CN107305557A - Content recommendation method and device - Google Patents
Content recommendation method and device Download PDFInfo
- Publication number
- CN107305557A CN107305557A CN201610249026.9A CN201610249026A CN107305557A CN 107305557 A CN107305557 A CN 107305557A CN 201610249026 A CN201610249026 A CN 201610249026A CN 107305557 A CN107305557 A CN 107305557A
- Authority
- CN
- China
- Prior art keywords
- content
- video
- object content
- recommendation
- frame
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—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
- G06F16/7867—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, title and artist information, manually generated time, location and usage information, user ratings
-
- 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
- G06F16/73—Querying
- G06F16/735—Filtering based on additional data, e.g. user or group profiles
-
- 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
Abstract
The embodiment of the present invention provides a kind of content recommendation method and device.Methods described includes：Ask the video played to send to player during the video playback is asked when the video playback for receiving player is asked, trigger video described in the player plays；The content recommendation matched with the object content is obtained also after the object content in detecting and obtaining the video simultaneously, when the object content is played, to trigger the prompt message that the player exports the content recommendation in the corresponding viewing area of the object content.Content recommendation in the embodiment of the present invention has more specific aim.
Description
Technical field
The application is related to video information process field, relates more specifically to a kind of content recommendation method and device.
Background technology
Quick popularization and digital image collection, the development for the treatment of technology with internet, video industry are special
It is not that Internet video industry emerges rapidly.It is used as a kind of multi informations such as image, sound, word of including
There is comprehensive media, video powerful information to carry and transmission capacity, be played in people's daily life
More and more important effect.
In the playing process of video, in order to recommend more contents to user, in the prior art,
Generally content recommendation is being exported before video is commenced play out or in the playing process of video, is temporarily being interrupted
Video is played to export content recommendation, wherein, content recommendation can be the information such as advertisement, news, consulting.
Seen from the above description, current commending contents mode, is using pre-setting, fixed mostly
The content recommendation of pattern.For a user, video-see can be both influenceed, again user can be made passively to connect
The content recommendation fixed, content recommendation whether be easily accepted by a user or whether useful to user etc. be all
Unknown, therefore the specific aim of content recommendation is poor.
The content of the invention
The embodiment of the present invention provides a kind of content recommendation method and device, to solve content in the prior art
The technical problem for recommending specific aim poor.
The embodiment of the present invention provides a kind of content recommendation method, including：
Receive the video playback request of player；
Ask the video played to send to player during the video playback is asked, trigger the player
Play the video；
Detect and obtain the object content in the video；
Obtain the content recommendation matched with the object content；
When the object content is played, the player is triggered in the corresponding viewing area of the object content
The prompt message of the content recommendation is exported in domain.
Preferably, the content recommendation matched with the object content that obtains includes：
According to the order with the object content similarity from high to low, it is determined that being matched with the object content
Multiple pictures；
Obtain the user characteristics of the user of the triggering video playback request；
For each picture, by the user characteristics, the content characteristic of picture correspondence alternating content
And the content characteristic of the object content is combined, and obtains recommended characteristics；The alternating content it is interior
Holding feature includes history recommendation rate and/or content type；The content characteristic of the object content includes the mesh
There is the similarity of duration and/or the object content and the picture in mark content；
Using the recommended models of training in advance, the recommendation rate of each recommended characteristics is estimated；
The corresponding alternating content of selection recommendation rate highest picture is used as content recommendation.
Preferably, it is described to detect and obtain the object content in the video, including：
Detect and obtain the object content extracted in advance in the video；
The object content is extracted in advance from the video as follows：
Extract the key frame in the video；
Using the object content detection model of training in advance, object content is extracted from the key frame.
Preferably, it is described from key frame extract object content after, methods described also includes：
Target following is carried out to the object content, appearance of the object content in the video is determined
Duration；
The content recommendation matched with the object content that obtains includes：
When the object content duration occurs more than preset duration, obtain what is matched with the object content
Content recommendation.
Preferably, the key frame extracted in the video, including：
It regard the first two field picture frame in the video as key frame；
For each two field picture frame after key frame, the comentropy of described image frame is calculated；
When described information entropy is more than entropy threshold, the phase of any frame picture frame and the key frame is calculated
Like degree；
When the similarity is less than similarity threshold, any frame picture frame is regard as key frame.
The embodiment of the present invention provides a kind of content recommendation device, including：
Playing request receiving module, the video playback for receiving player is asked；
Video sending module, for asking the video played to send to broadcasting during the video playback is asked
Device, triggers video described in the player plays；
Object content detection module, for detecting and obtaining the object content in the video；
Content recommendation acquisition module, for obtaining the content recommendation matched with the object content；
Display module is triggered, for when the object content is played, triggering the player in the mesh
The prompt message of the content recommendation is exported in the corresponding viewing area for marking content.
Preferably, the content recommendation acquisition module includes：
Arrangement units, for according to the order with the object content similarity from high to low, it is determined that and institute
State multiple pictures of object content matching and the content characteristic of picture；
First acquisition unit, the user characteristics of the user for obtaining the triggering video playback request；
Assembled unit, for each picture, by the user characteristics, picture correspondence alternating content
Content characteristic and the content characteristic of the object content be combined, obtain recommended characteristics；It is described to wait
Selecting the content characteristic of content at least includes history recommendation rate；The content characteristic of the object content includes described
There is the similarity of duration and/or the object content and the picture in object content；
Unit is estimated, for the recommended models using training in advance, the recommendation of each recommended characteristics is estimated
Rate；
Selecting unit, for selecting the corresponding content characteristic of recommendation rate highest picture as content recommendation.
Preferably, the object content detection module includes：
Detection unit, for detecting and obtaining the object content extracted in advance in the video；
Described device also includes：
Key-frame extraction module：For extracting the key frame in the video；
Object content extraction module：For the object content detection model using training in advance, closed from described
Object content is extracted in key frame.
Preferably, described device also includes：
Target tracking module, for carrying out target following to the object content, determines the object content
Appearance duration in the video；
In the object content specifically for there is duration more than preset duration in the content recommendation acquisition module
When, obtain the content recommendation matched with the object content.
Preferably, the key-frame extraction module includes：
First determining unit, for regarding the first two field picture frame in the video as key frame；
Comentropy computing unit, for for each two field picture frame after key frame, calculating described image
The comentropy of frame；
Similarity calculated, for when described information entropy is more than entropy threshold, calculating any frame figure
As the similarity of frame and the key frame；
Second determining unit, for when the similarity is less than similarity threshold, by any frame figure
As frame is used as key frame.
A kind of content recommendation method and device provided in an embodiment of the present invention, from the video watched according to user
Content is set out, obtain video in object content, and obtain object content matching content recommendation, regarding
The prompt message of synchronism output content recommendation in frequency playing process, is regarded in whole recommendation process without interrupting
Frequency is played, and improves Consumer's Experience.And content recommendation is the content matched with the object content of video content,
So that content recommendation is more targeted, personalized recommendation is realized.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to reality
The accompanying drawing used required for applying in example or description of the prior art is briefly described, it should be apparent that, under
Accompanying drawing in the description of face is some embodiments of the present invention, for those of ordinary skill in the art,
On the premise of not paying creative work, other accompanying drawings can also be obtained according to these accompanying drawings.
Fig. 1 is a kind of one embodiment flow chart of content recommendation method in the present invention；
Fig. 2 is an a kind of output information entry example figure of content recommendation method in the present invention；
Fig. 3 is a kind of another embodiment flow chart of content recommendation method in the present invention；
Fig. 4 is a kind of another embodiment flow chart of content recommendation method in the present invention；
Fig. 5 is a kind of one embodiment structural representation of content recommendation device in the present invention；
Fig. 6 is a kind of another example structure schematic diagram of content recommendation device in the present invention.
Embodiment
To make the purpose, technical scheme and advantage of the embodiment of the present invention clearer, below in conjunction with this hair
Accompanying drawing in bright embodiment, the technical scheme in the embodiment of the present invention is clearly and completely described,
Obviously, described embodiment is a part of embodiment of the invention, rather than whole embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained under the premise of creative work is not made
The every other embodiment obtained, belongs to the scope of protection of the invention.
The technical scheme of the application is mainly used in Internet video and played in scene, mainly in Internet video
Playing process in carry out commending contents.Quick popularization and the development of digital audio-effect processing in internet
Support under, Internet video industry emerges rapidly.Internet video is carried and propagated energy with its powerful information
Power is rapidly developed.Conventional Internet video is when playing, as described in the background art, with solid
Determine the form of content and commence play out content recommendation in video playback, or interrupt the video of user's broadcasting regarding
Commending contents are completed in the playing process of frequency.The way of recommendation of this immobilized substance easily causes user to pushing away
The conflict psychology of content is recommended, the drawbacks such as recommendation effect reduction are caused.
To solve this technical problem, inventor passes through after a series of researchs, proposes present techniques side
Case.In embodiments of the present invention, when service end receives the playing request from player, detection is simultaneously
The corresponding object content of video that request is played is obtained, and obtains the content recommendation matched with object content；
And when object content is played, trigger and export the prompting of content recommendation in the corresponding viewing area of object content
Information, the prompt message is used to prompt the user with the content recommendation.In the present embodiment, content recommendation exists
Synchronism output in the playing process of video, without in the broadcasting of break of video, and content recommendation and video
Object content matches so that content recommendation is more targeted.For different video and different
Object content, content recommendation may also can be different, it is achieved that the personalized recommendation of content recommendation, is used
Family is easier to receive the content recommendation related to object content, improves Consumer's Experience.
Technical scheme is described in detail below in conjunction with accompanying drawing.
As shown in figure 1, a kind of one embodiment flow chart of content recommendation method for the embodiment of the present invention,
This embodiment scheme is mainly used in server, and this method can include following steps：
101：Receive the video playback request of player；
In the application, player can be the videoconference client that mobile terminal is installed；It can also be browser
In play control；It can also be the videoconference client for being installed on computer.
Server in the application can be video server.
As a kind of possible implementation, video playback request can be selected according to user by player
Select some video to perform playing request action and generate.
102：Will the video playback ask in ask the video played to send to player, broadcast described in triggering
Put device and play the video；
After the video is found, the video can be sent to player by server.Connect in player
Receive after video, video described in player plays can be triggered.
103：Detect and obtain the object content in the video；
As a kind of possible implementation, in order to which the space complexity and time that reduce total algorithm are complicated
Degree, the object content can be the object content extracted in advance.After video playback request is received,
Server for its object content for extracting in advance of the video retrieval and can notify player automatically.
The object content can be set according to the actual requirements.For example：Can occur in video
Automobile occurred in a certain personal jacket, trousers, shoes, or video etc..Certain object content is not
The types such as clothing, shoes, automobile are only limitted to, can also be that mobile phone, building, beverage etc. have and use
Function or the article for showing function.
The object content can in advance be extracted using following methods from the video：
The key frame in the video is first extracted, the object content detection model of training in advance is reused, from
Object content is extracted on the key frame.
In order to obtain more recommend value object content, can also to the object content carry out target with
Track, determines the duration that the object content occurs in video.Therefore can be grown up only for when occurring
The operation of step 104 is performed in the object content of preset duration.
As another possible implementation, object content can also in real time be carried after video is got
Take, to reduce memory cost.
104：Obtain the content recommendation matched with the object content；
Server can pre-save different types of content recommendation, and the content recommendation can include figure
The contents such as Webpage, the text of piece.The picture for representing content recommendation can be stored in database.
When needing to obtain content recommendation, it can obtain and be matched with the object content from the picture database
Picture, and then the corresponding content recommendation of picture can be obtained.
Wherein, obtain the content recommendation that is matched with object content, the content recommendation can including one or
It is multiple.
Can be by calculating the recommendation that object content and the similarity of content recommendation determine to match with object content
Content.It can be specifically descending according to similarity, select one or more content recommendation.Recommend
When content is represented with picture, it is possible to use the box counting algorithm similarity such as image histogram.
Preferably, in order to further improve the degree of accuracy of content recommendation, it can also obtain and the target
User characteristics is combined during the content recommendation of content matching, namely is obtained and object content and user
The content recommendation that feature is matched.The user characteristics can be age of user, sex, region, duty
The message such as industry, these message can be obtained according to user's registration information.Add after user characteristics, can be with
The content recommendation more pressed close to user preferences is obtained, personalized recommendation is further realized.
Therefore, can be by content recommendation, object content, user as a kind of possible implementation
Feature etc. is combined, and obtains recommended characteristics, and by recommended models of the recommended characteristics by training in advance,
The recommendation rate of recommended characteristics is estimated, then the corresponding alternating content conduct of reselection recommendation rate highest picture
Content recommendation.
105：When the object content is played, the correspondence for triggering the player in the object content shows
Show the prompt message that the content recommendation is exported in region.
The prompt message of the content recommendation can be the summary letter of the content recommendation, the content recommendation
The items for information of the chained address of breath or the summary info comprising the content recommendation and content recommendation.
Link address one web page files of correspondence of the items for information, can be jumped by clicking on described information entry
The corresponding address page of content recommendation is gone to, so as to check the content recommendation.
Wherein it is possible to be when player commences play out object content, while exporting items for information.
Wherein, output items for information schematic diagram specifically can be as shown in Figure 2.Two are shown in fig. 2
Different types of object content, respectively trousers and jacket, and show corresponding entry
Information.Items for information 201 is carried for the content recommendations of jacket and trousers respectively with items for information 202
Show information, the prompt message has link address, and user, which clicks on this link, can just open correspondence webpage
The page completes browse, buy etc. to operate.
Wherein, user can also choose whether to export the content recommendation in player, in order to avoid to user's
Actual play causes other influences.So as to be receiving user triggering content recommendation export request
When, then perform step 103~105.
It should be noted that step 103~step 104 can receive video playback in step 101 to ask
Performed or in step 102 player plays video when asking, when being played to object content again
Perform.The present invention does not limit the execution step with the present embodiment.
In the present embodiment, by the analysis to video content, the object content related to video content is obtained,
And content recommendation corresponding with the object content is selected from database, then when playing video,
Show the prompt message of content recommendation.The content recommendation is the content based on video content, can be not
Selection pushes different contents during putting different video.Change fixed recommendation article in the prior art
Pattern, improve the specific aim of content recommendation, realize user watch video while user is provided
The commending contents easily received, improve the experience of user.
As one embodiment, as shown in figure 3, step 101~step 103 and step 105 and Fig. 1
Illustrated embodiment is identical, wherein, step 104 may comprise steps of the acquisition for carrying out content recommendation：
1041：According to the order with the object content similarity from high to low, it is determined that with the target
Hold multiple pictures of matching and the content characteristic of each picture correspondence alternating content.
Preferably, in the database picture and the content characteristic of picture can be beforehand through network canal
Picture concerned and its database of content composition that road or other channels are obtained.According to the object content
It can use and be retrieved based on two kinds of strategies of classification and histogram similarity from the database and the mesh
The high picture of mark content similarity and the content spy that the corresponding alternating content of picture is obtained by inquiring about database
Levy., can be suitable according to from high to low by the result of calculation of similarity as a kind of possible implementation
Sequence is arranged, and chooses some pictures in the top alternately picture.
1042：Obtain the user characteristics of the user of the triggering video playback request.
The user characteristics of the user can mainly include：Age of user, sex, region, occupation,
Platform etc..User characteristics can be obtained from the log-on message of user.The log-on message can be stored in
, it is necessary to which during user characteristics, the log-on message of user can be inquired about from server in server.
1043：For each picture, by the user characteristics, picture correspondence alternating content
The content characteristic for holding feature and the object content is combined, and obtains recommended characteristics；
The content characteristic of the object content can include the Similarity Measure result of object content and picture,
The duration that object content occurs in video can also be included.
The content characteristic of the alternating content can include history recommendation rate；Content type can also be included；
In alternating content correspondence actual object, item price, the article place of production can also be included and be used as content recommendation
When, when prompt message includes chained address, the history recommendation rate is to refer to the information such as historic click-through rate.
Due to equal in the feature of user characteristics, the content characteristic of the alternating content and the object content
Continuous feature and discrete features are potentially included, discrete features refer to that the numerical value of feature is centrifugal pump, continuous special
Levy and refer to that the numerical value of feature is continuous.For example：User's sex character etc. can obtain centrifugal pump, be from
Dissipate feature；Price, overall clicking rate or picture and the phase of object content feature in image content feature
Successive value can be obtained like degree etc., is continuous feature.
In order to which user characteristics, the image content feature, the object content feature are combined,
Need continuous feature carrying out sliding-model control.The processing mode of discretized features is as follows：
B1：The scope of continuous character numerical value is 0~X, is divided into N sections.Then continuous feature is changed into N bit
Binary system discrete features.
B2:Which section judging characteristic falls in, and such as correspondence position falls at the 3rd section, then discrete features are expressed as
00000100.(assuming that being divided into 8 sections).
, can be by above-mentioned three kinds for the accuracy rate that effective lift scheme is predicted after discrete features are completed
Hold and carry out characteristic crossover, that is to say, that two features can be reconfigured to form new feature, then will combination
Feature is drawn into a vector, obtains recommended characteristics.For example：By sex and content type (m classes, m
Represent an integer) the discrete features of 2m can be produced after combination.
As a kind of possible implementation, recommended characteristics can be set to vector x, it is assumed that the dimension of feature
Spend for 113.Wherein x1~x10 is age of user characteristic segments；X11~x18 is user's regional feature section；
X19~x25 is user's job characteristics section；X26~x30 is player name feature section；X31~x38 is candidate
Content type characteristic segments；X39~x50 is alternating content price feature section；X51~x58 is alternating content region
Characteristic segments；X59~x60 is alternating content clicking rate characteristic segments；X61~x65 is that object content duration spy occurs
Levy section；X66~x75 is object content and corresponding picture similarity characteristic segments；X76~x91 is alternating content class
Not/user's sex combination characteristic segments；X92~x113 is user's sex/alternating content price assemblage characteristic section.
1044：Using the recommended models of training in advance, the recommendation rate of each recommended characteristics is estimated；
The purpose for estimating the recommendation rate of each recommended characteristics is the ranking for obtaining recommendation rate, can will recommend rate
The higher corresponding content of one or several recommended characteristics is set to content recommendation, and the number of content recommendation is herein
Excessive restriction is not done.
As a kind of possible implementation, the recommendation rate can specifically refer to clicking rate, that is, user exists
The probability clicked on during viewing video to the content recommendation prompt message of appearance.Clicking rate can be for table
Show the consumers' acceptable degree of content recommendation, recommended by the content recommendation for selecting clicking rate high, can be with
Further improve the degree of accuracy of content recommendation.
Predicted the outcome to obtain accurately clicking rate, the recommended models can use Logic Regression Models
(Logic Regression, LR).
Wherein, Logic Regression Models be assumed to be conditional probability meet parameter be θ probability-distribution function：
Wherein, T represents transposition, and θ refers to model parameter, and x refers to the variate-value of some input, and y=1 is represented a little
Hit, y=-1 represents not clicking on, g (h) here is S type functions (also referred to as S-shaped growth curve).
Above-mentioned Logic Regression Models are substituted into using recommended characteristics as variable x, the probable value P obtained is calculated and both may be used
To represent clicking rate.
But at this moment parameter θ is unknown, it is necessary to be solved to parameter θ.Therefore the process of training recommended models
Namely it is to solve for optimal model parameter θ process.Optimized parameter θ solution can be estimated using maximum likelihood
The mode of meter solves optimal model parameters θ, can specifically include：
First, training sample, D=(x are predefined1,y1),(x2,y2),…,(xn,yn)；
Wherein,For the recommended characteristics of structure, k value 1,2,3 ... n, n are
The number of the total number of sample namely the corresponding multiple pictures of an object content；M represents recommended characteristics
Dimension；ynFor sample label, such as its value is that 1 expression is clicked on, and such as value represents not clicking on for -1.
Secondly, the training sample D is inputted into above-mentioned formula A1, obtains Logic Regression Models corresponding most
Maximum-likelihood degree function：
L (θ)=P (D | θ)=∏ p (y | x；θ)=∏ g (θTx)y(1-g(θTx))1-y…………………(A2)
Logarithm is asked to it, log likelihood function can be obtained：
L (θ)=log (L (θ))=∏ ylogg (θTx)+(1-y)log(1-g(θTx)……………………(A3)
Again, iterative optimized parameter θ can be declined using gradient, specifically chooses log likelihood letter
Number (A3), which changes a most fast direction, to be carried out adjusting parameter to approach optimal solution.Basic step is as follows：
Select descent direction (gradient direction, ▽ J (θ))
Select step-length, undated parameter θi=θi-1-αi▽J(θi-1), wherein α be learning rate, Schistosomiasis control precision,
▽ J (θ) are current gradient.
Descent direction is most fast until meeting for two steps more than repeating, i.e., the Grad of loss function is maximum.
The gradient ▽ J (θ) of wherein loss function can be in the hope of log likelihood function on θ partial derivative obtain, have
The computational methods of body are：
By the above method, the value of parameter θ when log likelihood function changes most fast can be obtained.Then
By the value of the θ and recommended characteristics input formula (A1), the recommendation probability of recommended characteristics can be calculated.
1045：Selection estimates the corresponding alternating content of clicking rate highest picture as content recommendation.
The content recommendation, can be the content recommendation related to object content, for example, it may be and mesh
Mark the similar content of content type or content high with object content characteristic similarity etc..
In the above-described embodiments, the clicking rate of content recommendation is estimated using linear regression model (LRM), can be made
More preferable recommendation effect is obtained during the content recommendation for recommending selection, higher computational efficiency can also be obtained.
As another embodiment, the extraction of object content in step 103 can be according to following several
Individual step is completed：
C1：Extract the key frame in the video；
As a kind of possible implementation, the key frame in the video is extracted, image can be used to believe
Cease the key frame in the method extraction video of entropy and image similarity.The idiographic flow for extracting key frame can be with
Including following several steps：
C1a：It regard the first two field picture frame in the video as key frame；
, can be from the first frame figure of video when reading the picture frame of video in order to obtain more accurately data
As frame starts, on a frame-by-frame basis read, to the last a two field picture frame.If, will before without key frame
The first two field picture frame in the video performs step C1b afterwards as key frame.
C1b：For each two field picture frame after key frame, the comentropy of described image frame is calculated；
The value of the comentropy of the picture frame extracted is calculated using formula (A5).
Wherein, piThe probability that the pixel for being i for gradation of image occurs.
Comentropy represents the abundant degree of image information to a certain extent, and the value of comentropy shows more greatly figure
The detail content included as in is more, there is a possibility that object content is also bigger.The letter of solid-color image
The value for ceasing entropy is 0.
C1c：When described information entropy is more than entropy threshold, any two field picture and the key frame are calculated
Similarity；
Some useless frames can be filtered out with entropy threshold by the comentropy for calculating the picture frame extracted,
Reduce the complexity of algorithm.
Judge whether the comentropy of described image frame is more than information entropy threshold.
If the comentropy of picture frame is more than described information entropy threshold, the two field picture and upper one are calculated
The similarity of key frame, judges whether the value of the similarity is less than similarity threshold；If less than described
Then described image frame is key frame to similarity threshold.
Calculate the similarity of described image frame and a upper key frame；
Calculating the similarity of described image frame can use formula A6 to calculate, and formula A6 is as follows：
Wherein, N is the number of histogram piecemeal, piFall the probability in i-th section of pixel for gradation of image.
P in this applicationiWith p 'iRefer to the grey level histogram of the picture frame and previous frame key frame currently extracted
Statistical probability.The similarity calculating method of image grey level histogram is that a kind of time complexity is relatively low, quick
Estimate the method for image similarity, textural characteristics algorithm, such as SIFT (Scale-invariant can also be used
Feature transform, Scale invariant features transform) algorithm, HOG (Histograms of Oriented
Gradients, histograms of oriented gradients) algorithm scheduling algorithm calculate two field pictures frame similarity.
C1d：When the similarity is less than similarity threshold, any frame picture frame is regard as key frame.
Whether the result of the similarity calculated in judgment step C1c is less than some similarity threshold；
If the value for calculating two field pictures frame similarity is less than similarity threshold, new key frame is produced；
Step C1a is then back to if greater than similarity threshold.
C2：Using the detection model of the object content of training in advance, extracted from the key frame in target
Hold.
After step C1 extracts key frame, needed in step C2 in the target using training in advance
The detection model of appearance, object content is extracted from the key frame.
The detection model of the object content of training in advance can be used, is extracted from the key frame in target
Hold.The detection model of the object content of the training in advance can use algorithm of target detection to be trained.
In order to obtain higher feature representation ability, algorithm of target detection can be convolutional neural networks, institute
State target detection model and that is to say convolutional neural networks model, therefore can use convolutional neural networks mould
Type carries out the extraction of object content.Wherein, it can be wrapped using convolutional neural networks model extraction object content
Include：
Window sliding is carried out on key frame, to extract candidate region；Window function is entered with key frame
Row convolutional calculation；For a key frame, 2000 candidate regions can be extracted, then to each
Calculating is normalized in individual region, makes candidate region in the same size；
The high dimensional feature of candidate region is extracted using convolutional Neural pessimistic concurrency control；
By the full articulamentum of convolutional neural networks, candidate region is classified, to obtain object content；
If being judged as object content by candidate region after grader, when the window quilt of front slide
It is determined as target window, so as to obtain classification and the positional information of object content.
In the present embodiment, image information entropy is used with image similarity to judge picture frame whether for key
Frame, and using object content whole on convolutional neural networks model extraction key frame, make object content
Scope is more comprehensive.
But, if some content time of occurrence is short so that user can not preferably watch throwing very much
Put content, or when delivering multiple contents on same two field picture frame, user may also without it is enough when
Between viewing deliver content.In order to better control over advertisement putting number, efficient dispensing return, in step
Rapid C2 is extracted after the target in key frame, and the target can be tracked, with calculate target regarding
The duration occurred in frequency, the object content for choosing object content time of occurrence more than threshold value is used as final goal
Content.
As a kind of possible implementation, carrying out target following to object content can include：
From t+1 two field picture frames, multiple sweep corresponding with the object content region of t frames is extracted
Retouch window；
The target following grader obtained using being trained according to t two field pictures frame, enters to the scanning window
Row classified calculating, obtains classification fraction；
It is more than preset fraction in classification fraction, classification fraction highest scanning window is tracked as object content.
Wherein, if classification fraction is less than preset fraction, then it is assumed that do not include target in multiple scanning windows
Content, then object content tracking terminates.
Wherein target following grader can be Naive Bayes Classifier.
Target following grader training in advance can be obtained in such a way：
When t frames, several positive samples of the two field picture frame and the picture of negative sample are collected.
Can be specifically that positive sample is obtained by the nearer region sliding window of the object content in t frames,
Area sampling obtains negative sample to t frames distance objective farther out.Then to the positive sample collected and negative sample
This progress change of scale, carries out dimensionality reduction, then using sparseness measuring matrix to above-mentioned multi-scale image feature
Go to train Naive Bayes Classifier using the feature after dimensionality reduction.
Wherein, to adapt to prolonged target following, above-mentioned Naive Bayes Classifier needs continuous carry out
Training, to update the model of grader.Can be gone to calculate according to the positive sample newly detected and negative sample
The average and variance of the two, are updated using following manner：
Wherein, μ refers to average, and σ refers to variance, λ>0 is Studying factors, in actual applications for avoid miss
The accumulation of difference, can take λ=0.85.
The target following grader obtained according to t two field pictures frame is then utilized, multiple scanning windows are carried out
Classified calculating can be specifically：
Acquisition scanning window around the target location traced into is extracted to t+1 two field pictures frame (to avoid
Scan entire image frame), carry out after change of scale by sparseness measuring matrix to its dimensionality reduction, it is then sharp again
The Naive Bayes Classifier trained with t frames carries out classified calculating to each scanning window, obtains
The classification fraction of each scanning window.
It is more than preset fraction in classification fraction, and the maximum scanning window of fraction of classifying is as object content.
So it is achieved that the target following from t two field pictures frame to t+1 two field picture frames.
In the above-described embodiments, it can obtain and the object content that duration exceedes threshold value occur, can control to throw
Put content number, deliver content recommendation time of occurrence it is longer, user is more intensively watched
Content is delivered, higher dispensing effect is obtained.
As another embodiment, be different from Fig. 1, as shown in figure 4, in video object content extraction
I.e. step 103 can also be performed according to following steps：
10301：A two field picture frame is read from video；
10302：Judge whether current state is in tracking mode, if not, step 10303 is performed, it is no
Then perform step 10308；
10303：Carry out the extraction of key frame；
10304：Judge whether key-frame extraction succeeds；If performing step 10305, step is otherwise returned
Rapid 10301.
10305：The extraction of object content；
10306：Judge whether to extract object content, if it is, performing step 10307, otherwise return
Step 10301；
10307：Setting current state is tracking mode；
10308：Object content is tracked using compression track algorithm；
10309：Judge whether tracking mode terminates, if it is, setting current detection into step 10310
State, while also performing step 10311, calculates target and duration occurs；Otherwise return to step 1031；
The foundation for judging that tracking mode terminates is that object content is no longer come across in picture frame, namely use point
The window of class device training image frame obtains classification fraction and is less than preset value.
10310：Setting current state is tracking mode；
10311：Calculate object content and duration occur；
10312：Judge that object content occurs whether duration is more than threshold value, if performing step 10313
Operation；
10313：Final goal content is produced, the operation of step 104 is performed afterwards.
In embodiments of the present invention, there is provided the data of a sign current state, can sentence according to the data
It is disconnected whether to be in tracking mode, to judge tracking result, reduce the complexity of algorithm.Using target with
The method of track can avoid recommending excessive content, cause user to have no time to watch, on the whole, improve
Recommend efficiency.
Fig. 5 is a kind of structural representation of one embodiment of content recommendation device provided in an embodiment of the present invention
Figure, the device is mainly used in Video service end, and the device can include：
Playing request receiving module 501：Video playback for receiving player is asked；
Wherein, the video playback plays request and sent by playback equipment, and video playback request can
Some video request is selected to play out action and generation of setting out by player to be user.
Video sending module 502：For will the video playback ask in ask play video send to
Player, triggers video described in the player plays；
As a kind of possible implementation, after the video is found, server can be by the video
It is sent to player.After player receives video, video described in player plays can be triggered.Passing
During defeated video, can send the content recommendation related to video content simultaneously with video can also separate hair
Send.
Object content detection module 503：For detecting and obtaining the object content in the video；
Preferably, the object content detection module can specifically be detected with apparatus and obtain the mesh with first extracting
Mark content.The object content can be the object content extracted in advance, i.e., complete mesh before testing
The extraction work of content is marked, based on this, described device can also include：
Key-frame extraction module：For extracting the key frame in the video；
Object content extraction module：For the object content detection model using training in advance, closed from described
Object content is extracted in key frame.
Preferably, the key-frame extraction module can include：
First determining unit：For regarding the first two field picture frame in the video as key frame；
Comentropy computing unit：For for each two field picture frame after key frame, calculating described image
The comentropy of frame；
Similarity calculated：For when described information entropy is more than entropy threshold, calculating any frame figure
As the similarity of frame and the key frame；
Second determining unit：For when the similarity is less than similarity threshold, by any frame figure
As frame is used as key frame.
And after completing the advance extraction of object content, in order to obtain the object content for more recommending value,
Preferably, target following can also be carried out to the object content, determines the object content in video
The duration of appearance.I.e. described device can also include：
Target tracking module：For carrying out target following to the object content, the object content is determined
Appearance duration in the video；
When described device includes target tracking module, the content recommendation acquisition module is specifically for judging
The object content occurs whether duration is more than preset duration, when being judged as YES, in acquisition and the target
Hold the content recommendation of matching.
As another possible implementation, object content can also in real time be carried after video is got
Take, to reduce memory cost.
Content recommendation acquisition module 504：For obtaining the content recommendation matched with the object content；
Wherein, obtain the content recommendation that is matched with object content, the content recommendation can including one or
It is multiple.
Can be by calculating the recommendation that object content and the similarity of content recommendation determine to match with object content
Content.
Preferably, in order to further improve the degree of accuracy of content recommendation, it can also obtain and the target
User characteristics is combined during the content recommendation of content matching, namely is obtained and object content and user
The content recommendation that feature is matched.
Trigger display module 505：For when the object content is played, triggering the player in institute
The prompt message of the content recommendation is exported in the corresponding viewing area for stating object content.
The prompt message of the content recommendation can be the content recommendation or comprising the content recommendation
Summary info and/or content recommendation link address items for information, can by clicking on described information entry
To check the content recommendation.Therefore, described device can also include：
Display unit：Can be when player plays are to object content, while output information entry.Tool
Body can be as shown in Figure 2.
Wherein, user can also choose whether to export the content recommendation in player, in order to avoid to user's
Actual play causes other influences.So as to when user's selection exports the content recommendation, rerun
503~505 modules.
There is provided a kind of analysis by video content in the present embodiment, obtain related to video content
Object content, and select from database content recommendation corresponding with the object content, Ran Hou
When playing video, the device of content recommendation is exported.The pattern of fixed recommendation article in the prior art is changed,
User can be allow to obtain more targetedly content recommendation, user is realized and provided when watching video
Commending contents, improve the conversion ratio for recommending article.
As another embodiment, in order to further improve the degree of accuracy of recommendation information, the content recommendation
During with reference to user characteristics, module 504 can also include being different from Fig. 5, several units as shown in Figure 6：
Arrangement units 5041：For according to the order with the object content similarity from high to low, it is determined that
The multiple pictures matched with the object content；
First acquisition unit 5042：For the user characteristics for the user for obtaining the triggering video playback request；
Assembled unit 5043：For each picture, by the user characteristics, picture correspondence candidate
The content characteristic of the content characteristic of content and the object content is combined, and obtains recommended characteristics；Institute
Stating the content characteristic of alternating content at least includes history recommendation rate；The content characteristic of the object content includes
There is the similarity of duration and/or the object content and the picture in the object content；
Estimate unit 5044：For the recommended models using training in advance, each recommended characteristics is estimated
Recommendation rate.
Selecting unit 5045：For selecting the corresponding content characteristic of recommendation rate highest picture interior as recommending
Hold.
The recommendation rate highest content can be elected as after the recommendation rate of recommended characteristics has been estimated in recommendation
Hold, it is content recommendation that can also select recommendation rate ranking more forward multiple contents.
In the present embodiment, with reference to user characteristics to obtaining the recommendation related to user characteristics, video content
Content, the content recommendation specific aim is stronger, and accuracy is also stronger, it is hereby achieved that stronger recommendation
Effect.
Device embodiment described above is only schematical, wherein described illustrate as separating component
Unit can be or may not be it is physically separate, the part shown as unit can be or
Person may not be physical location, you can with positioned at a place, or can also be distributed to multiple networks
On unit.Some or all of module therein can be selected to realize the present embodiment according to the actual needs
The purpose of scheme.Those of ordinary skill in the art are not in the case where paying performing creative labour, you can with
Understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each reality
The mode of applying can add the mode of required general hardware platform to realize by software, naturally it is also possible to by hard
Part.Understood based on such, the portion that above-mentioned technical proposal substantially contributes to prior art in other words
Dividing can be embodied in the form of software product, and the computer software product can be stored in computer can
Read in storage medium, such as ROM/RAM, magnetic disc, CD, including some instructions are to cause one
Computer equipment (can be personal computer, server, or network equipment etc.) performs each implementation
Method described in some parts of example or embodiment.
Finally it should be noted that：The above embodiments are merely illustrative of the technical solutions of the present invention, rather than to it
Limitation；Although the present invention is described in detail with reference to the foregoing embodiments, the ordinary skill of this area
Personnel should be understood：It can still modify to the technical scheme described in foregoing embodiments, or
Person carries out equivalent substitution to which part technical characteristic；And these modifications or replacement, do not make corresponding skill
The essence of art scheme departs from the spirit and scope of various embodiments of the present invention technical scheme.
Claims (10)
1. a kind of content recommendation method, it is characterised in that including：
Receive the video playback request of player；
Ask the video played to send to player during the video playback is asked, trigger the player
Play the video；
Detect and obtain the object content in the video；
Obtain the content recommendation matched with the object content；
When the object content is played, the player is triggered in the corresponding viewing area of the object content
The prompt message of the content recommendation is exported in domain.
2. method according to claim 1, it is characterised in that the acquisition and the object content
The content recommendation of matching includes：
According to the order with the object content similarity from high to low, it is determined that being matched with the object content
Multiple pictures；
Obtain the user characteristics of the user of the triggering video playback request；
For each picture, by the user characteristics, the content characteristic of picture correspondence alternating content
And the content characteristic of the object content is combined, and obtains recommended characteristics；The alternating content it is interior
Holding feature includes history recommendation rate and/or content type；The content characteristic of the object content includes the mesh
There is the similarity of duration and/or the object content and the picture in mark content；
Using the recommended models of training in advance, the recommendation rate of each recommended characteristics is estimated；
The corresponding alternating content of selection recommendation rate highest picture is used as content recommendation.
3. method according to claim 1, it is characterised in that the detection is simultaneously obtained in the video
Object content, including：
Detect and obtain the object content extracted in advance in the video；
The object content is extracted in advance from the video as follows：
Extract the key frame in the video；
Using the object content detection model of training in advance, object content is extracted from the key frame.
4. method according to claim 3, it is characterised in that described to be extracted from key frame in target
After appearance, methods described also includes：
Target following is carried out to the object content, appearance of the object content in the video is determined
Duration；
The content recommendation matched with the object content that obtains includes：
When the object content duration occurs more than preset duration, obtain what is matched with the object content
Content recommendation.
5. method according to claim 3, it is characterised in that the key in the extraction video
Frame, including：
It regard the first two field picture frame in the video as key frame；
For each two field picture frame after key frame, the comentropy of described image frame is calculated；
When described information entropy is more than entropy threshold, the phase of any frame picture frame and the key frame is calculated
Like degree；
When the similarity is less than similarity threshold, any frame picture frame is regard as key frame.
6. a kind of content recommendation device, it is characterised in that including：
Playing request receiving module, the video playback for receiving player is asked；
Video sending module, for asking the video played to send to broadcasting during the video playback is asked
Device, triggers video described in the player plays；
Object content detection module, for detecting and obtaining the object content in the video；
Content recommendation acquisition module, for obtaining the content recommendation matched with the object content；
Display module is triggered, for when the object content is played, triggering the player in the mesh
The prompt message of the content recommendation is exported in the corresponding viewing area for marking content.
7. device according to claim 6, it is characterised in that the content recommendation acquisition module bag
Include：
Arrangement units, for according to the order with the object content similarity from high to low, it is determined that and institute
State multiple pictures of object content matching and the content characteristic of picture；
First acquisition unit, the user characteristics of the user for obtaining the triggering video playback request；
Assembled unit, for each picture, by the user characteristics, picture correspondence alternating content
Content characteristic and the content characteristic of the object content be combined, obtain recommended characteristics；It is described to wait
Selecting the content characteristic of content at least includes history recommendation rate；The content characteristic of the object content includes described
There is the similarity of duration and/or the object content and the picture in object content；
Unit is estimated, for the recommended models using training in advance, the recommendation of each recommended characteristics is estimated
Rate；
Selecting unit, for selecting the corresponding content characteristic of recommendation rate highest picture as content recommendation.
8. device according to claim 1, it is characterised in that the object content detection module includes：
Detection unit, for detecting and obtaining the object content extracted in advance in the video；
Described device also includes：
Key-frame extraction module：For extracting the key frame in the video；
Object content extraction module：For the object content detection model using training in advance, closed from described
Object content is extracted in key frame.
9. device according to claim 8, it is characterised in that described device also includes：
Target tracking module, for carrying out target following to the object content, determines the object content
Appearance duration in the video；
In the object content specifically for there is duration more than preset duration in the content recommendation acquisition module
When, obtain the content recommendation matched with the object content.
10. device according to claim 8, it is characterised in that the key-frame extraction module includes：
First determining unit, for regarding the first two field picture frame in the video as key frame；
Comentropy computing unit, for for each two field picture frame after key frame, calculating described image
The comentropy of frame；
Similarity calculated, for when described information entropy is more than entropy threshold, calculating any frame figure
As the similarity of frame and the key frame；
Second determining unit, for when the similarity is less than similarity threshold, by any frame figure
As frame is used as key frame.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610249026.9A CN107305557A (en) | 2016-04-20 | 2016-04-20 | Content recommendation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610249026.9A CN107305557A (en) | 2016-04-20 | 2016-04-20 | Content recommendation method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107305557A true CN107305557A (en) | 2017-10-31 |
Family
ID=60152719
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610249026.9A Pending CN107305557A (en) | 2016-04-20 | 2016-04-20 | Content recommendation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107305557A (en) |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107895019A (en) * | 2017-11-15 | 2018-04-10 | 百度在线网络技术（北京）有限公司 | A kind of information recommendation method, device, server and storage medium |
| CN107895025A (en) * | 2017-11-16 | 2018-04-10 | 成都视达科信息技术有限公司 | A kind of media resource information recommends method and system |
| CN108012162A (en) * | 2017-12-04 | 2018-05-08 | 北京小米移动软件有限公司 | Content recommendation method and device |
| CN108388624A (en) * | 2018-02-12 | 2018-08-10 | 科大讯飞股份有限公司 | Multimedia messages recommend method and device |
| CN108446390A (en) * | 2018-03-22 | 2018-08-24 | 百度在线网络技术（北京）有限公司 | Method and apparatus for pushed information |
| CN108776676A (en) * | 2018-02-02 | 2018-11-09 | 腾讯科技（深圳）有限公司 | Information recommendation method, device, computer-readable medium and electronic equipment |
| CN109543709A (en) * | 2018-10-11 | 2019-03-29 | 成都三零凯天通信实业有限公司 | Rapid picture preprocessing method |
| CN109582862A (en) * | 2018-10-31 | 2019-04-05 | 网易传媒科技（北京）有限公司 | Clicking rate predictor method, medium, system and calculating equipment |
| CN109819304A (en) * | 2017-11-22 | 2019-05-28 | 上海全土豆文化传播有限公司 | Barrage methods of exhibiting and device |
| CN110134828A (en) * | 2019-04-29 | 2019-08-16 | 北京物资学院 | A kind of video undercarriage detection method and device |
| CN110163066A (en) * | 2018-12-07 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Multi-medium data recommended method, device and storage medium |
| CN110505498A (en) * | 2019-09-03 | 2019-11-26 | 腾讯科技（深圳）有限公司 | Processing, playback method, device and the computer-readable medium of video |
| CN110688498A (en) * | 2019-09-27 | 2020-01-14 | 北京奇艺世纪科技有限公司 | Information processing method and device, electronic equipment and storage medium |
| CN110795235A (en) * | 2019-09-25 | 2020-02-14 | 北京邮电大学 | Method and system for deep learning and cooperation of mobile web |
| CN110809187A (en) * | 2019-10-31 | 2020-02-18 | Oppo广东移动通信有限公司 | Video selection method, video selection device, storage medium and electronic equipment |
| CN110968786A (en) * | 2019-11-29 | 2020-04-07 | 百度在线网络技术（北京）有限公司 | Visual information recommendation method, device, equipment and storage medium |
| CN111104550A (en) * | 2018-10-09 | 2020-05-05 | 北京奇虎科技有限公司 | Video recommendation method and device, electronic equipment and computer-readable storage medium |
| CN111291264A (en) * | 2020-01-23 | 2020-06-16 | 腾讯科技（深圳）有限公司 | Access object prediction method and device based on machine learning and computer equipment |
| CN111694983A (en) * | 2020-06-12 | 2020-09-22 | 百度在线网络技术（北京）有限公司 | Information display method, information display device, electronic equipment and storage medium |
| CN111708915A (en) * | 2020-06-12 | 2020-09-25 | 腾讯科技（深圳）有限公司 | Content recommendation method and device, computer equipment and storage medium |
| CN111935526A (en) * | 2019-05-13 | 2020-11-13 | 百度在线网络技术（北京）有限公司 | Video display method, system and storage medium thereof |
| CN112487300A (en) * | 2020-12-18 | 2021-03-12 | 上海众源网络有限公司 | Video recommendation method and device, electronic equipment and storage medium |
| CN113158061A (en) * | 2021-05-08 | 2021-07-23 | 王云森 | Data processing method and device |
| CN113378002A (en) * | 2021-08-11 | 2021-09-10 | 北京达佳互联信息技术有限公司 | Information delivery method and device, electronic equipment and storage medium |
| CN113763103A (en) * | 2021-01-19 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Method and device for determining collocation relationship of articles, electronic equipment and storage medium |
| CN115134636A (en) * | 2022-08-29 | 2022-09-30 | 北京达佳互联信息技术有限公司 | Information recommendation method and device |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101833552A (en) * | 2009-03-10 | 2010-09-15 | 郝瑞林 | Method for marking and recommending streaming media |
| CN103020172A (en) * | 2012-11-28 | 2013-04-03 | 北京京东世纪贸易有限公司 | Method and device utilizing video information to search articles |
| WO2013059904A1 (en) * | 2011-10-28 | 2013-05-02 | Research In Motion Limited | Factor-graph based matching systems and methods |
| CN103218734A (en) * | 2013-04-01 | 2013-07-24 | 天脉聚源(北京)传媒科技有限公司 | Advertisement information pushing method and device |
| CN103679754A (en) * | 2013-12-20 | 2014-03-26 | 河北汉光重工有限责任公司 | Real-time target searching and tracking technique |
| CN103839255A (en) * | 2013-12-05 | 2014-06-04 | 福建师范大学 | Method and device for detecting video image matting tampering |
| CN104090929A (en) * | 2014-06-23 | 2014-10-08 | 吕志雪 | Recommendation method and device of personalized picture |
| CN104156978A (en) * | 2014-07-04 | 2014-11-19 | 合肥工业大学 | Multi-target dynamic following method based on balloon platform |
| CN104715023A (en) * | 2015-03-02 | 2015-06-17 | 北京奇艺世纪科技有限公司 | Commodity recommendation method and system based on video content |
-
2016
- 2016-04-20 CN CN201610249026.9A patent/CN107305557A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101833552A (en) * | 2009-03-10 | 2010-09-15 | 郝瑞林 | Method for marking and recommending streaming media |
| WO2013059904A1 (en) * | 2011-10-28 | 2013-05-02 | Research In Motion Limited | Factor-graph based matching systems and methods |
| CN103020172A (en) * | 2012-11-28 | 2013-04-03 | 北京京东世纪贸易有限公司 | Method and device utilizing video information to search articles |
| CN103218734A (en) * | 2013-04-01 | 2013-07-24 | 天脉聚源(北京)传媒科技有限公司 | Advertisement information pushing method and device |
| CN103839255A (en) * | 2013-12-05 | 2014-06-04 | 福建师范大学 | Method and device for detecting video image matting tampering |
| CN103679754A (en) * | 2013-12-20 | 2014-03-26 | 河北汉光重工有限责任公司 | Real-time target searching and tracking technique |
| CN104090929A (en) * | 2014-06-23 | 2014-10-08 | 吕志雪 | Recommendation method and device of personalized picture |
| CN104156978A (en) * | 2014-07-04 | 2014-11-19 | 合肥工业大学 | Multi-target dynamic following method based on balloon platform |
| CN104715023A (en) * | 2015-03-02 | 2015-06-17 | 北京奇艺世纪科技有限公司 | Commodity recommendation method and system based on video content |
Cited By (36)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107895019A (en) * | 2017-11-15 | 2018-04-10 | 百度在线网络技术（北京）有限公司 | A kind of information recommendation method, device, server and storage medium |
| CN107895025A (en) * | 2017-11-16 | 2018-04-10 | 成都视达科信息技术有限公司 | A kind of media resource information recommends method and system |
| CN109819304A (en) * | 2017-11-22 | 2019-05-28 | 上海全土豆文化传播有限公司 | Barrage methods of exhibiting and device |
| CN108012162A (en) * | 2017-12-04 | 2018-05-08 | 北京小米移动软件有限公司 | Content recommendation method and device |
| CN108776676B (en) * | 2018-02-02 | 2020-05-05 | 腾讯科技（深圳）有限公司 | Information recommendation method and device, computer readable medium and electronic device |
| CN108776676A (en) * | 2018-02-02 | 2018-11-09 | 腾讯科技（深圳）有限公司 | Information recommendation method, device, computer-readable medium and electronic equipment |
| CN108388624A (en) * | 2018-02-12 | 2018-08-10 | 科大讯飞股份有限公司 | Multimedia messages recommend method and device |
| CN108446390A (en) * | 2018-03-22 | 2018-08-24 | 百度在线网络技术（北京）有限公司 | Method and apparatus for pushed information |
| CN108446390B (en) * | 2018-03-22 | 2022-01-04 | 百度在线网络技术（北京）有限公司 | Method and device for pushing information |
| CN111104550A (en) * | 2018-10-09 | 2020-05-05 | 北京奇虎科技有限公司 | Video recommendation method and device, electronic equipment and computer-readable storage medium |
| CN109543709A (en) * | 2018-10-11 | 2019-03-29 | 成都三零凯天通信实业有限公司 | Rapid picture preprocessing method |
| CN109582862A (en) * | 2018-10-31 | 2019-04-05 | 网易传媒科技（北京）有限公司 | Clicking rate predictor method, medium, system and calculating equipment |
| CN110163066A (en) * | 2018-12-07 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Multi-medium data recommended method, device and storage medium |
| CN110134828A (en) * | 2019-04-29 | 2019-08-16 | 北京物资学院 | A kind of video undercarriage detection method and device |
| CN110134828B (en) * | 2019-04-29 | 2021-02-23 | 北京物资学院 | Video off-shelf detection method and device |
| CN111935526A (en) * | 2019-05-13 | 2020-11-13 | 百度在线网络技术（北京）有限公司 | Video display method, system and storage medium thereof |
| CN110505498A (en) * | 2019-09-03 | 2019-11-26 | 腾讯科技（深圳）有限公司 | Processing, playback method, device and the computer-readable medium of video |
| CN110795235A (en) * | 2019-09-25 | 2020-02-14 | 北京邮电大学 | Method and system for deep learning and cooperation of mobile web |
| CN110795235B (en) * | 2019-09-25 | 2022-07-19 | 北京邮电大学 | Method and system for deep learning and cooperation of mobile web |
| CN110688498A (en) * | 2019-09-27 | 2020-01-14 | 北京奇艺世纪科技有限公司 | Information processing method and device, electronic equipment and storage medium |
| CN110688498B (en) * | 2019-09-27 | 2022-07-22 | 北京奇艺世纪科技有限公司 | Information processing method and device, electronic equipment and storage medium |
| CN110809187A (en) * | 2019-10-31 | 2020-02-18 | Oppo广东移动通信有限公司 | Video selection method, video selection device, storage medium and electronic equipment |
| CN110968786A (en) * | 2019-11-29 | 2020-04-07 | 百度在线网络技术（北京）有限公司 | Visual information recommendation method, device, equipment and storage medium |
| CN110968786B (en) * | 2019-11-29 | 2023-10-17 | 百度在线网络技术（北京）有限公司 | Visual information recommendation method, device, equipment and storage medium |
| CN111291264A (en) * | 2020-01-23 | 2020-06-16 | 腾讯科技（深圳）有限公司 | Access object prediction method and device based on machine learning and computer equipment |
| CN111291264B (en) * | 2020-01-23 | 2023-06-23 | 腾讯科技（深圳）有限公司 | Access object prediction method and device based on machine learning and computer equipment |
| CN111694983A (en) * | 2020-06-12 | 2020-09-22 | 百度在线网络技术（北京）有限公司 | Information display method, information display device, electronic equipment and storage medium |
| CN111708915A (en) * | 2020-06-12 | 2020-09-25 | 腾讯科技（深圳）有限公司 | Content recommendation method and device, computer equipment and storage medium |
| CN111694983B (en) * | 2020-06-12 | 2023-12-19 | 百度在线网络技术（北京）有限公司 | Information display method, information display device, electronic equipment and storage medium |
| CN112487300A (en) * | 2020-12-18 | 2021-03-12 | 上海众源网络有限公司 | Video recommendation method and device, electronic equipment and storage medium |
| CN113763103A (en) * | 2021-01-19 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Method and device for determining collocation relationship of articles, electronic equipment and storage medium |
| CN113158061A (en) * | 2021-05-08 | 2021-07-23 | 王云森 | Data processing method and device |
| CN113378002A (en) * | 2021-08-11 | 2021-09-10 | 北京达佳互联信息技术有限公司 | Information delivery method and device, electronic equipment and storage medium |
| US11632586B2 (en) | 2021-08-11 | 2023-04-18 | Beijing Dajia Internet Information Technology Co., Ltd. | Method for placing delivery information, electronic device, and storage medium |
| CN115134636A (en) * | 2022-08-29 | 2022-09-30 | 北京达佳互联信息技术有限公司 | Information recommendation method and device |
| CN115134636B (en) * | 2022-08-29 | 2023-01-20 | 北京达佳互联信息技术有限公司 | Information recommendation method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107305557A (en) | Content recommendation method and device | |
| US10169371B2 (en) | System and method for creating a preference profile from shared images | |
| CN106776619B (en) | Method and device for determining attribute information of target object | |
| US8738553B1 (en) | Image selection based on image quality | |
| CN108629224B (en) | Information demonstrating method and device | |
| CN111798879B (en) | Method and apparatus for generating video | |
| CN110390033B (en) | Training method and device for image classification model, electronic equipment and storage medium | |
| CN110019943B (en) | Video recommendation method and device, electronic equipment and storage medium | |
| CN111212303B (en) | Video recommendation method, server and computer-readable storage medium | |
| EP2568429A1 (en) | Method and system for pushing individual advertisement based on user interest learning | |
| CN104715023A (en) | Commodity recommendation method and system based on video content | |
| KR101385700B1 (en) | Method and apparatus for providing moving image advertisements | |
| CN110737783A (en) | method, device and computing equipment for recommending multimedia content | |
| CN109933782B (en) | User emotion prediction method and device | |
| CN113742567B (en) | Recommendation method and device for multimedia resources, electronic equipment and storage medium | |
| CN111984824A (en) | Multi-mode-based video recommendation method | |
| CN112364204A (en) | Video searching method and device, computer equipment and storage medium | |
| US20210326646A1 (en) | Automated generation of training data for contextually generated perceptions | |
| Akputu et al. | Emotion recognition using multiple kernel learning toward E-learning applications | |
| CN111597446B (en) | Content pushing method and device based on artificial intelligence, server and storage medium | |
| Yousaf et al. | Patch-CNN: Deep learning for logo detection and brand recognition | |
| CN112131477A (en) | Library book recommendation system and method based on user portrait | |
| McAllister et al. | Towards personalised training of machine learning algorithms for food image classification using a smartphone camera | |
| CN113704623B (en) | Data recommendation method, device, equipment and storage medium | |
| CN115129902A (en) | Media data processing method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20171031 Assignee: Apple R&D (Beijing) Co., Ltd. Assignor: BEIJING MOSHANGHUA TECHNOLOGY CO., LTD. Contract record no.: 2019990000054 Denomination of invention: Content recommending method and device based on interest point change License type: Exclusive License Record date: 20190211 |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171031 |