for intelligent management, e.g. learning user preferences for recommending movies characterized by learning algorithms
    • H04N21/4666—Learning process for intelligent management, e.g. learning user preferences for recommending movies characterized by learning algorithms using neural networks, e.g. processing the feedback provided by the user
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/80—Generation or processing of content or additional data by content creator independently of the distribution process; Content per se
    • H04N21/85—Assembly of content; Generation of multimedia applications
    • H04N21/854—Content authoring
    • H04N21/8549—Creating video summaries, e.g. movie trailer

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Computer Security & Cryptography (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computer Graphics (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a video distribution method, a video distribution device, video distribution equipment and a storage medium. Wherein, the method comprises the following steps: acquiring two or more video frames serving as candidate covers in a video to be published; respectively inputting the single candidate cover into different types of pre-constructed neural network models to obtain the characteristic information of the candidate cover under different attraction factors; determining a feature vector of the candidate cover according to the feature information of the candidate cover under different attraction factors; and determining the video cover of the video to be published according to the feature vectors corresponding to the two or more candidate covers. The technical scheme provided by the embodiment of the invention realizes the intelligent determination of the video cover in the video to be published, solves the problem that the diversified clicking and watching requirements of different users cannot be met by selecting the video cover through a single factor, has greater attraction to the users at the moment, and improves the clicking rate and the watching times of the video to be published after being published.

Description

Video publishing method, device, equipment and storage medium
Technical Field
The embodiment of the invention relates to the technical field of internet, in particular to a video publishing method, device, equipment and storage medium.
Background
With the rapid development of internet technology, short video tools or platforms attract a large number of mobile internet users and also occupy huge traffic. Users can use the short video tools or platforms to release a large number of short videos, so that the purposes of entertainment, sharing or communication and the like are achieved.
At present, before a user publishes a short video on a short video tool or a platform, a certain video frame is carefully selected from the short video as a cover page of the short video, but due to the use habits or operation and other reasons of the user, a great number of users skip the step of manually selecting the cover page, and directly publish the short video, and at the moment, the client usually adopts some simple strategies to determine the cover page, such as directly selecting the first frame.
Data analysis shows that the self-selected cover of the short video can bring higher click rate and more watching times than the default cover, namely, the cover generated by a certain strategy can further improve the click rate of the short video and increase the watching times of the short video by a user. However, because the amount of videos released every day is huge, all short videos cannot be selected manually, and a large amount of labor cost is consumed; meanwhile, a simple cover determining strategy only selects and determines the video cover in a single mode, and diversified clicking and watching requirements of different users cannot be met.
Disclosure of Invention
The embodiment of the invention provides a video publishing method, a video publishing device, video publishing equipment and a storage medium, which can be used for intelligently determining a video cover in a video to be published, improving the association degree of the video cover and the user preference and improving the click rate and the watching times of the video to be published after being published.
In a first aspect, an embodiment of the present invention provides a video publishing method, where the method includes:
acquiring two or more video frames serving as candidate covers in a video to be published;
respectively inputting a single candidate cover into different types of pre-constructed neural network models to obtain characteristic information of the candidate cover under different attraction factors, wherein the attraction factors represent factors for attracting a user to click or watch a video to be published;
determining a feature vector of the candidate cover according to the feature information of the candidate cover under different attraction factors;
and determining the video cover of the video to be published according to the feature vectors corresponding to two or more candidate covers.
In a second aspect, an embodiment of the present invention provides a video distribution apparatus, including:
the candidate cover acquiring module is used for acquiring two or more video frames serving as candidate covers in the video to be published;
the characteristic information determining module is used for respectively inputting the single candidate cover into different types of pre-constructed neural network models to obtain the characteristic information of the candidate cover under different attraction factors, wherein the attraction factors represent factors for attracting a user to click or watch a video to be published;
the characteristic vector determining module is used for determining the characteristic vector of the candidate cover according to the characteristic information of the candidate cover under different attraction factors;
and the video cover determining module is used for determining the video cover of the video to be published according to the feature vectors corresponding to two or more candidate covers.
In a third aspect, an embodiment of the present invention provides an apparatus, where the apparatus includes:
one or more processors;
storage means for storing one or more programs;
when executed by the one or more processors, cause the one or more processors to implement the video distribution method described in any embodiment of the present invention.
In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements the video distribution method described in any embodiment of the present invention.
The embodiment of the invention provides a video publishing method, a device, equipment and a storage medium, wherein two or more video frames are acquired from a video to be published as candidate covers, the single candidate cover is respectively input into neural network models corresponding to different attraction factors, obtaining the characteristic information under different attraction factors, generating a plurality of characteristic vectors under the attraction factor dimensionality, according to the feature vectors of the candidate covers, the corresponding video covers are determined, the association degree between the video covers and the user preferences corresponding to the attraction factors is improved, the video covers do not need to be selected manually, the intelligent determination of the video covers in the video to be published is achieved, the problem that the diversified clicking and watching requirements of different users cannot be met due to the fact that the video covers are selected through a single factor is solved, the video covers have great attraction to the users at the moment, and the clicking rate and the watching times of the video to be published after being published are improved.
Drawings
Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments made with reference to the following drawings:
fig. 1A is a flowchart of a video publishing method according to an embodiment of the present invention;
fig. 1B is a schematic diagram of a video distribution process according to an embodiment of the present invention;
FIG. 2 is a schematic diagram illustrating a process of constructing different types of neural network models and ranking models according to a second embodiment of the present invention;
fig. 3 is a schematic diagram of a neural network model training process according to a third embodiment of the present invention;
fig. 4 is a schematic diagram of a video distribution process according to a fourth embodiment of the present invention;
fig. 5 is a schematic structural diagram of a video distribution apparatus according to a fifth embodiment of the present invention;
fig. 6 is a schematic structural diagram of an apparatus according to a sixth embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be further noted that, for the convenience of description, only some of the structures related to the present invention are shown in the drawings, not all of the structures.
The embodiment of the invention mainly aims at different factors for attracting users to click or watch videos in the video covers, analyzes the characteristic information of the candidate covers under different attraction factors so as to describe the characteristics of each candidate cover from a plurality of dimensions corresponding to different attraction factors, further comprehensively analyzes the multi-dimensional characteristics of each candidate cover in the video to be published so as to meet diversified click and watch requirements of different users, obtains the video cover with the largest attraction for the users to click or watch the videos to be published, and improves the click rate and watch times of the videos to be published after being published.
Example one
Fig. 1A is a flowchart of a video publishing method according to an embodiment of the present invention, where the embodiment is applicable to any intelligent terminal configured with a video application capable of publishing various videos. The scheme of the embodiment of the invention can be suitable for the problem of how to intelligently determine the video cover of the video to be published. The video distribution method provided by this embodiment may be executed by the video distribution apparatus provided by the embodiment of the present invention, the apparatus may be implemented in a software and/or hardware manner, and is integrated in a device for executing the method, where the device may be any intelligent terminal configured with a video application program capable of distributing various videos, such as a smart phone, a tablet computer, or a notebook computer.
Specifically, referring to fig. 1A, the method may include the steps of:
s110, two or more video frames serving as candidate covers in the video to be published are obtained.
The video to be published means that any user can publish the video to the internet for wide spread through various video application programs generated in advance, so that video data corresponding to network social contact is realized; such as short video recorded by the user, live webcast video, etc. Specifically, when a user selects a certain video from a large number of videos existing in the internet to click or watch, the user selects a video with a large attraction according to the cover of each video, and ignores other videos with a small attraction.
In this embodiment, a certain video frame in a video to be published is used as a video cover of the video to be published, at this time, since a large number of video frames are included in one video to be published, and repeated video frames are included in continuous video frames within a certain time period, if a certain specific video frame is selected as a video cover by processing all video frames, the processing amount of the video frame is too large at this time, and a large number of repeated operations on similar video frames exist, which results in a long duration and low efficiency of a selection process of the video cover, so that in this embodiment, a corresponding video cover is obtained by processing a candidate cover, and the determination efficiency of the video cover is improved. The candidate cover refers to a video frame which meets a certain condition and can represent the characteristics of similar video frames in different types in all video frames contained in the video to be published, for example, the video frame with the best image effect in the similar video frames in the different types.
Optionally, when determining the video cover of the video to be published, first, two or more video frames which meet a certain condition and can represent the characteristics of various similar video frames are selected from all the video frames included in the video to be published and used as candidate covers of the video to be published, and then the candidate covers are processed to determine the video cover of the video to be published, so that the processing number of the video frames is reduced, and the processing efficiency is improved.
And S120, respectively inputting the single candidate cover into the pre-constructed neural network models of different types to obtain the characteristic information of the candidate cover under different attraction factors.
Wherein the attraction factor represents a factor that attracts a user to click on or view a video to be distributed. In particular, since the attraction factors in the video to be published are relatively complex, for example: for the same user, the video covers of different videos to be published attract the user to click or watch different factors, such as novelty or originality of the display content of the cover, the stress and stimulation degree of the display content, the age, the color value and the gender of the people of the cover, or the image definition; for the same video to be published, the first subjective feelings of different users are different, as if the factor of the same video attracting the first user to click or watch is possibly the character color value in the cover, and the factor attracting the second user to click or watch is possibly the scenario revealed in the display content of the cover; therefore, in this embodiment, for each attraction factor, a corresponding neural network model may be trained for the attraction factor in advance, where the neural network model is a deep machine learning model, and corresponding training parameters may be set for the neural network model in advance, and the training parameters initially set in the neural network model are optimally trained through a large number of video frames of historical videos, so that the neural network model has a certain feature extraction capability, and for each candidate cover, feature information of the candidate cover under the attraction factor can be accurately extracted.
In this embodiment, for different attraction factors, neural network models in the types corresponding to the attraction factors can be trained in advance, so that neural network models of different types are constructed in advance. It should be noted that, in this embodiment, the attraction factor corresponding to each neural network model is a pre-mined factor that has a positive influence on the user's clicking or watching video, and can reflect the degree of influence of the video content plots contained in the candidate cover on the number of times of attracting the user's clicking or watching to some extent, and combine the click rate or observation number of the user with the internal relationship between the content plots contained in the video cover to accurately measure the attraction of each candidate cover to the user.
Optionally, when two or more candidate covers in the video to be published are obtained, in order to determine the attraction of each candidate cover to the user for clicking or watching the video to be published, as shown in fig. 1B, a single candidate cover may be respectively input into different types of neural network models pre-constructed in this embodiment, and the different types of neural network models perform parallel processing on the candidate cover, so as to obtain feature information of the candidate cover output by the different types of neural network models under corresponding different attraction factors; according to the process, the same processing is carried out on each candidate cover, and the characteristic information of each candidate cover under different attraction factors is obtained. In this embodiment, the candidate covers are processed by different types of neural network models, and may be attraction scores of the same candidate cover for different attraction factors through the different types of neural network models, and the feature information of the candidate cover under different attraction factors is represented by the attraction scores, so that the influence degree of the candidate cover on respectively attracting the user to click or watch the video to be published under different single attraction factors is determined.
S130, determining the feature vector of the candidate cover according to the feature information of the candidate cover under different attraction factors.
Specifically, when obtaining the feature information of each candidate cover under different attraction factors, in order to more completely and accurately judge the attraction of each candidate cover to the video to be published for clicking or watching, the feature information of the candidate cover under different attraction factors may be subjected to corresponding feature processing, so as to obtain a feature vector of the candidate cover represented by multiple dimensions corresponding to different attraction factors, where the feature vector may describe the attraction of the candidate cover to the video to be published for a user to click or watch from multiple dimensions, and then the feature vector of each candidate cover is processed, and the attraction of each candidate cover to the video to be published for the user to click or watch is judged from multiple dimensions.
S140, determining the video cover of the video to be published according to the feature vectors corresponding to the two or more candidate covers.
The video cover refers to an image which is displayed in various video application programs and represents the video to be published after the video to be published is published on the line.
Specifically, after the feature vectors of each candidate cover are obtained, the feature vectors of two or more candidate covers may be analyzed respectively, that is, feature information corresponding to different attraction factors of the same candidate cover under multiple dimensions is subjected to fusion analysis, and the attraction of the candidate cover to the user clicking or watching the video to be published is comprehensively judged from the perspective of multiple attraction factors, so that the candidate cover with the highest attraction to the user clicking or watching the video to be published is selected from each candidate cover and used as the final video cover of the video to be published.
Optionally, in order to accurately measure the attraction of each candidate cover to the user for clicking or watching the video, in this embodiment, determining the video cover of the video to be published according to the feature vectors corresponding to two or more candidate covers may specifically include: inputting the feature vectors corresponding to the candidate covers into a pre-constructed ranking model to obtain ranking scores of the candidate covers; and determining the video cover of the video to be published according to the ranking score of the candidate cover.
Specifically, the ranking model is a machine learning model, training parameters and neuron structures in the model are trained by adopting a large number of training samples, and the ranking model has certain attraction judgment capability, so that the attraction of candidate covers to a user for clicking or watching a video to be published is accurately judged according to multi-dimensional characteristic vectors expressed by the candidate covers through different attraction factors; the ranking score refers to an attraction score for a user to click on or view a video to be published.
Optionally, in this embodiment, the feature vectors of the candidate covers are obtained through the feature information of the candidate covers under different attraction factors, at this time, the feature vectors of the candidate covers are respectively input into a ranking model which is constructed in advance through a large number of training samples, as shown in fig. 1B, the ranking model performs fusion analysis on the feature information of the candidate covers under different attraction factors corresponding to multiple dimensions contained in the feature vectors of the candidate covers, and obtains a ranking score representing the attraction of the candidate covers to the user for clicking or watching the video to be published; when the obtained ranking scores of the two or more candidate covers are obtained, the candidate cover with the highest ranking score is selected from the candidate covers to serve as the video cover of the video to be published, at the moment, the attraction of the video cover to a user for clicking or watching the video to be published is the highest, and the click rate and the watching times of the video to be published after being published are improved.
In the technical scheme provided by the embodiment, two or more video frames are acquired from the video to be published as candidate covers, the single candidate cover is respectively input into the neural network models corresponding to different attraction factors, obtaining the characteristic information under different attraction factors, generating a plurality of characteristic vectors under the attraction factor dimensionality, according to the feature vectors of the candidate covers, the corresponding video covers are determined, the association degree between the video covers and the user preferences corresponding to the attraction factors is improved, the video covers do not need to be selected manually, the intelligent determination of the video covers in the video to be published is achieved, the problem that the diversified clicking and watching requirements of different users cannot be met due to the fact that the video covers are selected through a single factor is solved, the video covers have great attraction to the users at the moment, and the clicking rate and the watching times of the video to be published after being published are improved.
Example two
Fig. 2 is a schematic diagram of a process for constructing different types of neural network models and ranking models according to a second embodiment of the present invention. The embodiment is optimized on the basis of the embodiment. Specifically, as shown in fig. 2, the embodiment mainly mines attraction factors in the video cover that affect the user to click or watch the video, so as to explain the building process of different types of neural network models and ranking models in detail.
Optionally, the method in this embodiment may specifically include the following steps:
s210, two or more preset training attraction factors are obtained.
The training attraction factor refers to factors which are preset according to habits of various users or preferences of the users and possibly influence the user to click or watch videos, factors which have forward influence on the user to click or watch the videos exist, and factors which have reverse influence on the user to click or watch the videos exist, for example, the color value of a cover page task may be the factor having the forward influence, and terrorist bloody smell contents displayed in the cover page may be the factor having the reverse influence.
Specifically, in this embodiment, when mining the attraction factors having a positive influence on the user's video click or viewing in the cover, two or more training attraction factors preset according to each user habit, user preference, or the like may be first obtained, then the corresponding video cover is obtained for a single training attraction factor in sequence, and the attraction of the video cover corresponding to the single training attraction factor to the user's video click or viewing is determined, so that the training attraction factors having a positive influence on the user's video click or viewing are screened out from the two or more training attraction factors, and thus different types of neural network models required in this embodiment are constructed.
And S220, sequentially acquiring click index quantities of the historical videos corresponding to the single training attraction factors after online release.
The click index is an index which can clearly reflect the attraction of a video envelope in the historical video to a user for clicking or watching the historical video; the click index quantity refers to the click rate or the watching frequency of a user for clicking or watching the historical video after the historical video is published on line according to the corresponding video cover.
Optionally, when two or more training attraction factors are obtained, the click index amount of the user clicking or watching the video corresponding to a single training attraction factor needs to be tested, and at this time, the click index amount corresponding to a single training attraction factor needs to be obtained first. Specifically, in this embodiment, a multi-round test mode is adopted to respectively obtain click index quantities corresponding to a single training attraction factor, a test sequence of the training attraction factor is first determined, a video cover of a historical video is specially re-determined for the single training attraction factor, and the historical video is distributed on line according to the video cover determined at this time, so that a round of click index quantity tests of the historical video is performed on the training attraction factor.
For example, since a new attraction of a single training attraction factor to a user for clicking or watching a video needs to be determined, an initial video cover of a historical video without any training attraction factor participating in determination needs to be obtained first, a click index amount obtained by issuing the historical video on line according to the video cover serves as a reference basis for subsequently determining the attraction of each training attraction factor, at this time, in a first click index amount test, any neural network model does not need to be set, a video frame is randomly selected from the historical video in a random sampling mode to serve as a corresponding video cover in the first test, at this time, the historical video is issued on line according to the video cover of the first test, and a click index amount of the first test is obtained.
In the second round of click index quantity test, the training attraction factor for the round of test can be selected according to the test sequence of the predetermined training attraction factor, the neural network model pre-trained on the training attraction factor for the round of test is obtained, so that two or more candidate covers selected from all video frames of the historical video in advance are input into the neural network model, the characteristic information of the candidate covers under the training attraction factor is obtained, the characteristic information is input into the ranking model, the ranking score of each candidate cover is obtained, the corresponding video cover in the second round of test is further determined, at the moment, the historical video is published on line according to the video cover corresponding to the second round of test, and the click index quantity of the second round of test is obtained.
In the third click index quantity test, the training attraction factor aimed by the test is selected again, the neural network model pre-trained by the training attraction factor aimed by the test is obtained, the neural network models corresponding to the training attraction factor tested by the front wheel are constructed together, two or more candidate covers selected from all video frames of the historical video in advance are input into the neural network models of different types pre-constructed by the test, the characteristic information of the candidate covers under different training attraction factors corresponding to the test and the front wheel test in the current test is obtained, the characteristic vector of the candidate covers is further obtained, the characteristic vector is input into the ranking model, the ranking score of each candidate cover is obtained, and the corresponding video cover in the third test is further determined, at the moment, the historical video is published on line according to the video cover corresponding to the third test to obtain a click index quantity of the third test; sequentially carrying out cyclic test on the click index quantity of the rear wheel according to the test process of the wheel to obtain the click index quantity corresponding to the training attraction factor in each click index quantity test; and until all the training attraction factors are subjected to corresponding click index quantity tests in a multi-round testing mode, determining the training attraction factors which have positive influence on the user to click or watch the video, and effectively mining the attraction factors in the video cover.
In addition, in the process of testing the multi-round click index amount, when the feature vectors of the historical videos are input into the ranking model, the ranking model can also be trained through a large number of historical videos, the feature vectors of the historical videos are determined through feature information under different training attraction factors obtained from different types of neural network models, and then the feature vectors are input into the ranking model for training, so that the accuracy of the ranking score output by the ranking model is improved. Optionally, in this embodiment, the ranking model may adopt Machine learning models such as a Multilayer Perceptron (MLP) and a Support Vector Machine (SVM), and any loss function may be adopted for training in the training process. Specifically, the loss function is defined as:
Figure BDA0001962268940000121
wherein the content of the first and second substances,
Figure BDA0001962268940000122
ranking scores output by the ranking model when a certain candidate cover is tested in the current round of click index quantity;
Figure BDA0001962268940000123
ranking scores output by the ranking model in the previous round of click index quantity test for the same candidate cover; required in this example
Figure BDA0001962268940000124
The positive influence of the training attraction factor of the current round of test on the user clicking or watching the video is ensured. Delta is an adjustable super parameter preset during model training and is a positive numberAnd the optimal super parameter is selected for machine learning, so that the performance of the machine learning is improved. At this time, in the present embodiment
Figure BDA0001962268940000125
And
Figure BDA0001962268940000126
is shown by setting the penalty measure of the loss function to two stages, u ≦ δ and u, respectively>δ two stages.
Wherein the content of the first and second substances,
Figure BDA0001962268940000127
the penalty degree of the loss function in the first stage is represented;
Figure BDA0001962268940000128
the penalty degree of the loss function in the second stage is represented;
Figure BDA0001962268940000129
the judgment indexes of the first stage and the second stage are obtained;
at this time, the process of the present invention,
Figure BDA00019622689400001210
representing the magnitude of the difference between the ranking scores of two adjacent rounds, at u<At time 0, description
Figure BDA00019622689400001211
Far greater than
Figure BDA00019622689400001212
Meet the training requirement, at this time lrankNo penalty is required when being equal to 0; when u.ltoreq.delta is 0. ltoreq. u.ltoreq.
Figure BDA0001962268940000131
Albeit greater than
Figure BDA0001962268940000132
But the difference is small, in order to meet the accuracy of the sequencing model, punishment with small strength can be carried out, and l is carried out at the momentrank＝0.5u2Since u is 0. ltoreq. u.ltoreq.delta, 0.5u2The smaller punishment degree is ensured; at u>Delta. time, description
Figure BDA0001962268940000133
Is less than
Figure BDA0001962268940000134
Not meeting the training requirement, when lrank＝δu-0.5δ2A penalty of greater strength can be made at this stage; therefore, training of the ranking model is achieved through the loss function, and accuracy of the ranking score output by the ranking model is guaranteed.
S230, judging whether the click index quantity corresponding to the current training attraction factor is larger than the click index quantity corresponding to the last training attraction factor or not, if so, executing S240; if not, go to S250.
Specifically, when the click index quantity corresponding to the currently trained attraction factor in each round of test is obtained through a multi-round test mode, in order to judge the attraction of the currently trained attraction factor, the click index quantity corresponding to the currently trained attraction factor may be compared with the click index quantity corresponding to the last trained attraction factor obtained when the neural network model corresponding to the currently trained attraction factor is not added in the previous round of click index quantity test, that is, the influence of the video cover determined when the neural network model corresponding to the currently trained attraction factor is added and the influence of the video cover determined before the addition on the click index quantity of the same historical video are judged, so that the attraction of the currently trained attraction factor on the user click or the video watching is judged.
S240, taking the neural network model corresponding to the current training attraction factor as one of the components of the neural network models of different types which are constructed in advance.
Optionally, if the click index amount corresponding to the current training attraction factor is larger than the click index amount corresponding to the previous training attraction factor, it indicates that after the historical video is published on line according to the video cover obtained after the neural network model corresponding to the current training attraction factor is added, the number of the users clicking or watching the historical video is increased, that is, the current training attraction factor has a positive influence on the users clicking or watching the historical video, at this time, the neural network model corresponding to the training attraction factor is used as one of the pre-constructed neural network models of different types, and the click index amount test of the next training attraction factor is continuously executed.
And S250, discarding the neural network model corresponding to the current training attraction factor.
Optionally, if the click index amount corresponding to the current training attraction factor is less than or equal to the click index amount corresponding to the previous training attraction factor, it indicates that after the historical video is published on line according to the video cover obtained after the neural network model corresponding to the current training attraction factor is added, the number of the historical videos clicked or watched by the user is reduced or unchanged, that is, the current training attraction factor does not have a forward effect on the user clicking or watching the historical video, and may also have a reverse effect, and at this time, the neural network model corresponding to the training attraction factor is directly discarded instead of being used as a component of the pre-constructed neural network models of different types, and the click index amount test of the next training attraction factor is continuously performed. Optionally, the currently discarded neural network model may not be used as different types of neural network models constructed in the next round of click index quantity test, that is, when the neural network model corresponding to the currently trained attraction factor is discarded in the current round of test, that is, the current round of test fails, at this time, when the next trained attraction factor is used as the currently trained attraction factor to be tested, the click index quantity corresponding to the training attraction factor that fails to pass the test is excluded, and the click index quantity corresponding to the training attraction factor that passes the test in the previous round is compared and judged, so that the effectiveness of the mined attraction factor is improved.
And S260, taking the next training attraction factor as the current training attraction factor, and continuously executing S230 until the training attraction factors are traversed to obtain different types of pre-constructed neural network models.
Optionally, after the attraction of the current training attraction factor to the user click or view the video is judged, the attraction of the next training attraction factor to the user click or view the video is continuously judged in a multi-round testing mode, that is, the next training attraction factor is used as the current training attraction factor, the judging process is continuously executed until the training attraction factor is traversed, and the neural network models of different types which are constructed in advance are obtained. Specifically, the training attraction factors are set in this embodiment, which may be to excavate the training attraction factors for multiple times in a multi-round click indicator quantity test process, and implement an iterative attraction factor excavation process through a multi-round click indicator quantity test until the click indicator quantity corresponding to the training attraction factors in successive multi-round click indicator quantity tests is less than or equal to the click indicator quantity in the previous round of tests, which indicates that no attraction factor with a forward influence can be excavated, at this time, excavation of the attraction factors is stopped, and in the multi-round click indicator quantity test, different types of neural network models constructed by the neural network models corresponding to the positively influenced attraction factors, which are required in this embodiment, are obtained.
According to the technical scheme provided by the embodiment, the click index quantity after the historical video corresponding to the current training attraction factor is issued on line is judged to be different from the click index quantity corresponding to the last training attraction factor, so that the attraction factor which has positive influence on the user clicking or watching the video is effectively mined in the training attraction factors, the neural network model corresponding to the attraction factor is constructed into the finally trained neural network models of different types, the attraction of the video envelope to the user is effectively improved, and the click rate and the watching times of the video to be issued after being issued are improved.
EXAMPLE III
Fig. 3 is a schematic diagram of a principle of a neural network model training process according to a third embodiment of the present invention. The embodiment is optimized on the basis of the embodiment. Specifically, in this embodiment, a detailed explanation is mainly given to a training process of off-line training a neural network model corresponding to different attraction factors.
Optionally, this embodiment may specifically include the following steps:
s310, two or more preset training attraction factors and two or more historical video frames serving as historical candidate covers in the historical videos corresponding to the single training attraction factor are obtained, and feature labels of the historical candidate covers under the corresponding training attraction factors are determined.
Specifically, when training the neural network model corresponding to each training attraction factor, first, two or more preset training attraction factors and two or more history video frames serving as history candidate covers in the history video corresponding to a single training attraction factor need to be obtained, the history candidate covers in the history video are taken as training samples, and feature labels of the history candidate covers under the corresponding training attraction factors are determined, where the feature labels may be scores of the history candidate covers under the corresponding training attraction factors marked in advance.
S320, inputting the single historical candidate cover into a preset model corresponding to the training attraction factor to obtain historical characteristic information of the historical candidate cover under the training attraction factor.
Specifically, when two or more candidate covers in the history video corresponding to a single training attraction factor are obtained, a single history candidate cover is directly input into the preset model set for the training attraction factor in the embodiment, at this time, the preset model is trained according to the feature information of the history candidate cover under the training attraction factor, the input history candidate cover is analyzed according to the relationship between the training parameters in the model and each neuron structure, the feature information of the history candidate cover under the training attraction factor is determined, so that the feature information is compared with the feature label of the history candidate cover under the training attraction factor, the training parameters and the neuron structure in the preset model are optimized according to the comparison result, and the preset model is subjected to iterative training.
S330, determining corresponding training loss according to historical characteristic information and characteristic labels of the historical candidate covers under the training attraction factors, adjusting parameters of a preset model by adopting an optimization method of random gradient descent to enable the training loss to be smaller than a set loss threshold value, and finally taking the latest preset model as a neural network model corresponding to the training attraction factors.
Specifically, when the feature information of the historical candidate cover under a single training attraction factor is obtained, the feature information is a pre-estimated value, at this time, the feature information is compared with the feature label of the historical candidate cover under the training attraction factor, that is, the pre-estimated value and the actual value of the feature information of the historical candidate cover are compared, so that the training loss existing in the preset model of the current training is determined, and the training loss can definitely indicate the accuracy of the preset model of the current training for feature extraction in the historical candidate cover. Optionally, in this embodiment, any existing loss function may be used to determine the training loss of the training, which is not limited herein. Meanwhile, when training loss exists when a plurality of historical candidate covers trained in batch is obtained as training samples, the training loss needs to be judged by adopting an optimization method with a random gradient descending, if the training loss of the training is more than or equal to a set loss threshold, the accuracy of the preset model of the training on feature extraction in the historical candidate covers is not high, and the training needs to be carried out again, so that parameters in the preset model are adjusted, and the training loss is correspondingly reduced; further obtaining a plurality of historical candidate covers of the next batch as training samples, determining corresponding training losses again, adjusting parameters in the preset model again by adopting an optimization method of random gradient descent, and circulating in sequence until the obtained training losses are smaller than a set loss threshold value, indicating that the preset model of the current training has reached certain accuracy in feature extraction of the historical candidate covers, and not needing to train again, and taking the current latest preset model as a neural network model corresponding to the training attraction factor; and training the neural network model corresponding to each training attraction factor according to the process to obtain different types of neural network models corresponding to different attraction factors. The optimization method of the random gradient descent is a very wide optimization algorithm used in machine learning, the optimization of the random gradient descent mainly aims at minimizing a loss function of each sample, at the moment, although the loss function obtained by each iteration is not towards the global optimal direction, the direction of the overall iteration is towards the global optimal solution, the final result is often near the global optimal solution, and therefore the obtained latest preset model can achieve certain accuracy for feature extraction in the historical candidate cover.
According to the technical scheme provided by the embodiment, the neural network model corresponding to the training attraction factor is trained through the historical characteristic information and the characteristic label of the historical candidate cover under the single training attraction factor, the accuracy of acquiring the characteristic information by the neural network model corresponding to each training attraction factor is improved, the association degree between the candidate cover characteristic and the user preference corresponding to each attraction factor is analyzed, and the attraction of the video cover to the user is effectively improved.
Example four
Fig. 4 is a schematic diagram of a video distribution process according to a fourth embodiment of the present invention. The embodiment is optimized on the basis of the embodiment. Specifically, as shown in fig. 4, the present embodiment mainly explains the acquisition process of the candidate cover page and the video distribution process in detail.
Optionally, this embodiment may specifically include the following steps:
s410, obtaining an initial video frame meeting preset conditions in the video to be issued.
Specifically, when the candidate cover pages in the video to be published are obtained, the preset condition that the initial screening should be performed on all the video frames may be determined, where the preset condition may be a condition that dark, fuzzy, or solid-color video frames can be filtered out from all the video frames, at this time, the initial video frames meeting the preset condition are screened out from all the video frames included in the video to be published, and then, the corresponding candidate cover pages are continuously screened out from the initial video frames.
And S420, processing the initial video frame by adopting a clustering algorithm to obtain two or more clustering sets.
Specifically, when an initial video frame in a video to be published is obtained, the initial video frame may be processed by using a clustering algorithm, so that video frames with similar characteristics in the initial video frame are classified into one type, and two or more cluster sets are obtained, where each cluster set may include at least one initial video frame.
And S430, selecting a target video frame from the clustering set as a candidate cover of the video to be published.
Optionally, after two or more cluster sets are obtained, one target video frame may be selected from each cluster set to serve as a candidate cover corresponding to the cluster set, so as to obtain two or more candidate covers in the video to be published, at this time, repeated video frames may be excluded, repeated operations on pictures with similar video frames are reduced, and accuracy of the video covers in the video to be published is improved.
S440, inputting the single candidate cover into the pre-constructed neural network models of different types respectively to obtain the characteristic information of the candidate cover under different attraction factors.
S450, carrying out normalization processing on the feature information of the candidate cover under different attraction factors to obtain the feature vector of the candidate cover.
Optionally, when obtaining the feature information of the candidate cover under different attraction factors, the feature information is directed at different features under different dimensions, and at this time, considering the dimension problem of each dimension, the feature information of the candidate cover under different attraction factors may be preprocessed in a normalization manner, so as to obtain a feature vector of the candidate cover. Illustratively, if the characteristic information of each candidate cover under different attraction factors is
Figure BDA0001962268940000191
Wherein n is the serial number mark of the candidate cover, t is the serial number mark of the attraction factor,
Figure BDA0001962268940000192
characteristic information of the nth candidate cover under the t attraction factor is represented; at the moment, the feature information of the candidate cover under different attraction factors is normalized through the following formula:
Figure BDA0001962268940000193
wherein the content of the first and second substances,
Figure BDA0001962268940000194
n is the number of candidate covers;
Figure BDA0001962268940000195
the feature vector of each candidate cover after normalization is obtained at the moment
Figure BDA0001962268940000196
And S460, determining the video cover of the video to be published according to the feature vectors corresponding to the two or more candidate covers.
And S470, publishing the video to be published on line according to the video cover.
Optionally, when the video cover of the video to be published is determined, the video to be published is published on line according to the video cover, so that the click rate and the watching times of the video to be published are improved.
According to the technical scheme provided by the embodiment, the initial video frames meeting the preset conditions in the video to be published are clustered, the target video frames are respectively selected from the obtained clustering sets to serve as the candidate cover of the video to be published, the processing quantity of the video frames determined by the video cover is reduced, the determination rate of the video cover is improved, the corresponding video to be published is published on line according to the video cover, and the click rate and the watching times of the video to be published after being published are improved.
EXAMPLE five
Fig. 5 is a schematic structural diagram of a video distribution apparatus according to a fifth embodiment of the present invention, specifically, as shown in fig. 5, the apparatus may include:
a candidate cover acquiring module 510, configured to acquire two or more video frames serving as candidate covers in a video to be published;
the characteristic information determining module 520 is configured to input the single candidate cover into different types of neural network models which are constructed in advance, so as to obtain characteristic information of the candidate cover under different attraction factors, where the attraction factors represent factors for attracting a user to click or watch a video to be published;
a feature vector determining module 530, configured to determine a feature vector of the candidate cover according to feature information of the candidate cover under different attraction factors;
the video cover determining module 540 is configured to determine a video cover of the video to be published according to the feature vectors corresponding to two or more candidate covers.
In the technical scheme provided by the embodiment, two or more video frames are acquired from the video to be published as candidate covers, the single candidate cover is respectively input into the neural network models corresponding to different attraction factors, obtaining the characteristic information under different attraction factors, generating a plurality of characteristic vectors under the attraction factor dimensionality, according to the feature vectors of the candidate covers, the corresponding video covers are determined, the association degree between the video covers and the user preferences corresponding to the attraction factors is improved, the video covers do not need to be selected manually, the intelligent determination of the video covers in the video to be published is achieved, the problem that the diversified clicking and watching requirements of different users cannot be met due to the fact that the video covers are selected through a single factor is solved, the video covers have great attraction to the users at the moment, and the clicking rate and the watching times of the video to be published after being published are improved.
Further, the video distribution apparatus may further include:
the training factor acquisition module is used for acquiring two or more preset training attraction factors;
and the network model building module is used for taking the neural network model corresponding to the current training attraction factor as one of the components of the pre-built different types of neural network models until the training attraction factor is traversed to obtain the pre-built different types of neural network models if the click index quantity of the historical video corresponding to the current training attraction factor after being issued on line is larger than the click index quantity corresponding to the last training attraction factor.
Further, the video distribution apparatus may further include:
the training sample acquisition module is used for acquiring two or more preset training attraction factors and two or more historical video frames which serve as historical candidate covers in the historical videos corresponding to the single training attraction factor, and determining feature labels of the historical candidate covers under the corresponding training attraction factors;
the historical characteristic determining module is used for inputting the single historical candidate cover into a preset model corresponding to the training attraction factor to obtain historical characteristic information of the historical candidate cover under the training attraction factor;
and the network model training module is used for determining corresponding training loss according to the historical characteristic information and the characteristic label of the historical candidate cover under the training attraction factor, adjusting the parameters of the preset model by adopting an optimization method of random gradient descent to enable the training loss to be smaller than a set loss threshold value, and finally taking the latest preset model as a neural network model corresponding to the training attraction factor.
Further, the video cover determination module may include:
the score determining unit is used for inputting the feature vectors corresponding to the candidate covers into a pre-constructed ranking model to obtain the ranking scores of the candidate covers;
and the video cover determining unit is used for determining the video cover of the video to be published according to the ranking score of the candidate cover.
Further, the video distribution apparatus further includes:
and the video publishing module is used for publishing the video to be published on line according to the video cover after the video cover of the video to be published is determined.
Further, the feature vector determination module may be specifically configured to:
and carrying out normalization processing on the feature information of the candidate cover under different attraction factors to obtain the feature vector of the candidate cover.
Further, the candidate cover page obtaining module may include:
the system comprises an initial frame acquisition unit, a frame distribution unit and a frame distribution unit, wherein the initial frame acquisition unit is used for acquiring an initial video frame meeting preset conditions in a video to be issued;
the video frame clustering unit is used for processing the initial video frames by adopting a clustering algorithm to obtain two or more clustering sets;
and the candidate cover determining unit is used for selecting the target video frame from the clustering set as a candidate cover of the video to be published.
The video distribution device provided by the embodiment can be applied to the video distribution method provided by any embodiment, and has corresponding functions and beneficial effects.
EXAMPLE six
Fig. 6 is a schematic structural diagram of an apparatus according to a sixth embodiment of the present invention, as shown in fig. 6, the apparatus includes a processor 60, a storage device 61, and a communication device 62; the number of processors 60 in the device may be one or more, and one processor 60 is taken as an example in fig. 6; the processor 60, the storage means 61 and the communication means 62 in the device may be connected by a bus or other means, as exemplified by the bus connection in fig. 6.
The storage device 61 is a computer-readable storage medium, and can be used for storing software programs, computer-executable programs, and modules, such as program instructions/modules corresponding to the video distribution method provided in the embodiment of the present invention. The processor 60 executes various functional applications and data processing of the apparatus by executing software programs, instructions and modules stored in the storage device 61, that is, implements the above-described video distribution method.
The storage device 61 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the terminal, and the like. Further, the storage device 61 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid state storage device. In some examples, the storage 61 may further include memory located remotely from the processor 60, which may be connected to the device over a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The communication means 62 may be used to enable a network connection or a mobile data connection between devices.
The device provided by the embodiment can be used for executing the video distribution method provided by any embodiment, and has corresponding functions and beneficial effects.
EXAMPLE seven
Seventh, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, can implement the video distribution method in any of the above embodiments. The method specifically comprises the following steps:
acquiring two or more video frames serving as candidate covers in a video to be published;
respectively inputting the single candidate cover into different types of pre-constructed neural network models to obtain characteristic information of the candidate cover under different attraction factors, wherein the attraction factors represent factors for attracting a user to click or watch a video to be published;
determining a feature vector of the candidate cover according to the feature information of the candidate cover under different attraction factors;
and determining the video cover of the video to be published according to the feature vectors corresponding to the two or more candidate covers.
Of course, the storage medium provided by the embodiment of the present invention contains computer-executable instructions, and the computer-executable instructions are not limited to the method operations described above, and may also perform related operations in the video distribution method provided by any embodiment of the present invention.
From the above description of the embodiments, it is obvious for those skilled in the art that the present invention can be implemented by software and necessary general hardware, and certainly, can also be implemented by hardware, but the former is a better embodiment in many cases. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as a floppy disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a FLASH Memory (FLASH), a hard disk or an optical disk of a computer, and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network device) to execute the methods according to the embodiments of the present invention.
It should be noted that, in the embodiment of the video distribution apparatus, the units and modules included in the embodiment are only divided according to functional logic, but are not limited to the above division as long as the corresponding functions can be implemented; in addition, specific names of the functional units are only for convenience of distinguishing from each other, and are not used for limiting the protection scope of the present invention.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (9)

1. A video distribution method, characterized in that,
acquiring two or more video frames serving as candidate covers in a video to be published;
respectively inputting a single candidate cover into different types of pre-constructed neural network models to obtain characteristic information of the candidate cover under different attraction factors, wherein the attraction factors represent factors for attracting a user to click or watch a video to be published;
determining a feature vector of the candidate cover according to the feature information of the candidate cover under different attraction factors;
determining a video cover of the video to be published according to the feature vectors corresponding to two or more candidate covers;
wherein the different types of neural network models are constructed by adopting the following steps:
acquiring two or more preset training attraction factors;
in the first round of testing, randomly selecting a frame of video cover from a historical video to be distributed on line so as to determine a click index quantity corresponding to the first round of testing;
aiming at each round of test after the first round of test, continuously adding the neural network model corresponding to the training attraction factor corresponding to the round of test on the neural network model corresponding to each training attraction factor reserved in the previous round of test, and the training attraction factor targeted by the round of test is taken as the current training attraction factor of the round of test, if the click index quantity of the historical video corresponding to the current training attraction factor in the round of test after online release is larger than the click index quantity corresponding to each training attraction factor reserved in the previous round of test, and if not, discarding the neural network model corresponding to the current training attraction factor until the training attraction factor is traversed through multiple rounds of tests to obtain the neural network models of different types.
2. The method according to claim 1, before acquiring two or more video frames as candidate covers in the video to be published, further comprising:
acquiring two or more preset training attraction factors and two or more historical video frames serving as historical candidate covers in a historical video corresponding to a single training attraction factor, and determining feature labels of the historical candidate covers under the corresponding training attraction factors;
inputting a single historical candidate cover into a preset model corresponding to the training attraction factor to obtain historical characteristic information of the historical candidate cover under the training attraction factor;
and determining corresponding training loss according to the historical characteristic information and the characteristic label of the historical candidate cover under the training attraction factor, adjusting the parameters of a preset model by adopting an optimization method of random gradient descent to enable the training loss to be smaller than a set loss threshold value, and finally taking the latest preset model as a neural network model corresponding to the training attraction factor.
3. The method of claim 1, wherein determining the video cover of the video to be published according to the feature vectors corresponding to two or more candidate covers comprises:
inputting the feature vector corresponding to the candidate cover into a pre-constructed ranking model to obtain a ranking score of the candidate cover;
and determining the video cover of the video to be published according to the ranking score of the candidate cover.
4. The method of claim 1, after determining the video cover of the video to be distributed, further comprising:
and publishing the video to be published on line according to the video cover.
5. The method of claim 1, wherein determining the feature vector of the candidate cover based on the feature information of the candidate cover under different attraction factors comprises:
and carrying out normalization processing on the feature information of the candidate cover under different attraction factors to obtain the feature vector of the candidate cover.
6. The method of claim 1, wherein obtaining two or more video frames of the video to be published as candidate covers comprises:
acquiring an initial video frame meeting preset conditions in the video to be published;
processing the initial video frame by adopting a clustering algorithm to obtain two or more cluster sets;
and selecting a target video frame from the clustering set as a candidate cover of the video to be published.
7. A video distribution apparatus, comprising:
the candidate cover acquiring module is used for acquiring two or more video frames serving as candidate covers in the video to be published;
the characteristic information determining module is used for respectively inputting the single candidate cover into different types of pre-constructed neural network models to obtain the characteristic information of the candidate cover under different attraction factors, wherein the attraction factors represent factors for attracting a user to click or watch a video to be published;
the characteristic vector determining module is used for determining the characteristic vector of the candidate cover according to the characteristic information of the candidate cover under different attraction factors;
the video cover determining module is used for determining the video cover of the video to be published according to the feature vectors corresponding to two or more candidate covers;
wherein the different types of neural network models are constructed by adopting the following steps:
acquiring two or more preset training attraction factors;
in the first round of testing, randomly selecting a frame of video cover from a historical video to be distributed on line so as to determine a click index quantity corresponding to the first round of testing;
aiming at each round of test after the first round of test, continuously adding the neural network model corresponding to the training attraction factor corresponding to the round of test on the neural network model corresponding to each training attraction factor reserved in the previous round of test, and the training attraction factor targeted by the round of test is taken as the current training attraction factor of the round of test, if the click index quantity of the historical video corresponding to the current training attraction factor in the round of test after online release is larger than the click index quantity corresponding to each training attraction factor reserved in the previous round of test, and if not, discarding the neural network model corresponding to the current training attraction factor until the training attraction factor is traversed through multiple rounds of tests to obtain the neural network models of different types.
8. A computer device, characterized in that the computer device comprises:
one or more processors;
storage means for storing one or more programs;
when executed by the one or more processors, cause the one or more processors to implement the video distribution method of any of claims 1-6.
9. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the video distribution method according to any one of claims 1 to 6.
CN201910087567.XA 2019-01-29 2019-01-29 Video publishing method, device, equipment and storage medium Active CN111491202B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201910087567.XA CN111491202B (en) 2019-01-29 2019-01-29 Video publishing method, device, equipment and storage medium
PCT/CN2020/072191 WO2020156171A1 (en) 2019-01-29 2020-01-15 Video publishing method, apparatus and device, and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910087567.XA CN111491202B (en) 2019-01-29 2019-01-29 Video publishing method, device, equipment and storage medium

Publications (2)

Publication Number Publication Date
CN111491202A CN111491202A (en) 2020-08-04
CN111491202B true CN111491202B (en) 2021-06-15

Family

ID=71794177

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910087567.XA Active CN111491202B (en) 2019-01-29 2019-01-29 Video publishing method, device, equipment and storage medium

Country Status (2)

Country Link
CN (1) CN111491202B (en)
WO (1) WO2020156171A1 (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112689187A (en) * 2020-12-17 2021-04-20 北京达佳互联信息技术有限公司 Video processing method and device, electronic equipment and storage medium
CN112800276B (en) * 2021-01-20 2023-06-20 北京有竹居网络技术有限公司 Video cover determining method, device, medium and equipment
CN113111222B (en) * 2021-03-26 2024-03-19 北京达佳互联信息技术有限公司 Short video template generation method, device, server and storage medium
CN113315984B (en) * 2021-05-21 2022-07-08 北京达佳互联信息技术有限公司 Cover display method, device, system, equipment and storage medium
CN113821678B (en) * 2021-07-21 2024-04-12 腾讯科技（深圳）有限公司 Method and device for determining video cover

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107832725A (en) * 2017-11-17 2018-03-23 北京奇虎科技有限公司 Video front cover extracting method and device based on evaluation index
CN108595493A (en) * 2018-03-15 2018-09-28 腾讯科技（深圳）有限公司 Method for pushing and device, storage medium, the electronic device of media content

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4958748B2 (en) * 2007-11-27 2012-06-20 キヤノン株式会社 Audio processing device, video processing device, and control method thereof
US9848228B1 (en) * 2014-05-12 2017-12-19 Tunespotter, Inc. System, method, and program product for generating graphical video clip representations associated with video clips correlated to electronic audio files
CN104881798A (en) * 2015-06-05 2015-09-02 北京京东尚科信息技术有限公司 Device and method for personalized search based on commodity image features
CN106503693B (en) * 2016-11-28 2019-03-15 北京字节跳动科技有限公司 The providing method and device of video cover
CN106599208B (en) * 2016-12-15 2022-05-06 腾讯科技（深圳）有限公司 Content sharing method and user client
CN107958030B (en) * 2017-11-17 2021-08-24 北京奇虎科技有限公司 Video cover recommendation model optimization method and device
CN108650524B (en) * 2018-05-23 2022-08-16 腾讯科技（深圳）有限公司 Video cover generation method and device, computer equipment and storage medium
CN109165301B (en) * 2018-09-13 2021-04-20 北京字节跳动网络技术有限公司 Video cover selection method, device and computer readable storage medium

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107832725A (en) * 2017-11-17 2018-03-23 北京奇虎科技有限公司 Video front cover extracting method and device based on evaluation index
CN108595493A (en) * 2018-03-15 2018-09-28 腾讯科技（深圳）有限公司 Method for pushing and device, storage medium, the electronic device of media content

Also Published As

Publication number Publication date
WO2020156171A1 (en) 2020-08-06
CN111491202A (en) 2020-08-04

Similar Documents

Publication Publication Date Title
CN111491202B (en) Video publishing method, device, equipment and storage medium
CN109165249B (en) Data processing model construction method and device, server and user side
CN111241985B (en) Video content identification method and device, storage medium and electronic equipment
CN109582903B (en) Information display method, device, equipment and storage medium
CN110737783A (en) method, device and computing equipment for recommending multimedia content
CN110413867B (en) Method and system for content recommendation
CN110968767B (en) Ranking engine training method and device, and business card ranking method and device
US20160055243A1 (en) Web crawler for acquiring content
US20170344899A1 (en) Automatic generation of training sets using subject matter experts on social media
CN111597446B (en) Content pushing method and device based on artificial intelligence, server and storage medium
CN109344314A (en) A kind of data processing method, device and server
CN109189889B (en) Bullet screen recognition model establishing method, device, server and medium
CN111783712A (en) Video processing method, device, equipment and medium
CN109729377B (en) Anchor information pushing method and device, computer equipment and storage medium
CN110245310B (en) Object behavior analysis method, device and storage medium
CN110032678A (en) Service resources method for pushing and device, storage medium and electronic device
CN109978575A (en) A kind of method and device excavated customer flow and manage scene
CN113011886B (en) Method and device for determining account type and electronic equipment
CN117194772B (en) Content pushing method and device based on user tag
CN110309753A (en) A kind of race process method of discrimination, device and computer equipment
CN114245185A (en) Video recommendation method, model training method, device, electronic equipment and medium
CN113127778A (en) Information display method and device, server and storage medium
CN117235371A (en) Video recommendation method, model training method and device
CN113836388A (en) Information recommendation method and device, server and storage medium
CN112182460A (en) Resource pushing method and device, storage medium and electronic device

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
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20220608

Address after: 31a, 15 / F, building 30, maple mall, bangrang Road, Brazil, Singapore

Patentee after: Baiguoyuan Technology (Singapore) Co.,Ltd.

Address before: 511400 floor 5-13, West Tower, building C, 274 Xingtai Road, Shiqiao street, Panyu District, Guangzhou City, Guangdong Province

Patentee before: GUANGZHOU BAIGUOYUAN INFORMATION TECHNOLOGY Co.,Ltd.