CN110008375A - Video is recommended to recall method and apparatus - Google Patents
Video is recommended to recall method and apparatus Download PDFInfo
- Publication number
- CN110008375A CN110008375A CN201910224166.4A CN201910224166A CN110008375A CN 110008375 A CN110008375 A CN 110008375A CN 201910224166 A CN201910224166 A CN 201910224166A CN 110008375 A CN110008375 A CN 110008375A
- Authority
- CN
- China
- Prior art keywords
- video
- vector
- user
- portrait
- cluster
- 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
- 238000000034 method Methods 0.000 title claims abstract description 54
- 239000013598 vector Substances 0.000 claims abstract description 324
- 238000010276 construction Methods 0.000 claims description 18
- 238000013528 artificial neural network Methods 0.000 claims description 17
- 238000004364 calculation method Methods 0.000 claims description 16
- 238000000354 decomposition reaction Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 2
- 208000001491 myopia Diseases 0.000 description 21
- 238000013527 convolutional neural network Methods 0.000 description 14
- 238000010586 diagram Methods 0.000 description 10
- 230000004913 activation Effects 0.000 description 8
- 230000006870 function Effects 0.000 description 8
- 230000008569 process Effects 0.000 description 8
- 230000006399 behavior Effects 0.000 description 6
- 238000004590 computer program Methods 0.000 description 5
- 230000004927 fusion Effects 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 239000000463 material Substances 0.000 description 4
- 238000000605 extraction Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 239000000203 mixture Substances 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000013517 stratification Methods 0.000 description 2
- 206010000234 Abortion spontaneous Diseases 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000035800 maturation Effects 0.000 description 1
- 208000015994 miscarriage Diseases 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000011524 similarity measure Methods 0.000 description 1
- 208000000995 spontaneous abortion Diseases 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
It discloses a kind of recommendation video based on vector and recalls method and apparatus.The described method includes: the current video based on user clicks history, video presentation vector corresponding with each video is searched；According to the multiple video presentation vectors found, at least one active user portrait vector of the user is generated；And recall video presentation vector Candidate Recommendation video similar at least one described active user portrait vector.Pass through the description vectors that video and user are drawn a portrait, the present invention can fast and efficiently realize personalized recommendation on the basis of lacking user's portrait, and it can be conveniently based on user video viewing history building user's portrait vector, and the portrait of the user based on vectorization easily recalls the recommendation video with similar vector.
Description
Technical field
The present invention relates to recommender systems more particularly to a kind of recommendation video based on vector to recall method and apparatus.
Background technique
Video industry develops rapidly in recent years, wherein short-sighted frequency You Yiqi sociability and propagate its belief on a large scale, and appear in one
Large quantities of short-sighted frequency social applications based on the short-sighted frequency of the personal situation of distribution, this kind of application are mainly social based on short video consumer
Supplemented by carry out system building.
In order to improve user-friendliness and reduce inefficient hind computation, these applications need to exist suitable video content
The suitable time recommends suitable user in an appropriate manner.Content properly refers to that content agrees with target user's taste, time
Properly refer to the context that recommendation is done something for the occasion at that time, mode properly refers to that the content ways of presentation of recommendation allows user to feel easypro
Suitable, user properly refers to the user of recommendation really to having stronger degree of recognition in recommended.For this purpose, before efficiently being recommended
Mentioning is to understand user and video content, this needs to create user's portrait just to state user and be divided video content
It class or is labeled.
In the prior art, all kinds of recommendations recall method all and be calculated offline for each user according to user's portrait it is each
Zipper is recalled, these are recalled zipper later and is stored in caching, is drawn a portrait when recommending according to user and is pulled not with context is recommended
Zipper conduct of recalling together is recalled.This mode of recalling needs to store a plurality of of each user and recalls zipper, usually each recalls
Zipper length all thousand ranks and be not aware that when off-line calculation recalls zipper user within the next recommendation period whether
It will do it application access, so it is generally necessary to calculating 90 days or 60 days all any active ues.For medium-and-large-sized userbase
For, as above storage resource needed for operation is considerable, but effective rate of utilization is not high.In addition, such recall mode
It needs to expend huge manpower and material resources and carries out user's portrait construction, and be difficult to realize in platform initial go-live period.
For this reason, it may be necessary to a kind of more light weight and efficient video be recommended to recall scheme.
Summary of the invention
In view of this, the invention proposes a kind of videos based on vector to recall algorithm, user's portrait can lacked
On the basis of fast and efficiently realize personalized recommendation.By the way that ground can be facilitated by the description vectors of video and user's portrait
In user video viewing history building user's portrait vector, and light recall of the portrait of the user based on vectorization has similar vector
Recommendation video.
According to an aspect of the present invention, it proposes a kind of recommendation videos to recall method, comprising: works as forward sight based on user
Frequency point hits history, searches video presentation vector corresponding with each video；According to the multiple video presentation vectors found, generate
At least one active user portrait vector of the user；And recall video presentation vector and at least one described active user picture
As the similar Candidate Recommendation video of vector.As a result, by that can be conveniently based on the description vectors of video and user's portrait
User video watches history building user's portrait vector, and the portrait of the user based on vectorization is easily recalled with similar vector
Recommend video.
Preferably, video presentation vector Candidate Recommendation similar at least one described active user portrait vector is recalled to regard
Frequency includes: to recall a Candidate Recommendation video zipper based on each of multiple active users portrait vector；And fusion is more
Current similar recommendation video of the Candidate Recommendation video zipper as the user, for example, being drawn a portrait vector based on active user
Weight determines the accounting and/or presentation mode of each Candidate Recommendation video zipper in similar recommendation video, wherein each current
The weight of user's portrait vector is determined based at least one of following: generating the video presentation vector of active user portrait vector
Number；The video presentation vector for generating active user portrait vector corresponds to the viewing time of video.Thereby, it is possible to based on difference
User preference, obtain and corresponding recommend video.
Preferably, the method for recalling of the invention can also include: to calculate each existing video for multiple existing videos
Video presentation vector；And construction video ID corresponds to table to its video vector for corresponding to video presentation vector.Thus it is possible to be based on
User's current video click history in include video ID, corresponded to from the video vector searched in table corresponding video presentation to
Amount.
Preferably, the method for recalling of the invention can also include: to gather to the video presentation vector of multiple existing videos
Class；And the Video clustering of construction video ID to its affiliated cluster ID correspond to table.
Then, recall operation may include: to calculate cluster belonging to each active user's portrait vector；Based on each affiliated
Cluster corresponded to from the Video clustering and find out the included video ID of each cluster in table；Based on the video ID for being included from institute
It states video vector and corresponds to and find out corresponding video presentation vector in table；It calculates each active user's portrait and its affiliated cluster is interior each
The similitude of the video presentation vector of video；And the video ID of the highest predetermined number of similarity is returned as the active user
The Candidate Recommendation video zipper that portrait vector is recalled.As a result, platform need to only store two corresponding tables, so that it may based on use
The video historical data at family easily carries out the generation of active user portrait and recommends recalling for video.
Preferably, the method for recalling of the invention can also include: the density being distributed in each cluster based on newly-increased video
Degree, the cluster for carrying out merging or the decomposition of existing cluster update operation；And institute is updated according to the result that cluster updates operation
It states Video clustering and corresponds to table.As a result, cluster is capable of the content of video belonging to reflection more promptly and accurately, and promoted subsequent
The computational efficiency and accuracy of recall operation.
Active user's portrait vector of user can be generated in real time under following at least one occasion and recall corresponding time
Video is recommended in choosing: in the case where user opens Video Applications；In the case where user newly clicks the video of predetermined number；With
And in predetermined time interval under.As a result, be based on concrete application scene and system performance, greatly promoted recommendation real-time and
Flexibility.
Preferably, the corresponding video presentation vector of each video is the key that from artificial neural network (ANN) to the video
The key frame vector for the classification results construction that frame is classified, and active user's portrait vector of user is from the user
Merging and/or the cluster of the key frame vector of the video of viewing and generate.Thus, it is possible to using maturation ANN (for example, CNN schemes
As disaggregated model) feature extraction functions, construct the key frame vector for accurately reflecting picture material, and for subsequent merging and
The accuracy of operation is recommended to provide basis.
According to another aspect of the present invention, it proposes a kind of recommendation video and recalls device, comprising: video presentation vector is looked into
Unit is looked for, history is clicked for the current video based on user, searches video presentation vector corresponding with each video；User draws
As vector generation unit, for generating at least one active user of the user according to the multiple video presentation vectors found
Portrait vector；And Candidate Recommendation video recalls unit, for recalling video presentation vector and at least one described active user
The similar Candidate Recommendation video of portrait vector.
Preferably, the Candidate Recommendation video, which recalls unit, can be used for: based in multiple active users portrait vector
Each recalls a Candidate Recommendation video zipper；And a plurality of Candidate Recommendation video zipper of fusion is as the current of the user
Similar recommendation video.
Preferably, the mixing operation that the Candidate Recommendation video recalls unit may include: based on active user draw a portrait to
The weight of amount determines the accounting and/or presentation mode of each Candidate Recommendation video zipper in similar recommendation video, wherein each
The draw a portrait weight of vector of active user is based at least one of following determine: generate the video presentation of active user portrait vector to
The number of amount；And it generates the draw a portrait video presentation vector of vector of the active user and corresponds to the viewing time of video.
Preferably, recalling device can also include: video presentation vector generation unit and cluster calculation unit, wherein view
Frequency description vectors generation unit is used for: being directed to multiple existing videos, is calculated the video presentation vector of each existing video；And structure
It makes video ID and corresponds to table to its video vector for corresponding to video presentation vector, the cluster calculation unit is used for: to multiple existing
The video presentation vector of video is clustered；And the Video clustering of construction video ID to its affiliated cluster ID correspond to table.
Preferably, video presentation vector search unit can be used for: based on user's current video click history in include
Video ID is corresponded to from the video vector and is searched corresponding video presentation vector in table, and the Candidate Recommendation video is recalled
Unit is used for: calculating cluster belonging to each active user's portrait vector；Based on each affiliated cluster from the Video clustering
The included video ID of each cluster is found out in corresponding table；It is corresponded in table and is looked for from the video vector based on the video ID for being included
Corresponding video presentation vector out；The video presentation vector of each video in calculating each active user's portrait and being clustered belonging to it
Similitude；And return to the candidate that the video ID of the highest predetermined number of similarity is recalled as active user portrait vector
Recommend video zipper.
Preferably, the cluster calculation unit can be also used for: the density being distributed in each cluster based on newly-increased video
Degree, the cluster for carrying out merging or the decomposition of existing cluster update operation；And institute is updated according to the result that cluster updates operation
It states Video clustering and corresponds to table.
Preferably, the video presentation vector generation unit is from artificial neural network (ANN) to the key frame of a certain video
Video presentation vector of the classification results construction key frame vector classified as the video, and user portrait vector
The key frame vector for the video that generation unit watches a certain user merges and/or clusters the active user for generating the user
Portrait vector.
Preferably, the real-time generation recalled device and carry out active user's portrait vector under following at least one occasion
Recommend recalling in real time for video with corresponding candidate: in the case where user opens Video Applications；Predetermined is newly clicked in user
In the case where several videos；And in predetermined time interval under.
According to a further aspect of the invention, a kind of calculating equipment is proposed, comprising: processor；And memory, thereon
It is stored with executable code, when the executable code is executed by the processor, executes the processor as described above
Recommendation video recall method.
According to a further aspect of the invention, a kind of non-transitory machinable medium is proposed, is stored thereon with
Executable code executes the processor as described above when the executable code is executed by the processor of electronic equipment
Recommendation video recall method.
The present invention indicates video content using vector by the vectorization of video content, being capable of broadcasting in real time according to user
It puts history and calculates the vector portrait of user in real time, and indicate to carry out similitude recommendation using specific vector.By the way that behaviour will be recalled
Make to be changed into the operation for vector, so that entirely framework is recommended more to be simple and efficient, and cold open can be solved to a certain extent
Dynamic problem.
Detailed description of the invention
Disclosure illustrative embodiments are described in more detail in conjunction with the accompanying drawings, the disclosure above-mentioned and its
Its purpose, feature and advantage will be apparent, wherein in disclosure illustrative embodiments, identical reference label
Typically represent same parts.
Fig. 1 shows the flow diagram that recommendation video according to an embodiment of the invention recalls method.
Fig. 2 shows the examples that key frame generates.
Fig. 3 shows the example of the group stratification of typical CNN.
Fig. 4 shows an example of memory data structure building process of the invention.
Fig. 5 shows an example for recalling calculation process in real time of the invention.
Fig. 6 shows the composition block diagram that recommendation video according to an embodiment of the invention recalls device.
Fig. 7, which is shown, can be used for realizing according to an embodiment of the present invention the calculating equipment that above-mentioned recommendation video recalls method
Structural schematic diagram.
Specific embodiment
The preferred embodiment of the disclosure is more fully described below with reference to accompanying drawings.Although showing the disclosure in attached drawing
Preferred embodiment, however, it is to be appreciated that may be realized in various forms the disclosure without the embodiment party that should be illustrated here
Formula is limited.On the contrary, these embodiments are provided so that this disclosure will be more thorough and complete, and can be by the disclosure
Range is completely communicated to those skilled in the art.
In recent years due to Feed miscarriage product it is prevailing, individual character recommend have become video class application (for example, cell phone application) mark
Match.A critically important link is exactly to recall in personalized recommendation, is to be obtained according to the broadcasting context of user's portrait or user
Take a collection of video that family most possibly then plays.
Currently there are all kinds of recommendations to recall method all and be to be calculated offline for each user according to user's portrait and each recall drawing
These are recalled zipper later and are stored in caching by chain, pull different call together from context is recommended according to user's portrait when recommending
Chain of pulling back is used as and recalls.This mode of recalling needs to store a plurality of of each user and recalls zipper, and it is long usually each to recall zipper
Degree all in thousand ranks and when off-line calculation recalls zipper and is unaware of user and whether can access this in the next recommendation period
Using so it is generally necessary to calculating 90 days or 60 days all any active ues.For the application of medium-and-large-sized userbase, this
Storage resource needed for kind method is considerable, but effective rate of utilization is lower.In addition, such mode of recalling need to expend it is huge
Manpower and material resources carry out user's portrait construction, and are difficult to realize in platform initial go-live period.
For this purpose, the invention proposes a kind of real-time videos based on vector to recall algorithm, it can lack user's portrait
On the basis of fast and efficiently realize personalized recommendation, without with lie in consumer consumption behavior number, solve product initial stage it is cold
The problem of starting.
Specifically, the scheme of recalling of the invention can be the encoding video content that needs are recommended at vector, each video item
A corresponding vector, these vectors have following characteristic: a. indicates the content of a video item by multi-C vector, and b. is in video item
Vector space in the similar vector of content assemble together, their similarity (for example, cosine similarity value) is all bigger.
It can be according to the associated video that the play history of user, recommended user may like to realize similar recommendation when recommending to recall.
This will greatly reduce the demand of memory space because for recommendation alternating content pond size with 90 days in number of users understand phase
The several orders of magnitude of difference.It calculates in real time and recommends to recall and can for greater flexibility feed back the nearest behavior of user, can preferably capture
The interest of user has significant effect especially for new user, and is capable of the interest of fast understanding user.
Fig. 1 shows the flow diagram that recommendation video according to an embodiment of the invention recalls method.This method is logical
The vectorization operation for video and user's portrait is crossed, can be realized and recommend recalling in real time for video, thus greatly reduce existing
There is technology kind to draw a portrait and recall the storage needs of zipper for user, and improves the flexibility and accuracy recalled.
In step S110, current video based on user clicks history, search video presentation corresponding with each video to
Amount.In step S120, according to the multiple video presentation vectors found, generate the user at least one active user draw a portrait to
Amount.In step S130, recalls video presentation vector Candidate Recommendation similar at least one described active user portrait vector and regard
Frequently.
In step S110, the lookup of video presentation vector may include clicking history according to the video of user first, obtain
The video information that the user watched searches video presentation corresponding to video then further according to the video information of acquisition
Vector.Here, the viewing history of different range can be obtained based on different strategies.For example, newly using an application in user
When, the video information for each video that available user is clicked.With increasing for user click data, can choose most
A recent period of time (for example, nearest seven days) or predetermined number (for example, 100 the watched recently) video watched recently simultaneously obtain
Corresponding video information.In one embodiment, it can also be now to further according to the viewing completeness of video for calculating
The video of user's portrait vector.For example, can choose user click all videos, can also only select user watched or
Watch the video that completeness is more than predetermined percentage or duration.
Here, video presentation vector is capable of the vector of describing video contents, can be by image contained by video into
Row analysis and generate.Here, can handle video that is existing in video library and/or newly uploading, to obtain video
Corresponding video presentation vector.In one embodiment, the key frame vector of video can be used as video of the invention
Description vectors.Video content description, which is carried out, using key frame vector is particularly suitable for typically more single short-sighted of Behaviour theme
Frequently.
Fig. 2 shows the examples that key frame generates.In step S210, the key frame of video is extracted.Here, can be
For the respective key frame of multiple video extractions.For the considerations of calculating cost, preferred pin is to one key of each video selection
Frame is handled.In the case where short-sighted frequency has cover, cover can be directly selected as key frame.As replacement or mend
It fills, a certain frame in known key frame extraction algorithm (for example, ffmpeg) selecting video also can be used as the video
Key frame.
In step S220, classified using ANN classification device to each key frame.The ANN being used in the present invention
Classifier especially can be the CNN classifier suitable for image classification.
Fig. 3 shows the example of the group stratification of typical CNN.As shown in figure 3, typical CNN is by a series of orderly functions
Layer composition.
CNN neural network is composed in series by input layer, output layer and multiple hidden layers.The first layer of CNN reads input value,
Such as input picture, and export a series of activation value (alternatively referred to as characteristic pattern).What following layer reading was generated by upper one layer
Activation value, and export new activation value.The last one classifier (classifier) export the input picture may belong to it is every
A kind of other probability.
These layers are broadly divided into the layer (such as convolutional layer, full articulamentum, batch normalization layer) and not cum rights of Weight
The layer (such as pond layer, ReLU layers, Softmax layers) of weight.Among these, CONV layers of (Convolutional layers, convolution
Layer) using series of features figure as input, and output activation value is obtained with convolution kernels convolution.Pond layer usually with CONV layers of phase
Even, it for exporting the maximum value or average value of each subregion in each characteristic pattern (sub area), is dropped from there through sub-sampling
Low calculation amount, while keeping displacement, scale and deformation invariance to a certain degree.It may include convolutional layer and pond in one CNN
Change multiple alternatings between layer, thus gradually reduces spatial resolution and increase the quantity of Feature Mapping.Then can connect to
At least one full articulamentum obtains including the one-dimensional of multiple characteristic values by the linear transformation being applied on input feature value
Vector output.
On the whole, the operation of the layer of Weight can indicate are as follows:
Y=WX+b,
Wherein W is weighted value, and b is biasing, and X is input activation value, and Y is output activation value.
The operation of the layer of Weight can not indicate are as follows:
Y=f (X),
Wherein f (X) is nonlinear function.
Here, " weight " (weights) refers to the parameter in hidden layer, it is logical that understanding in a broad sense, which may include biasing,
The numerical value of training process acquistion is crossed, and is remained unchanged in reasoning；Activation value refers to each layer of the output since input layer
It is obtained by input value and weighted value by operation, the numerical value transmitted between the layers, also referred to as characteristic value.It is different from weighted value,
The distribution of activation value can according to input data sample dynamic change.
Before making inferences (for example, image classification) using CNN, it is necessary first to be trained to CNN.Pass through training number
According to a large amount of importings, determine the parameter of each layer of neural network model, such as weight and biasing.
In one embodiment, the present invention can be directly using the picture recognition model of TensorFlow come to short-sighted frequency
Key frame carries out image recognition to obtain each classification and its probability value.Used CNN model can be in the industry it is generally acknowledged at
Ripe model, such as the CNN model Inception of Google open source, such as utilize the number in large-scale image data base ImageNet
It is formed according to training and error rate is only 3.5% V3 version.It can be with using existing model (for example, Inception_v3 model)
Industry achievement is directly utilized, to substantially reduce the cost and calculate power demand.
In step S230, each respective key frame vector of video can be constructed based on classification results.
Since classification results can embody the image correlation between key frame well, and being converted to vector can be from
Multiple dimensions embody above-mentioned correlation, therefore suggested design of the invention even can be more more objective and accurate than manual sort
Find out the correlation between video content.In one embodiment, can using class categories as vector dimension, and will classification belonging to
Probability value constructs key frame vector as vector classification value.It in a preferred embodiment, can be by seeking between vector
Cosine similarity characterizes the similarity of vector.
It, can be 1000 classifications of image point, each in the case where for example being classified using Inception_v3 model
Classification has the probability value of a generic.If using this 1000 classifications short-sighted frequency if the dimension of the vector of short-sighted frequency
Vector has 1000 dimensions, and the probability value of generic obtains the vector unique identification one of one 1000 dimension as the value of each dimension in turn
A short-sighted frequency.
If directly vector (for example, the 1000 dimension) vector as above obtained is stored and is calculated, will lead to resource and
Calculate the inefficient use of power.Therefore, vector is simplified in the case where classification accuracy can not be significantly affected.Implement at one
In example, can choose probability value maximum top n classification belonging to classifying in each classification results construct the key frame of the video to
Amount, wherein N is scheduled positive integer, for example, can based on experience value or system calculate power chosen.For example, above-mentioned
In the case where 1000 disaggregated models, for the key frame of each short-sighted frequency, it can only take such as probability value first 10 maximum, preceding
50 or preceding 100 classification, and by other classification value zero setting so that the vector dimensionality reductions of 1000 dimensions to only 10,50 or
It is the sparse vector in 100 dimensions with nonzero value.It as replacement or adds, chooses in each classification results belonging to classification
The classification that probability value is greater than predetermined threshold constructs the key frame vector of the video.For example, can dimension by probability value less than 0.001
Spend zero setting.Equally alternatively or additionally, M after decimal point rounding-offs can be carried out to probability value belonging to each classification,
To construct the respective key frame vector of each video, wherein M is scheduled positive integer.For example, still for 1000 classification
Each component value can be taken 3 decimals (average value under 1000 classification is 0.001) by model, and it is lesser to cast out probability value
Dimension.Thus, it is possible to for each existing video, generate be easy to subsequent arithmetic, the key frame of simplified (for example, dimensionality reduction) to
Amount.
It, can be with it is understood that other than as above generating according to the key frame vector of ANN classification result vector
It is handled according to other video features with extracting method and realizes the vectorization to video presentation.
In addition, though can be dropped by the operation that simplifies as above to video presentation vector (for example, key frame vector)
Dimension, but huge calculation power still can be consumed for the multi-C vector of the short-sighted frequency of magnanimity calculating (for example, calculating of cosine similarity).
As replacement or add, can also be by the way that video vector be grouped, and only carry out Similarity measures in group and come substantially
Power is calculated needed for reducing.For this purpose, in one embodiment, the method for recalling of the invention can also include the view to multiple existing videos
The step of frequency description vectors are clustered, for example, being clustered to its key frame vector.Obviously, the classification number of cluster wants small
In the class categories number of ANN classification device.For example, the k- of 200 clusters can be carried out for the CNN model of 1000 classification
Means clusters (or cluster of other known methods).Preferably, it can be directed to each video, in addition to storing key frame for it
Except vector, the cluster ID belonging to it can also be stored.Further, recall step S130 may include based on user draw a portrait to
Cluster belonging to amount, recommends key frame vector to belong to the similar video in the cluster.It is clustered, can be determined first by introducing
The grouping of user's portrait vector, then degree of approximation calculating is confined in the grouping.
Here, the cluster centre vector of each cluster can be sought in advance, for example, by short-sighted frequency vectors all under the cluster
Addition normalizing obtain.Then, the similarity of user's portrait vector and each cluster centre vector can be calculated, and selects and is somebody's turn to do
User draw a portrait vector it is closest cluster centre vector institute to it is corresponding cluster as cluster belonging to the target video, example
The cluster ID being such as calculated to the video distribution.Thus, it is possible to carry out phase only for the existing video with identical cluster ID
It is sought like degree and video recommendations.
It in one embodiment, can also be to cluster originally with the transfer of amount of video on platform enriched constantly with hot spot
Body is adjusted.Recommended method of the invention can also include the density being distributed in each cluster based on current video as a result,
Degree carries out merging or the operation splitting of existing cluster.For example, the especially more cluster of affiliated video can be carried out further
Refinement is split, and the less similar topic cluster of affiliated video is merged.Above-mentioned cluster update can periodically or according to
Newly-increased number of videos carries out, and after cluster updates, can accordingly update the cluster ID of every video.
The subsequent use to video presentation vector sum cluster data for convenience, the method for recalling of the invention can also include
The preparatory construction of data structure and update, and store it in memory so that subsequent user portrait vector constructs and recommend view
Frequency recalls use.For this purpose, the present invention is being directed to multiple existing videos, after the video presentation vector for calculating each existing video,
Video ID can also be constructed and correspond to table to its video vector for corresponding to video presentation vector.The finding step of S110 can be with as a result,
Include: based on user's current video click history in include video ID, corresponded to from the video vector searched in table it is corresponding
Video presentation vector.Further, the present invention is after the video presentation vector to multiple existing videos clusters, can be with
The Video clustering of construction video ID to its affiliated cluster ID correspond to table.Above-mentioned corresponding table can with the update of video and cluster and
It is updated.
Fig. 4 shows an example of memory data structure building process of the invention.Retouching for each video is obtained first
The ID clustered belonging to vector and each short-sighted frequency is stated, the generation of vector and the generation for clustering ID are as described above.Then, it constructs short-sighted
Mapping data structure and each set that clusters ID to affiliated video ID of the frequency id to short-sighted frequency vector.Here, can be used
Map data structure respectively obtains itemID → vec (video ID to video presentation vector) and clusterID → list
[itemID] (cluster ID gathers to affiliated video ID) two corresponding tables, the query time complexity of each corresponding table is O
(1).Then, above-mentioned two Map data structure is obtained, in memory to facilitate the use of subsequent step S120 and S130.
It may include the video presentation vector for watching the user video in the generation of step S120, user's portrait vector
(for example, key frame vector) is merged or is clustered to realize.Here, merge and cluster based on be similar between vector
Property, for example, cosine similarity.As a result, by the vectorization to picture material contained by video, so realize to user draw a portrait to
Quantization.User's portrait of vectorization can be more objective and directly reaction user is also easier to carry out needle to the preference of video
Pair or based on user portrait various calculating and update.
In one embodiment, the generation of user's portrait vector may include clicking history according to the video that dynamic updates,
Seek to real-time or near real-time user's portrait vector.It is highly preferred that according to the quantity and/or video of the video information of acquisition
The distribution situation of description vectors, dynamic select directly use, merge or cluster the video presentation vector as the user
The operation of portrait vector.
According to the quantity and distribution of the video presentation vector that can be used for generating user's portrait vector, different sides can be used
Method generates user's portrait vector.For example, in the case where user only has viewed a video or only finishes watching a video, it can
Directly to indicate user's portrait vector using the video presentation vector of the video (for example, key frame vector above).With
User viewing number of videos increase, then can flexibly using merge or cluster operation come generate user draw a portrait vector.For example,
When the video that user watched reaches two or more, the similarity between multiple video presentation vectors can be calculated two-by-two
(for example, cosine similarity), and merge similar two or more video presentation vectors as at least one user draw a portrait to
Amount.And when the number of videos that user watched further increases when even up to a hundred a (for example, reach tens), then it can introduce poly-
Class method, such as the similarity based on multiple video presentation vectors cluster multiple video presentation vectors, and seek gathering
Class center is as the portrait vector of user described at least one.
In the case where merging and clustering, since there may be multiple merging and cluster centre (for example, user concentrates
Have viewed multiple videos of three different themes), therefore it is directed to a certain user, it will usually multiple users' portrait vectors are obtained, are used
In the different aspect for showing user's subject of interest.
For same user obtain two or more users draw a portrait vector in the case where, can draw a portrait for each user to
The associated weighted value of the amount distribution video presentation vector number that merges or cluster with it.Since the similar video of user's viewing is got over
It is more, generally indicate that user is interested in the theme.Therefore, introduce relevant to number weight, can also be convenient for it is subsequent to
The use of family portrait vector.
When merging and/or cluster operation (for example, K-Means cluster), merging and cluster can be pre-established
Restrictive rule.In one embodiment, the video presentation vector that only similarity is greater than predetermined similarity threshold can just be closed
And and/or cluster.For example, in watching history only there are two video when, if the description vectors of the two videos it is similar (that is,
Similarity is greater than predetermined similarity threshold), then it can merge to obtain user's portrait vector.Otherwise, then this can be based respectively on
Two video presentation vectors directly obtain two different users' portrait vectors.And in cluster operation, can give up can not be by
The dispersion video presentation vector of cluster, can also be incorporated into immediate cluster centre.In the case where cluster operation, may be used also
To require the video presentation vector of each cluster to be no less than predetermined number.In another embodiment, it can specify that same user
User draw a portrait vector the upper limit so that for same user generate user draw a portrait vector be not more than scheduled number threshold
Value.Alternatively, or additionally, the number of user's portrait vector needs to meet the predetermined constraints item with the video number watched
Part, for example, the cluster numbers generated are not more than 1/10th of the video presentation vector number for being used to generate cluster.
As described above, the usual more than one of number of user's portrait vector for the same user, is consequently for indicating
The difference preference of user.For this purpose, step S130 may include: to recall one based on each of multiple active users portrait vector
Bar Candidate Recommendation video zipper；And current similar recommendation view of a plurality of Candidate Recommendation video zipper of fusion as the user
Frequently.
It in fusion, needs to consider the weight of each user's portrait vector, determines that each item is waited in similar recommendation video with this
The accounting and/or presentation mode of video zipper are recommended in choosing.In one embodiment, the weight of each active user's portrait vector can
To be determined based on the number for the video presentation vector for generating active user portrait vector.In other embodiments, Ke Yigen
According to other factors, for example, the video presentation vector for generating active user portrait vector corresponds to the viewing time of video to determine
The weight of active user's portrait vector.
In one embodiment, to recommend video in real time recall especially may rely on it is stored in memory above-mentioned
ItemID → vec (video ID to video presentation vector) and clusterID → list [itemID] (cluster ID to affiliated video
ID set) correspond to table.For this purpose, step S130 may include: to calculate cluster belonging to each active user's portrait vector；Based on every
Cluster belonging to a is corresponded to from the Video clustering finds out the included video ID of each cluster in table；Based on the video for being included
ID is corresponded to from the video vector and is found out corresponding video presentation vector in table；It is affiliated poly- with it to calculate each active user's portrait
The similitude of the video presentation vector of each video in class；And the video ID of the return highest predetermined number of similarity is used as and deserves
The Candidate Recommendation video zipper that preceding user's portrait vector is recalled.
Fig. 5 shows an example for recalling calculation process in real time of the invention.
As shown in figure 5, obtaining user first is recently played list, i.e., the nearest short video playing of user is obtained from caching
List, the ID of short-sighted frequency, for example, taking nearest 10, which can modify according to business and computing capability.Then,
According to the ID of this 10 short-sighted frequencies, the expression of vector corresponding to this 10 short-sighted frequencies is obtained from the Map constructed before.
Being drawn a portrait according to the user for being recently played history and calculating in real time active user is indicated with vector.The calculating process of vectorization
It may include merging and clustering.In the case where only having 10 video presentation vectors, merging method can be used, such as: meter
Similar video addition of vectors, is obtained new vector by the cosine similarity between calculating this 10 vectors two-by-two, as obtained 3 after being added
A vector, their corresponding weights are 5,3,2.Weighted value is the vector number being added.As a result, according to from user's play history meter
Calculate the real-time vector portrait for obtaining 3 vectors as active user.
Later, Similar content is carried out using this 3 vectors to recall, obtain three and recall link.Process is recalled with these three
User vector portrait is used as object vector, calculates cosine similarity with the vector of each cluster, obtains the maximum cluster ID of similarity,
It is assumed to be 18.It calculates the cosine similarity value of the short-sighted frequency of target and all short-sighted frequencies under cluster 18 and inverted order is arranged, O before taking.
As a result, by this 3 users draw a portrait vectors do it is similar recall to obtain three recall zipper.
Finally, can according to recommending context and other strategies of recalling to recall zipper to this three and merge, before taking P it is complete
At recalling for recommendation.Here, O and P can be identical or different.
Above-mentioned real-time recall operation can carry out in all cases, such as the case where user opens Video Applications
Under, in the case where user newly clicks the video of predetermined number, and in predetermined time interval under.Compared with the prior art
In the recommendation videos of all any active ues in 90 days be stored in advance recall zipper, it is of the invention recall scheme in real time needed for storage
The result that space is smaller, calculates more light weight, obtains is also more accurate flexible.
The recommendation video scheme of recalling of the invention be primarily based on ANN (especially existing CNN model) to video (for example,
The key frame of short-sighted frequency) processing generation vector is carried out for identifying video, the cosine similarity of these video presentation vectors is bigger,
Two videos are more similar, realize the associated recommendation of video according to the real-time broadcasting behavior of user accordingly, i.e., similar to recall.The present invention
The scheme of recalling do not need a large amount of consumer consumption behavior to generate user's portrait, and without user tag between different terminals
The operation of additional mapping etc..
In other embodiments, the present invention can also be implemented as a kind of recommendation video and recall device.Fig. 6 is shown according to this
The recommendation video of invention one embodiment recalls the composition block diagram of device.As shown, recommending video to recall device 600 can wrap
It includes video presentation vector search unit 610, user's portrait vector generation unit 620 and Candidate Recommendation video and recalls unit 630.
Video presentation vector search unit 610 can be used for the current video based on user and click history, search and each
The corresponding video presentation vector of video；User's portrait vector generation unit 620 can be used for being retouched according to the multiple videos found
Vector is stated, at least one active user portrait vector of the user is generated；Candidate Recommendation video, which recalls unit 630, then can be used for calling together
Return video presentation vector Candidate Recommendation video similar at least one described active user portrait vector.
Same user is being directed to there are when two or more active user portrait vector, the Candidate Recommendation video recalls list
Member 630 can be also used for: recall a Candidate Recommendation video zipper based on each of multiple active users portrait vector；With
And current similar recommendation video of a plurality of Candidate Recommendation video zipper of fusion as the user.Mixing operation may include: base
It draws a portrait the weight of vector in active user, determines the accounting of each Candidate Recommendation video zipper in similar recommendation video and/or be in
Existing mode, wherein the weight of each active user's portrait vector is determined based at least one of following: generating active user portrait
The number of the video presentation vector of vector；And it generates the draw a portrait video presentation vector of vector of the active user and corresponds to the sight of video
See the time.
Preferably, recalling device 600 can also include: video presentation vector generation unit and cluster calculation unit, wherein
Video presentation vector generation unit is used for: being directed to multiple existing videos, is calculated the video presentation vector of each existing video；And
Construction video ID to its correspondence video presentation vector video vector correspond to table, the cluster calculation unit is used for: to it is multiple
There is the video presentation vector of video to be clustered；And the Video clustering of construction video ID to its affiliated cluster ID correspond to table.
Preferably, video presentation vector search unit 610 can be used for: based on user's current video click history in include
Video ID, corresponded to from the video vector and search corresponding video presentation vector in table, and the Candidate Recommendation video is called together
Receipt member 630 can be used for: calculate cluster belonging to each active user's portrait vector；Based on each affiliated cluster from described
Video clustering, which corresponds to, finds out the included video ID of each cluster in table；Based on the video ID for being included from the video vector pair
It answers and finds out corresponding video presentation vector in table；Video of each video is retouched in calculating each active user's portrait and clustering belonging to it
State the similitude of vector；And the video ID of the return highest predetermined number of similarity is called together as active user portrait vector
The Candidate Recommendation video zipper returned.
Preferably, the cluster calculation unit can be also used for: the density being distributed in each cluster based on newly-increased video
Degree, the cluster for carrying out merging or the decomposition of existing cluster update operation；And institute is updated according to the result that cluster updates operation
It states Video clustering and corresponds to table.
Preferably, the video presentation vector generation unit is from artificial neural network (ANN) to the key frame of a certain video
Video presentation vector of the classification results construction key frame vector classified as the video, and user portrait vector
The key frame vector for the video that generation unit watches a certain user merges and/or clusters the active user for generating the user
Portrait vector.
Preferably, the real-time of active user's portrait vector can be carried out under following at least one occasion by recalling device 600
It generates and corresponding candidate recommends recalling in real time for video: in the case where user opens Video Applications；It is newly clicked in user pre-
In the case where the video for determining number；And in predetermined time interval under.
Fig. 7, which is shown, can be used for realizing according to an embodiment of the present invention the calculating equipment that above-mentioned recommendation video recalls method
Structural schematic diagram.
Referring to Fig. 7, calculating equipment 700 includes memory 710 and processor 720.
Processor 720 can be the processor of a multicore, also may include multiple processors.In some embodiments,
Processor 720 may include a general primary processor and one or more special coprocessors, such as graphics process
Device (GPU), digital signal processor (DSP) etc..In some embodiments, the circuit reality of customization can be used in processor 720
It is existing, such as application-specific IC (ASIC, Application Specific Integrated Circuit) or scene
Programmable gate array (FPGA, Field Programmable Gate Arrays).
Memory 710 may include various types of storage units, such as Installed System Memory, read-only memory (ROM), and forever
Long storage device.Wherein, ROM can store the static data of other modules needs of processor 720 or computer or refer to
It enables.Permanent storage can be read-write storage device.Permanent storage can be after computer circuit breaking not
The non-volatile memory device of the instruction and data of storage can be lost.In some embodiments, permanent storage device uses
Mass storage device (such as magnetically or optically disk, flash memory) is used as permanent storage.In other embodiment, permanently deposit
Storage device can be removable storage equipment (such as floppy disk, CD-ROM drive).Installed System Memory can be read-write storage equipment or
The read-write storage equipment of volatibility, such as dynamic random access memory.Installed System Memory can store some or all processors
The instruction and data needed at runtime.In addition, memory 710 may include the combination of any computer readable storage medium,
Including various types of semiconductor memory chips (DRAM, SRAM, SDRAM, flash memory, programmable read only memory), disk and/or
CD can also use.In some embodiments, memory 710 may include that removable storage that is readable and/or writing is set
It is standby, for example, laser disc (CD), read-only digital versatile disc (such as DVD-ROM, DVD-dual layer-ROM), read-only Blu-ray Disc,
Super disc density, flash card (such as SD card, min SD card, Micro-SD card etc.), magnetic floppy disc etc..It is computer-readable to deposit
It stores up medium and does not include carrier wave and the momentary electron signal by wirelessly or non-wirelessly transmitting.
It is stored with executable code on memory 710, when executable code is handled by processor 720, can make to handle
Device 720 executes the recommendation video addressed above and recalls method.
Recommendation video according to the present invention above is described in detail by reference to attached drawing and recalls method and apparatus.The present invention
Video content is indicated using vector by the vectorization of video content, can be calculated in real time according to the real-time play history of user
The vector of user is drawn a portrait, and indicates to carry out similitude recommendation using specific vector.By by recall operation be changed into for
The operation of amount so that entirely framework is recommended more to be simple and efficient, and can solve the problems, such as cold start-up to a certain extent.
In addition, being also implemented as a kind of computer program or computer program product, the meter according to the method for the present invention
Calculation machine program or computer program product include the calculating for executing the above steps limited in the above method of the invention
Machine program code instruction.
Alternatively, the present invention can also be embodied as a kind of (or the computer-readable storage of non-transitory machinable medium
Medium or machine readable storage medium), it is stored thereon with executable code (or computer program or computer instruction code),
When the executable code (or computer program or computer instruction code) by electronic equipment (or calculate equipment, server
Deng) processor execute when, so that the processor is executed each step according to the above method of the present invention.
Those skilled in the art will also understand is that, various illustrative logical blocks, mould in conjunction with described in disclosure herein
Block, circuit and algorithm steps may be implemented as the combination of electronic hardware, computer software or both.
The flow chart and block diagram in the drawings show the possibility of the system and method for multiple embodiments according to the present invention realities
Existing architecture, function and operation.In this regard, each box in flowchart or block diagram can represent module, a journey
A part of sequence section or code, a part of the module, section or code include one or more for realizing defined
The executable instruction of logic function.It should also be noted that in some implementations as replacements, the function of being marked in box can also
To be occurred with being different from the sequence marked in attached drawing.For example, two continuous boxes can actually be basically executed in parallel,
They can also be executed in the opposite order sometimes, and this depends on the function involved.It is also noted that block diagram and/or stream
The combination of each box in journey figure and the box in block diagram and or flow chart, can the functions or operations as defined in executing
Dedicated hardware based system realize, or can realize using a combination of dedicated hardware and computer instructions.
Various embodiments of the present invention are described above, above description is exemplary, and non-exclusive, and
It is not limited to disclosed each embodiment.Without departing from the scope and spirit of illustrated each embodiment, for this skill
Many modifications and changes are obvious for the those of ordinary skill in art field.The selection of term used herein, purport
In the principle, practical application or improvement to the technology in market for best explaining each embodiment, or make the art
Other those of ordinary skill can understand each embodiment disclosed herein.
Claims (20)
1. a kind of recommendation video recalls method, comprising:
Current video based on user clicks history, searches video presentation vector corresponding with each video；
According to the multiple video presentation vectors found, at least one active user portrait vector of the user is generated；And
Recall video presentation vector Candidate Recommendation video similar at least one described active user portrait vector.
2. the method for claim 1, wherein recall video presentation vector and at least one described active user draw a portrait to
Measuring similar Candidate Recommendation video includes:
A Candidate Recommendation video zipper is recalled based on each of multiple active users portrait vector；And
Merge current similar recommendation video of a plurality of Candidate Recommendation video zipper as the user.
3. method according to claim 2, wherein merge current phase of a plurality of Candidate Recommendation video zipper as the user
Include: like recommendation video
Based on the weight of active user's portrait vector, the accounting of each Candidate Recommendation video zipper in similar recommendation video is determined
And/or presentation mode, wherein
The weight of each active user's portrait vector is determined based at least one of following:
Generate the number of the video presentation vector of active user portrait vector；
The video presentation vector for generating active user portrait vector corresponds to the viewing time of video.
4. the method as described in claim 1, further includes:
For multiple existing videos, the video presentation vector of each existing video is calculated；And
Construction video ID corresponds to table to its video vector for corresponding to video presentation vector.
5. method as claimed in claim 4, wherein the current video based on user clicks history, searches and each video pair
The video presentation vector answered includes:
The video ID for including in history is clicked based on user's current video, is corresponded to from the video vector and is searched corresponding view in table
Frequency description vectors.
6. method as claimed in claim 4, further includes:
The video presentation vector of multiple existing videos is clustered；And
The Video clustering of construction video ID to its affiliated cluster ID correspond to table.
7. method as claimed in claim 6, wherein recall video presentation vector and at least one described active user draw a portrait to
Measuring similar Candidate Recommendation video includes:
Calculate cluster belonging to each active user's portrait vector；
It is corresponded to based on each affiliated cluster from the Video clustering and finds out the included video ID of each cluster in table；
It is corresponded to based on the video ID for being included from the video vector and finds out corresponding video presentation vector in table；
The similitude of the video presentation vector of each video in calculating each active user's portrait and being clustered belonging to it；And
Return to the Candidate Recommendation view that the video ID of the highest predetermined number of similarity is recalled as active user portrait vector
Frequency zipper.
8. method as claimed in claim 6, further includes:
Based on the density degree that newly-increased video is distributed in each cluster, the cluster for carrying out merging or the decomposition of existing cluster updates
Operation；And
The Video clustering, which is updated, according to the result that cluster updates operation corresponds to table.
9. the active user for the method for claim 1, wherein generating user in real time under following at least one occasion draws
As vector and recall corresponding Candidate Recommendation video:
In the case where user opens Video Applications；
In the case where user newly clicks the video of predetermined number；And
Under in predetermined time interval.
10. the method for claim 1, wherein the corresponding video presentation vector of each video is from artificial neural network
(ANN) the key frame vector for the classification results construction classified to the key frame of the video, and the current use of user
Family portrait vector is generated from merging and/or the cluster of the key frame vector of the video of user viewing.
11. a kind of recommendation video recalls device, comprising:
Video presentation vector search unit is clicked history for the current video based on user, is searched corresponding with each video
Video presentation vector；
User's portrait vector generation unit, for generating at least the one of the user according to the multiple video presentation vectors found
A active user's portrait vector；And
Candidate Recommendation video recalls unit, for recalling video presentation vector and at least one described active user portrait vector phase
As Candidate Recommendation video.
12. device as claimed in claim 11, wherein the Candidate Recommendation video is recalled unit and is used for:
A Candidate Recommendation video zipper is recalled based on each of multiple active users portrait vector；And
Merge current similar recommendation video of a plurality of Candidate Recommendation video zipper as the user.
13. device as claimed in claim 12, wherein the mixing operation that the Candidate Recommendation video recalls unit includes:
Based on the weight of active user's portrait vector, the accounting of each Candidate Recommendation video zipper in similar recommendation video is determined
And/or presentation mode, wherein
The weight of each active user's portrait vector is determined based at least one of following:
Generate the number of the video presentation vector of active user portrait vector；
The video presentation vector for generating active user portrait vector corresponds to the viewing time of video.
14. device as claimed in claim 11, further includes: video presentation vector generation unit and cluster calculation unit, wherein
Video presentation vector generation unit is used for:
For multiple existing videos, the video presentation vector of each existing video is calculated；And
Construction video ID corresponds to table to its video vector for corresponding to video presentation vector,
The cluster calculation unit is used for:
The video presentation vector of multiple existing videos is clustered；And
The Video clustering of construction video ID to its affiliated cluster ID correspond to table.
15. device as claimed in claim 14, wherein video presentation vector search unit is used for:
The video ID for including in history is clicked based on user's current video, is corresponded to from the video vector and is searched corresponding view in table
Frequency description vectors, and
The Candidate Recommendation video is recalled unit and is used for:
Calculate cluster belonging to each active user's portrait vector；
It is corresponded to based on each affiliated cluster from the Video clustering and finds out the included video ID of each cluster in table；
It is corresponded to based on the video ID for being included from the video vector and finds out corresponding video presentation vector in table；
The similitude of the video presentation vector of each video in calculating each active user's portrait and being clustered belonging to it；And
Return to the Candidate Recommendation view that the video ID of the highest predetermined number of similarity is recalled as active user portrait vector
Frequency zipper.
16. device as claimed in claim 14, wherein the cluster calculation unit is also used to:
Based on the density degree that newly-increased video is distributed in each cluster, the cluster for carrying out merging or the decomposition of existing cluster updates
Operation；And
The Video clustering, which is updated, according to the result that cluster updates operation corresponds to table.
17. device as claimed in claim 11, wherein the video presentation vector generation unit is from artificial neural network
(ANN) the classification results construction key frame vector classified to the key frame of a certain video as the video video presentation to
Amount, and the user draw a portrait vector generation unit the key frame vector for the video that a certain user watches is merged and/or
Cluster generates active user's portrait vector of the user.
18. device as claimed in claim 11, wherein described device carries out active user's picture under following at least one occasion
As real-time generate of vector recommends recalling in real time for video with corresponding candidate:
In the case where user opens Video Applications；
In the case where user newly clicks the video of predetermined number；And
Under in predetermined time interval.
19. a kind of calculating equipment, comprising:
Processor；And
Memory is stored thereon with executable code, when the executable code is executed by the processor, makes the processing
Device executes such as method of any of claims 1-10.
20. a kind of non-transitory machinable medium, is stored thereon with executable code, when the executable code is electric
When the processor of sub- equipment executes, the processor is made to execute such as method of any of claims 1-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910224166.4A CN110008375A (en) | 2019-03-22 | 2019-03-22 | Video is recommended to recall method and apparatus |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910224166.4A CN110008375A (en) | 2019-03-22 | 2019-03-22 | Video is recommended to recall method and apparatus |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110008375A true CN110008375A (en) | 2019-07-12 |
Family
ID=67167876
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910224166.4A Pending CN110008375A (en) | 2019-03-22 | 2019-03-22 | Video is recommended to recall method and apparatus |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110008375A (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390033A (en) * | 2019-07-25 | 2019-10-29 | 腾讯科技（深圳）有限公司 | Training method, device, electronic equipment and the storage medium of image classification model |
| CN110909200A (en) * | 2019-10-09 | 2020-03-24 | 上海麦克风文化传媒有限公司 | Offline and real-time combined audio album recommendation system recall method and system |
| CN110929209A (en) * | 2019-12-06 | 2020-03-27 | 北京百度网讯科技有限公司 | Method and device for sending information |
| CN110990695A (en) * | 2019-11-22 | 2020-04-10 | 厦门美柚股份有限公司 | Recommendation system content recall method and device |
| CN111241388A (en) * | 2019-12-13 | 2020-06-05 | 北京三快在线科技有限公司 | Multi-policy recall method and device, electronic equipment and readable storage medium |
| CN111400546A (en) * | 2020-03-18 | 2020-07-10 | 腾讯科技（深圳）有限公司 | Video recall method and video recommendation method and device |
| CN111552884A (en) * | 2020-05-13 | 2020-08-18 | 腾讯科技（深圳）有限公司 | Method and apparatus for content recommendation |
| CN111918104A (en) * | 2020-07-29 | 2020-11-10 | 有半岛(北京)信息科技有限公司 | Video data recall method and device, computer equipment and storage medium |
| CN111951043A (en) * | 2020-07-27 | 2020-11-17 | 深圳市欢太科技有限公司 | Information delivery processing method and device, storage medium and electronic equipment |
| CN112153423A (en) * | 2020-09-15 | 2020-12-29 | 济南雪景网络技术有限公司 | User-self-driven short video intelligent recommendation method, electronic equipment and storage medium |
| CN112287167A (en) * | 2020-10-29 | 2021-01-29 | 四川长虹电器股份有限公司 | Video recommendation recall method and device |
| CN112528164A (en) * | 2020-12-14 | 2021-03-19 | 建信金融科技有限责任公司 | User collaborative filtering recall method and device |
| CN112804567A (en) * | 2021-01-04 | 2021-05-14 | 青岛聚看云科技有限公司 | Display device, server and video recommendation method |
| CN113573097A (en) * | 2020-04-29 | 2021-10-29 | 北京达佳互联信息技术有限公司 | Video recommendation method and device, server and storage medium |
| CN113704596A (en) * | 2020-05-21 | 2021-11-26 | 北京沃东天骏信息技术有限公司 | Method and apparatus for generating a set of recall information |
| CN114071237A (en) * | 2021-11-23 | 2022-02-18 | 四川长虹电器股份有限公司 | Intelligent television personalized topic recommendation method based on user portrait |
| CN114915844A (en) * | 2021-11-08 | 2022-08-16 | 海看网络科技(山东)股份有限公司 | Method for realizing real-time intelligent recommendation on IPTV |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105930425A (en) * | 2016-04-18 | 2016-09-07 | 乐视控股（北京）有限公司 | Personalized video recommendation method and apparatus |
| US20170300571A1 (en) * | 2016-04-13 | 2017-10-19 | Google Inc. | Video Competition Discovery and Recommendation |
| CN107748759A (en) * | 2017-09-21 | 2018-03-02 | 广州华多网络科技有限公司 | Video pushing method and its equipment |
| CN108228911A (en) * | 2018-02-11 | 2018-06-29 | 北京搜狐新媒体信息技术有限公司 | The computational methods and device of a kind of similar video |
| CN109451334A (en) * | 2018-11-22 | 2019-03-08 | 青岛聚看云科技有限公司 | User, which draws a portrait, generates processing method, device and electronic equipment |
-
2019
- 2019-03-22 CN CN201910224166.4A patent/CN110008375A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170300571A1 (en) * | 2016-04-13 | 2017-10-19 | Google Inc. | Video Competition Discovery and Recommendation |
| CN105930425A (en) * | 2016-04-18 | 2016-09-07 | 乐视控股（北京）有限公司 | Personalized video recommendation method and apparatus |
| CN107748759A (en) * | 2017-09-21 | 2018-03-02 | 广州华多网络科技有限公司 | Video pushing method and its equipment |
| CN108228911A (en) * | 2018-02-11 | 2018-06-29 | 北京搜狐新媒体信息技术有限公司 | The computational methods and device of a kind of similar video |
| CN109451334A (en) * | 2018-11-22 | 2019-03-08 | 青岛聚看云科技有限公司 | User, which draws a portrait, generates processing method, device and electronic equipment |
Non-Patent Citations (1)
| Title |
|---|
| 匡文波等: "论视频智能推荐的算法模型设计", 《新闻传播》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390033B (en) * | 2019-07-25 | 2023-04-21 | 腾讯科技（深圳）有限公司 | Training method and device for image classification model, electronic equipment and storage medium |
| CN110390033A (en) * | 2019-07-25 | 2019-10-29 | 腾讯科技（深圳）有限公司 | Training method, device, electronic equipment and the storage medium of image classification model |
| CN110909200A (en) * | 2019-10-09 | 2020-03-24 | 上海麦克风文化传媒有限公司 | Offline and real-time combined audio album recommendation system recall method and system |
| CN110990695A (en) * | 2019-11-22 | 2020-04-10 | 厦门美柚股份有限公司 | Recommendation system content recall method and device |
| CN110929209A (en) * | 2019-12-06 | 2020-03-27 | 北京百度网讯科技有限公司 | Method and device for sending information |
| CN110929209B (en) * | 2019-12-06 | 2023-06-20 | 北京百度网讯科技有限公司 | Method and device for transmitting information |
| CN111241388A (en) * | 2019-12-13 | 2020-06-05 | 北京三快在线科技有限公司 | Multi-policy recall method and device, electronic equipment and readable storage medium |
| CN111400546A (en) * | 2020-03-18 | 2020-07-10 | 腾讯科技（深圳）有限公司 | Video recall method and video recommendation method and device |
| CN111400546B (en) * | 2020-03-18 | 2020-12-01 | 腾讯科技（深圳）有限公司 | Video recall method and video recommendation method and device |
| CN113573097A (en) * | 2020-04-29 | 2021-10-29 | 北京达佳互联信息技术有限公司 | Video recommendation method and device, server and storage medium |
| CN111552884A (en) * | 2020-05-13 | 2020-08-18 | 腾讯科技（深圳）有限公司 | Method and apparatus for content recommendation |
| CN111552884B (en) * | 2020-05-13 | 2024-05-14 | 腾讯科技（深圳）有限公司 | Method and apparatus for content recommendation |
| CN113704596A (en) * | 2020-05-21 | 2021-11-26 | 北京沃东天骏信息技术有限公司 | Method and apparatus for generating a set of recall information |
| CN111951043A (en) * | 2020-07-27 | 2020-11-17 | 深圳市欢太科技有限公司 | Information delivery processing method and device, storage medium and electronic equipment |
| CN111918104A (en) * | 2020-07-29 | 2020-11-10 | 有半岛(北京)信息科技有限公司 | Video data recall method and device, computer equipment and storage medium |
| CN112153423A (en) * | 2020-09-15 | 2020-12-29 | 济南雪景网络技术有限公司 | User-self-driven short video intelligent recommendation method, electronic equipment and storage medium |
| CN112287167A (en) * | 2020-10-29 | 2021-01-29 | 四川长虹电器股份有限公司 | Video recommendation recall method and device |
| CN112528164A (en) * | 2020-12-14 | 2021-03-19 | 建信金融科技有限责任公司 | User collaborative filtering recall method and device |
| CN112528164B (en) * | 2020-12-14 | 2023-02-03 | 中国建设银行股份有限公司 | User collaborative filtering recall method and device |
| CN112804567A (en) * | 2021-01-04 | 2021-05-14 | 青岛聚看云科技有限公司 | Display device, server and video recommendation method |
| CN114915844A (en) * | 2021-11-08 | 2022-08-16 | 海看网络科技(山东)股份有限公司 | Method for realizing real-time intelligent recommendation on IPTV |
| CN114071237A (en) * | 2021-11-23 | 2022-02-18 | 四川长虹电器股份有限公司 | Intelligent television personalized topic recommendation method based on user portrait |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110008375A (en) | Video is recommended to recall method and apparatus | |
| Elezi et al. | The group loss for deep metric learning | |
| CN107220365B (en) | Accurate recommendation system and method based on collaborative filtering and association rule parallel processing | |
| CN112785397A (en) | Product recommendation method, device and storage medium | |
| CN110008376A (en) | User's portrait vector generation method and device | |
| CN110019950A (en) | Video recommendation method and device | |
| CN109511015B (en) | Multimedia resource recommendation method, device, storage medium and equipment | |
| CN112364204B (en) | Video searching method, device, computer equipment and storage medium | |
| CN112749330B (en) | Information pushing method, device, computer equipment and storage medium | |
| CN111241394B (en) | Data processing method, data processing device, computer readable storage medium and electronic equipment | |
| CN112380453B (en) | Article recommendation method and device, storage medium and equipment | |
| CN110351318A (en) | Using the method, terminal and computer storage medium of recommendation | |
| CN116975615A (en) | Task prediction method and device based on video multi-mode information | |
| CN113570416B (en) | Method and device for determining delivered content, electronic equipment and storage medium | |
| Khoshaba et al. | Machine learning algorithms in Bigdata analysis and its applications: A Review | |
| CN116823410B (en) | Data processing method, object processing method, recommending method and computing device | |
| CN108804491A (en) | item recommendation method, device, computing device and storage medium | |
| CN110569447B (en) | Network resource recommendation method and device and storage medium | |
| CN114817692A (en) | Method, device and equipment for determining recommended object and computer storage medium | |
| CN110851708B (en) | Negative sample extraction method, device, computer equipment and storage medium | |
| US20230214676A1 (en) | Prediction model training method, information prediction method and corresponding device | |
| CN116910357A (en) | Data processing method and related device | |
| CN115525831A (en) | Recommendation model training method, recommendation device and computer readable storage medium | |
| CN114330519A (en) | Data determination method and device, electronic equipment and storage medium | |
| CN114996435A (en) | Information recommendation method, device, equipment and storage medium based on artificial intelligence |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190712 |