CN109558514A - Video recommendation method, its device, information processing equipment and storage medium - Google Patents
Video recommendation method, its device, information processing equipment and storage medium Download PDFInfo
- Publication number
- CN109558514A CN109558514A CN201910014819.6A CN201910014819A CN109558514A CN 109558514 A CN109558514 A CN 109558514A CN 201910014819 A CN201910014819 A CN 201910014819A CN 109558514 A CN109558514 A CN 109558514A
- Authority
- CN
- China
- Prior art keywords
- video
- user
- vector
- article
- sequence
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention discloses a kind of video recommendation method, its device, information processing equipment and storage mediums, and sample of multiple video sequences with forward-backward correlation as training neural network model is determined in all user's history behavioral datas.After being trained using such video sequence sample to neural network model, the article vector for each video can be determined, so that each video be indicated with vector.In view of the real-time interest of user and time correlation, the real-time interest of user can be embodied by closely determining the corresponding user vector of user to remote watched video and the corresponding article vector of each video according to the time according to user.Thus, the similar sequence of video interested to each user can be determined according to the similarity degree of the corresponding article vector user vector corresponding to the user of each video, with this tendency and real-time interest for recommending video more to meet user to user that sort, the accuracy for being conducive to improve video recommendations, promotes user experience.
Description
Technical field
The present invention relates to video technique field more particularly to a kind of video recommendation method, its device, information processing equipment and
Storage medium.
Background technique
With the continuous development of Internet technology, network video becomes increasingly abundant, and user watches video and is no longer limited to TV,
Can also be by the interested video-see of internet hunt, the broadcasting time limit of no longer limited TV.In addition to this, internet regards
Frequency can also facilitate user to select to user recommended user.
Currently, the mode for generalling use collaborative filtering recommends video to user, this method watches video dependent on user
Historical behavior, need based on a large amount of historical behavior data carry out video recommendations, data volume to be treated is huge, and efficiency is not
It is high.And what interest when user watches video may vary over, but existing video recommendation method is not examined
Consider the real-time interest of user, the accuracy of recommendation is to be improved.
Summary of the invention
The present invention provides a kind of video recommendation method, its device, information processing equipment and storage medium, to optimize video
The accuracy of recommendation.
In a first aspect, the present invention provides a kind of video recommendation method, comprising:
According to the historical behavior data of users all in database determine multiple video sequence samples and each user by
According to by closely to the video sequence of remote the watched video of time sequencing；Each video and the video phase in the video sequence sample
There are correlations between adjacent video；
The neural network model of setting is trained using the video sequence sample, determines that each video is corresponding
Article vector；
According to video each in the video sequence of each user put in order and the corresponding article vector of each video, really
Determine the corresponding user vector of each user；
It successively determines the similarity between the corresponding article vector of each video user vector corresponding with each user, presses
Recommend similar video to the user according to the similarity sequence from high to low of user vector corresponding to the user.
In a kind of achievable embodiment, in the above method provided by the invention, the neural network model is
Item2vec model；
It is described that the neural network model of setting is trained using the video sequence sample, determine each video pair
The article vector answered, comprising:
The item2vec model is trained using the video sequence sample；
The latent space parameter for the item2vec model that training is completed is expressed as article corresponding with each video
Vector；
Wherein, the quantity for the component that the article vector is included is less than the video that the video sequence sample is included
Quantity.
In a kind of achievable embodiment, in the above method provided by the invention, the user vector use with
Lower formula determines:
Wherein,Indicate the user vector,Indicate the article vector of j-th of video, wjIndicate j-th of video
Time decaying weight.
In a kind of achievable embodiment, in the above method provided by the invention, the time decaying weight is adopted
It is determined with following formula:
Wherein, wjIndicate that time decaying weight, α are to indicate time attenuation coefficient, order indicates video in the video sequence
Sequence in column.
In a kind of achievable embodiment, in the above method provided by the invention, the article vector with it is described
The similarity of user vector is determined using following formula:
Wherein, sim (ij,uj) indicate the similarity of the article vector and the user vector,Indicate the article
Vector,Indicate the user vector.
Second aspect, the present invention provide a kind of video recommendations device, comprising:
Video sequence determination unit, for determining multiple video sequences according to the historical behavior data of users all in database
Column sample and each user are according to by closely to the video sequence of remote the watched video of time sequencing；The video sequence sample
In each video video adjacent with the video between there are correlations；
Article vector determination unit, for being instructed using the video sequence sample to the neural network model of setting
Practice, determines the corresponding article vector of each video；
User vector determination unit, in the video sequence according to each user each video put in order and it is each described
The corresponding article vector of video, determines the corresponding user vector of each user；
Recommendation unit, for successively determine the corresponding article vector of each video user vector corresponding with each user it
Between similarity, recommend similar view to the user according to the similarity sequence from high to low of user vector corresponding to the user
Frequently.
In a kind of achievable embodiment, in above-mentioned apparatus provided by the invention, the article vector determines single
Member, specifically for being trained using the video sequence sample to the item2vec model；It will be described in training completion
The latent space parameter of item2vec model is expressed as article vector corresponding with each video；
Wherein, the quantity for the component that the article vector is included is less than the video that the video sequence sample is included
Quantity.
In a kind of achievable embodiment, in above-mentioned apparatus provided by the invention, the user vector determines single
Member determines the user vector using following formula:
Indicate the user vector,Indicate the article vector of j-th of video, wjIndicate j-th of video when
Between decaying weight；
Wherein, the time decaying weight is determined using following formula:
α is to indicate time attenuation coefficient, and order indicates sequence of the video in the video sequence.
In a kind of achievable embodiment, in above-mentioned apparatus provided by the invention, the recommendation unit use with
Lower formula determines the similarity of the article vector and the user vector:
Wherein, sim (ij,uj) indicate the similarity of the article vector and the user vector,Indicate the article
Vector,Indicate the user vector.
The third aspect, the present invention provide a kind of information processing equipment, comprising:
Memory, for storing program instruction；
Processor is executed for calling the described program stored in the memory to instruct according to the program of acquisition: according to
The historical behavior data of all users determine multiple video sequence samples and each user according to by closely to remote in database
The video sequence of the watched video of time sequencing；The neural network model of setting is instructed using the video sequence sample
Practice, determines the corresponding article vector of each video；According to the video sequence of each user and the article vector of each video,
Determine the corresponding user vector of each user；Successively determine the corresponding article vector of each video user corresponding with each user to
Similarity between amount is recommended according to the similarity sequence from high to low of user vector corresponding to the user to the user similar
Video；
Wherein, there are correlations between each video in video sequence sample video adjacent with the video.
Fourth aspect, the present invention provides a kind of computer-readable non-volatile memory medium, described computer-readable
Non-volatile memory medium is stored with computer executable instructions, and the computer executable instructions are above-mentioned for executing calculating
Any video recommendation method.
Video recommendation method, its device, information processing equipment and storage medium provided by the invention, according to institute in database
There are the historical behavior data of user to determine multiple video sequence samples and each user according to by closely to remote time sequencing institute
Watch the video sequence of video；It is trained using neural network model of the video sequence sample to setting, determines each video pair
The article vector answered；According to video each in the video sequence of each user put in order and the corresponding article vector of each video,
Determine the corresponding user vector of each user；Successively determine the corresponding article vector of each video user vector corresponding with each user it
Between similarity, recommend similar view to the user according to the similarity sequence from high to low of user vector corresponding to the user
Frequently.Determine multiple video sequences with forward-backward correlation as training neural network in all user's history behavioral datas
The sample of model.After being trained using such video sequence sample to neural network model, it can determine for every
The article vector of a video, so that each video be indicated with vector.User vector is related with article vector, embodies each use
Family tends to which video watched, it is contemplated that the real-time interest of user and time correlation, the current interest of user and recently viewing
The correlation of the correlation of video watched video when may be far longer than earlier, thus according to user according to the time by closely to remote
Watched video and the corresponding article vector of each video determine the corresponding user vector of user, can not only embody user
Which tend to that video watched, additionally it is possible to embody the real-time interest of user.As a result, according to the corresponding article vector of each video with
The similarity degree of the corresponding user vector of user can determine the similar sequence of video interested to each user, be arranged with this
Sequence recommends video more to meet the tendency and real-time interest of user to user, is conducive to the accuracy for improving video recommendations, is promoted and used
Family experience.
Detailed description of the invention
Fig. 1 is the flow chart of video recommendation method provided in an embodiment of the present invention；
Fig. 2 is the structural schematic diagram of video recommendations device provided in an embodiment of the present invention；
Fig. 3 is the structural schematic diagram of information processing equipment provided in an embodiment of the present invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing to the present invention make into
It is described in detail to one step, it is clear that described embodiments are only a part of the embodiments of the present invention, rather than whole implementation
Example.Based on the embodiments of the present invention, obtained by those of ordinary skill in the art without making creative efforts
All other embodiment, shall fall within the protection scope of the present invention.
It to video recommendation method provided in an embodiment of the present invention, its device, information processing equipment and deposits with reference to the accompanying drawing
Storage media is described in detail.
The embodiment of the present invention in a first aspect, a kind of video recommendation method is provided, as shown in Figure 1, the embodiment of the present invention mentions
The video recommendation method of confession, comprising:
S101, multiple video sequence samples and each use are determined according to the historical behavior data of users all in database
Family is according to by closely to the video sequence of remote the watched video of time sequencing；
S102, it is trained using neural network model of the video sequence sample to setting, determines the corresponding object of each video
Product vector；
In S103, the video sequence according to each user each video put in order and the corresponding article vector of each video,
Determine the corresponding user vector of each user；
S104, similarity between the corresponding article vector of each video user vector corresponding with each user is successively determined,
Recommend similar video to the user according to the similarity sequence from high to low of user vector corresponding to the user.
Wherein, there are correlations between each video in video sequence sample video adjacent with the video.User's viewing
The historical behavior data of video reflect the correlation between video.This is because user is when watching video, at current time
The real-time interest generated can be related with the video watched at this time, then user when having watched current video, is more likely to
Search or performer similar to current video type is similar, or directs similar etc. video-see.The embodiment of the present invention is based on using
The These characteristics that video generates are watched at family, and multiple views with forward-backward correlation are determined in all user's history behavioral datas
Sample of the frequency sequence as training neural network model.Neural network model is trained using such video sequence sample
Later, the article vector for each video can be determined, so that each video be indicated with vector.User vector and article
Vector is related, embodies each user tends to which video watched, in embodiments of the present invention, it is contemplated that user's is real-time emerging
Interest and time correlation, the current interest of user and the correlation of nearest viewing video watched video when may be far longer than earlier
Correlation, therefore according to user according to the time by closely respectively being regarded into the video sequence and the video sequence of remote watched video
Frequently corresponding article vector determines the corresponding user vector of user, can not only embody user tends to which view watched
Frequently, additionally it is possible to embody the real-time interest of user.As a result, according to the corresponding article vector user corresponding to the user of each video to
The similarity degree of amount can determine the similar sequence of video interested to each user, sorted with this and recommend video to user
The tendency and real-time interest for more meeting user are conducive to the accuracy for improving video recommendations, promote user experience.
Specifically, item2vec model can be used in neural network model in embodiments of the present invention.It adopts in the prior art
Video is recommended to there are problems that two with collaborative filtering model, the quantity of user and the quantity of video first is very huge, then
It is not high to carry out matrix decomposition computational valid time rate；Secondly, what the video of user's viewing was distributed there is long hair, therefore association can not be used
It is indicated well with filtering model.And the article vector that item2vec model used in the embodiment of the present invention is determined then more can
Expression is in the article of long-tail, and this model computational efficiency under coordinates data amount is better than the prior art.
It is item2vec model in neural network model, in above-mentioned step S102, using video sequence sample to setting
Fixed neural network model is trained, and is determined the corresponding article vector of each video, be can specifically include:
Item2vec model is trained using video sequence sample；
The latent space parameter for the item2vec model that training is completed is expressed as article vector corresponding with each video；
Wherein, the quantity for the component that article vector is included is less than the quantity for the video that video sequence sample is included.
Multiple video sequence samples are determined in the historical behavior data of all users, both include positive sample in these samples
It also include negative sample.Positive negative sample is encoded to be indicated with vector, the equal length of the vector of positive negative sample, and in positive sample
Video and adjacent video between there are correlation, correlation is not present between the video and adjacent video in negative sample.Meet
The video sequence of the historical behavior of user produces positive sample, and the video sequence for not meeting the historical behavior of user can give birth to
At negative sample.Under normal conditions, the mode generated at random can be used in negative sample, and the quantity for the negative sample that can be generated is significantly larger than
The quantity of positive sample.Come after determining positive negative sample, item2vec model is trained using these positive negative samples.
Item2vec model includes CBOW model and Skip-gram model, can be carried out in practical applications using any one model
Training, it is not limited here.For example, can be used for positive negative sample is trained Skip-gram model, can be adopted using negative
The mode of sample is trained Skip-gram model, and speed, improved model quality can be improved in the mode of negative sampling.It sets hidden
The specific value of the quantity of spatial parameter and the length of prediction window, the quantity of latent space parameter can be rule of thumb and real
Border situation is set, and is not specifically limited herein.But the quantity of latent space parameter is far smaller than video sequence sample under normal conditions
This length, therefore the latent space parameter finally obtained is expressed as a vector the corresponding article vector of each video, it can play
Effect to the vector dimensionality reduction of characterization video, can substantially reduce operand in this way, improve efficiency.
After the quantity of latent space parameter has been determined, and has taken the window that prediction length is 2r+1, by above-mentioned positive negative sample
It sequentially inputs and model is trained into model, so that the model that final training is completed is in some view inputted in positive sample
When the vector of frequency, preceding r video and rear r video of the video of input in positive sample can be exported.Thus, it is possible to be instructed
The latent space parameter of the model after the completion of white silk.By the latent space parameter with vector table be shown as each video corresponding to article to
Amount.Wherein, the component of the corresponding article vector of each video is divided into continuous partial parameters in latent space parameter, and each video
Continuous parameter corresponding to corresponding article vector can be different.For example, the corresponding article vector of certain video may be expressed as:
Wherein,Indicate the corresponding article vector of video r, w1,w2,w3......wkIt is retrieved as in above-mentioned latent space parameter
Continuous k floating-point values.So, each video in database can use length to indicate for the vector of k.Such as
The total quantity of fruit video is n, then when parameter is arranged to Skip-gram model, quantity should take nk, in practical applications, can be with
Set particular number according to actual needs, herein without limitation.
Further, after having obtained the corresponding article vector of each video, it is also necessary to which each user is watched video
Historical behavior is indicated with vector, the real-time interest of user is considered in embodiments of the present invention, in the historical behavior of all users
Get the history viewing record of each user in data first, and can by video that user is watched according to the time by closely to
Remote sequentially forms video sequence corresponding with the user, thus constitutes the historical behavior table for being directed to each user, can be with
It indicates are as follows:
Wherein, uiIndicating user i, the length of S is the quantity m for indicating user,Indicate m-th of user's
Video sequence, the video sequence have order, come the more video of front and the interaction time of user is closer from now.
In the specific implementation, according in above-mentioned video sequence video put in order and each video corresponding to article to
Amount, can be used following formula to determine user vector:
Wherein,Indicate user vector,Indicate the article vector of j-th of video, wjIndicate j-th of video when
Between decaying weight.
Since user can deposit real-time interest in a short time, i.e., the video of interaction closer apart from present time more can shadow
Ring the next interbehavior of user.Therefore in order to reasonably embody this feature in user vector, the embodiment of the present invention is adopted
User vector is characterized with the linear weighted function article vector that the time decays.
It in another embodiment, can also be by the corresponding article vector of video each in the corresponding video sequence of user
Average vector is as user vector.
For example, being directed to the video sequence of m-th of userThe corresponding user vector of the user can be with
It is indicated using following relationship:
Wherein,Indicate user vector,Indicate the article vector of j-th of video.
In practical applications, either above-mentioned formula can be used to determine the corresponding user vector of user, declined using the time
The article vector for subtracting linear weighted function can fully demonstrate the real-time interest of user, more succinct by the way of average vector.It removes
Except this, user vector can also be characterized using other can characterize by the way of the real-time interest of user, it is not limited here.
Further, above-mentioned time decaying weight can be determined using following formula:
Wherein, wjIndicate that time decaying weight, α are to indicate time attenuation coefficient, order indicates video in the video sequence
Sequence.As can be seen that the more forward video that sorts in video sequence, corresponding time decaying weight wjValue it is bigger, that
Influence of the corresponding article vector of the video to user vector is bigger.The reality of user can be sufficiently characterized in this manner
When interest, be conducive to improve video recommendations accuracy.
After obtaining the corresponding article vector of each video and the corresponding user vector of each user, for each user
Successively corresponding with each video article vector carries out the calculating of similarity to corresponding user vector, it is hereby achieved that with each use
Family will more meet the phase of user according to the sequence of similarity from high to low to the interest level of each video to user's recommendation video
It hopes, promotes user experience.
In the specific implementation, it is similar to user vector to calculate article vector that the calculation method of cosine similarity can be used
Degree, may refer to following formula:
Wherein, sim (ij,uj) indicate article vector and user vector similarity,Indicate article vector,Table
Show user vector.Molecule is the product of article vector sum user vector, and denominator is the long product of two vector field homoemorphisms.
It is possible thereby to calculate the matching result of each video in each user and database, user vector and article to
The similarity of amount is higher, then says it is viewing tendency that the video is more in line with user, and is more in line with current real-time emerging of user
User experience can be improved then recommending video to user with the sequence of such similarity from high to low in interest.In addition to this,
It can also the number appropriate for adjusting paid video in recommendation list of the ratio according to shared by paid video in the video of user's interaction
Thus amount improves the payment conversion ratio of product.When user is a new user, there is no about the new user's in the database
Historical behavior data.The attribute information of already present user in the attribute information and database of new user can then be compared
Compared with, for example, the region of new user and old user can be compared, the attributes such as gender, by the old user's high with new user's matching degree
Recommendation results recommend new user, and paid video accounting example may be configured as 50% at this time.
The second aspect of the embodiment of the present invention provides a kind of video recommendations device, as shown in Fig. 2, the embodiment of the present invention mentions
The video recommendations device of confession, comprising:
Video sequence determination unit 21, for determining multiple videos according to the historical behavior data of users all in database
Sequence samples and each user are according to by closely to the video sequence of remote the watched video of time sequencing；In video sequence sample
Each video video adjacent with the video between there are correlations；
Article vector determination unit 22, for being trained using video sequence sample to the neural network model of setting,
Determine the corresponding article vector of each video；
User vector determination unit 23 putting in order and respectively regarding for each video in the video sequence according to each user
Frequently corresponding article vector determines the corresponding user vector of each user；
Recommendation unit 34, for successively determining between the corresponding article vector of each video user vector corresponding with each user
Similarity, recommend similar video to the user according to the similarity sequence from high to low of user vector corresponding to the user.
Optionally, article vector determination unit 22 is specifically used for carrying out item2vec model using video sequence sample
Training；The latent space parameter for the item2vec model that training is completed is expressed as article vector corresponding with each video；
Wherein, the quantity for the component that article vector is included is less than the quantity for the video that video sequence sample is included.
Reliably, user vector determination unit 23 determines user vector using following formula:
Indicate user vector,Indicate the article vector of j-th of video, wjIndicate the time decaying of j-th of video
Weight；
Wherein, time decaying weight is determined using following formula:
α is to indicate time attenuation coefficient, and order indicates the sequence of video in the video sequence.
Optionally, recommendation unit 24 determines the similarity of article vector and user vector using following formula:
Wherein, sim (ij,uj) indicate article vector and user vector similarity,Indicate article vector,Table
Show user vector.
Above-mentioned video recommendations device provided in an embodiment of the present invention determination in all user's history behavioral datas is multiple
Sample of the video sequence with forward-backward correlation as training neural network model.Using such video sequence sample to mind
After being trained through network model, the article vector for each video can be determined, thus by each video vector
It indicates.User vector is related with article vector, embodies each user tends to which video watched, it is contemplated that user's is real-time
Interest and time correlation, the current interest of user and the correlation of nearest viewing video watch view when may be far longer than earlier
The correlation of frequency, thus according to user according to the time by closely to remote watched video and the corresponding article vector of each video come really
Determine the corresponding user vector of user, can not only embody user tends to which video watched, additionally it is possible to embody user's
Real-time interest.It can be determined according to the similarity degree of the corresponding article vector user vector corresponding to the user of each video as a result,
The similar sequence of video interested to each user out, with this tendency and reality for recommending video more to meet user to user that sort
When interest, be conducive to improve video recommendations accuracy, promoted user experience.
The third aspect of the embodiment of the present invention provides a kind of information processing equipment, as shown in figure 3, the embodiment of the present invention mentions
The information processing equipment of confession includes:
Memory, for storing program instruction；
Processor executes: according in database for calling the program instruction stored in memory according to the program of acquisition
The historical behavior data of all users determine multiple video sequence samples and each user according to by closely to remote time sequencing
The video sequence of watched video；It is trained using neural network model of the video sequence sample to setting, determines each video
Corresponding article vector；According to the video sequence of each user and the article vector of each video, the corresponding user of each user is determined
Vector；Successively determine the similarity between the corresponding article vector of each video user vector corresponding with each user, according to with
The sequence of the similarity of the corresponding user vector in family from high to low recommends similar video to the user；
Wherein, there are correlations between each video in video sequence sample video adjacent with the video.
The fourth aspect of the embodiment of the present invention provides a kind of computer-readable non-volatile memory medium, the computer
Readable non-volatile memory medium is stored with computer executable instructions, which executes for making to calculate
Any of the above-described video recommendation method.
Video recommendation method, its device, information processing equipment and storage medium provided by the invention, according to institute in database
There are the historical behavior data of user to determine multiple video sequence samples and each user according to by closely to remote time sequencing institute
Watch the video sequence of video；It is trained using neural network model of the video sequence sample to setting, determines each video pair
The article vector answered；According to video each in the video sequence of each user put in order and the corresponding article vector of each video,
Determine the corresponding user vector of each user；Successively determine the corresponding article vector of each video user vector corresponding with each user it
Between similarity, recommend similar view to the user according to the similarity sequence from high to low of user vector corresponding to the user
Frequently.Determine multiple video sequences with forward-backward correlation as training neural network in all user's history behavioral datas
The sample of model.After being trained using such video sequence sample to neural network model, it can determine for every
The article vector of a video, so that each video be indicated with vector.User vector is related with article vector, embodies each use
Family tends to which video watched, it is contemplated that the real-time interest of user and time correlation, the current interest of user and recently viewing
The correlation of the correlation of video watched video when may be far longer than earlier, thus according to user according to the time by closely to remote
Watched video and the corresponding article vector of each video determine the corresponding user vector of user, can not only embody user
Which tend to that video watched, additionally it is possible to embody the real-time interest of user.As a result, according to the corresponding article vector of each video with
The similarity degree of the corresponding user vector of user can determine the similar sequence of video interested to each user, be arranged with this
Sequence recommends video more to meet the tendency and real-time interest of user to user, is conducive to the accuracy for improving video recommendations, is promoted and used
Family experience.
The present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
The processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices is instructed, so that
A stream in flow chart can be achieved by the instruction that the computer or the processor of other programmable data processing devices execute
The function of being specified in journey or multiple processes and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one process or multiple processes and/or block diagrams of flow chart
One box or multiple boxes in specify function the step of.
Although preferred embodiments of the present invention have been described, it is created once a person skilled in the art knows basic
Property concept, then additional changes and modifications can be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as
It selects embodiment and falls into all change and modification of the scope of the invention.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (11)
1. a kind of video recommendation method characterized by comprising
According to the historical behavior data of users all in database determine multiple video sequence samples and each user according to by
Closely to the video sequence of remote the watched video of time sequencing；Each video in the video sequence sample is adjacent with the video
There are correlations between video；
The neural network model of setting is trained using the video sequence sample, determines the corresponding article of each video
Vector；
According to video each in the video sequence of each user put in order and the corresponding article vector of each video, determine each
The corresponding user vector of user；
Successively determine the similarity between the corresponding article vector of each video user vector corresponding with each user, according to
The sequence of the similarity of the corresponding user vector of user from high to low recommends similar video to the user.
2. the method as described in claim 1, which is characterized in that the neural network model is item2vec model；
It is described that the neural network model of setting is trained using the video sequence sample, determine that each video is corresponding
Article vector, comprising:
The item2vec model is trained using the video sequence sample；
The latent space parameter for the item2vec model that training is completed is expressed as article vector corresponding with each video；
Wherein, the quantity for the component that the article vector is included is less than the number for the video that the video sequence sample is included
Amount.
3. the method as described in claim 1, which is characterized in that the user vector is determined using following formula:
Wherein,Indicate the user vector,Indicate the article vector of j-th of video, wjIndicate j-th of video when
Between decaying weight.
4. method as claimed in claim 3, which is characterized in that the time decaying weight is determined using following formula:
Wherein, wjIndicate that time decaying weight, α are to indicate time attenuation coefficient, order indicates video in the video sequence
Sequence.
5. the method as described in claim 1, which is characterized in that the similarity of the article vector and the user vector uses
Following formula determines:
Wherein, sim (ij,uj) indicate the similarity of the article vector and the user vector,Indicate the article to
Amount,Indicate the user vector.
6. a kind of video recommendations device characterized by comprising
Video sequence determination unit, for determining multiple video sequence samples according to the historical behavior data of users all in database
This and each user are according to by closely to the video sequence of remote the watched video of time sequencing；In the video sequence sample
There are correlations between each video video adjacent with the video；
Article vector determination unit, for being trained using the video sequence sample to the neural network model of setting, really
Determine the corresponding article vector of each video；
User vector determination unit, in the video sequence according to each user each video put in order and each video
Corresponding article vector determines the corresponding user vector of each user；
Recommendation unit, for successively determining between the corresponding article vector of each video user vector corresponding with each user
Similarity recommends similar video to the user according to the similarity sequence from high to low of user vector corresponding to the user.
7. device as claimed in claim 6, which is characterized in that the article vector determination unit is specifically used for described in use
Video sequence sample is trained the item2vec model；The latent space ginseng for the item2vec model that training is completed
Number is expressed as article vector corresponding with each video；
Wherein, the quantity for the component that the article vector is included is less than the number for the video that the video sequence sample is included
Amount.
8. device as claimed in claim 6, which is characterized in that the user vector determination unit determines institute using following formula
State user vector:
Indicate the user vector,Indicate the article vector of j-th of video, wjIndicate the time decaying of j-th of video
Weight；
Wherein, the time decaying weight is determined using following formula:
α is to indicate time attenuation coefficient, and order indicates sequence of the video in the video sequence.
9. device as claimed in claim 6, which is characterized in that the recommendation unit using following formula determine the article to
The similarity of amount and the user vector:
Wherein, sim (ij,uj) indicate the similarity of the article vector and the user vector,Indicate the article to
Amount,Indicate the user vector.
10. a kind of information processing equipment characterized by comprising
Memory, for storing program instruction；
Processor executes: according to data for calling the described program stored in the memory to instruct according to the program of acquisition
The historical behavior data of all users determine multiple video sequence samples and each user according to by closely to the remote time in library
The video sequence of the watched video of sequence；The neural network model of setting is trained using the video sequence sample, really
Determine the corresponding article vector of each video；According to the video sequence of each user and the article vector of each video, determine
The corresponding user vector of each user；Successively determine the corresponding article vector of each video user vector corresponding with each user it
Between similarity, recommend similar view to the user according to the similarity sequence from high to low of user vector corresponding to the user
Frequently；
Wherein, there are correlations between each video in video sequence sample video adjacent with the video.
11. a kind of computer-readable non-volatile memory medium, which is characterized in that described computer-readable non-volatile to deposit
Storage media is stored with computer executable instructions, and the computer executable instructions calculate perform claim requirement 1-5 for making
Video recommendation method described in one.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910014819.6A CN109558514B (en) | 2019-01-08 | 2019-01-08 | Video recommendation method, device thereof, information processing equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910014819.6A CN109558514B (en) | 2019-01-08 | 2019-01-08 | Video recommendation method, device thereof, information processing equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109558514A true CN109558514A (en) | 2019-04-02 |
| CN109558514B CN109558514B (en) | 2023-04-11 |
Family
ID=65872545
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910014819.6A Active CN109558514B (en) | 2019-01-08 | 2019-01-08 | Video recommendation method, device thereof, information processing equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109558514B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110971973A (en) * | 2019-12-03 | 2020-04-07 | 北京奇艺世纪科技有限公司 | Video pushing method and device and electronic equipment |
| CN111125428A (en) * | 2019-12-17 | 2020-05-08 | 东北大学 | Time-dependent movie recommendation method based on score prediction function fitting structure |
| CN111782925A (en) * | 2019-04-04 | 2020-10-16 | 阿里巴巴集团控股有限公司 | Item recommendation method, device, equipment, system and readable storage medium |
| CN111813992A (en) * | 2020-07-14 | 2020-10-23 | 四川长虹电器股份有限公司 | Sorting system and method for movie recommendation candidate set |
| CN112000872A (en) * | 2019-05-27 | 2020-11-27 | 北京地平线机器人技术研发有限公司 | Recommendation method based on user vector, training method of model and training device |
| CN112528071A (en) * | 2020-10-30 | 2021-03-19 | 百果园技术(新加坡)有限公司 | Video data sorting method and device, computer equipment and storage medium |
| CN113065067A (en) * | 2021-03-31 | 2021-07-02 | 达而观信息科技（上海）有限公司 | Article recommendation method and device, computer equipment and storage medium |
| CN111782925B (en) * | 2019-04-04 | 2024-05-14 | 阿里巴巴集团控股有限公司 | Item recommendation method, device, equipment, system and readable storage medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150178265A1 (en) * | 2013-12-20 | 2015-06-25 | Google Inc. | Content Recommendation System using a Neural Network Language Model |
| US9471314B1 (en) * | 2015-12-15 | 2016-10-18 | International Business Machines Corporation | Auxiliary perceptron branch predictor with magnitude usage limit |
| CN106227793A (en) * | 2016-07-20 | 2016-12-14 | 合网络技术（北京）有限公司 | A kind of video and the determination method and device of Video Key word degree of association |
| TW201713131A (en) * | 2015-06-11 | 2017-04-01 | Sony Corp | Control device, method, and computer program |
| CN108304440A (en) * | 2017-11-01 | 2018-07-20 | 腾讯科技（深圳）有限公司 | Method, apparatus, computer equipment and the storage medium of game push |
| CN108830543A (en) * | 2018-04-24 | 2018-11-16 | 平安科技（深圳）有限公司 | A kind of trip based reminding method and terminal device |
| CN109104620A (en) * | 2018-07-26 | 2018-12-28 | 腾讯科技（深圳）有限公司 | A kind of short video recommendation method, device and readable medium |
-
2019
- 2019-01-08 CN CN201910014819.6A patent/CN109558514B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150178265A1 (en) * | 2013-12-20 | 2015-06-25 | Google Inc. | Content Recommendation System using a Neural Network Language Model |
| TW201713131A (en) * | 2015-06-11 | 2017-04-01 | Sony Corp | Control device, method, and computer program |
| US9471314B1 (en) * | 2015-12-15 | 2016-10-18 | International Business Machines Corporation | Auxiliary perceptron branch predictor with magnitude usage limit |
| CN106227793A (en) * | 2016-07-20 | 2016-12-14 | 合网络技术（北京）有限公司 | A kind of video and the determination method and device of Video Key word degree of association |
| CN108304440A (en) * | 2017-11-01 | 2018-07-20 | 腾讯科技（深圳）有限公司 | Method, apparatus, computer equipment and the storage medium of game push |
| CN108830543A (en) * | 2018-04-24 | 2018-11-16 | 平安科技（深圳）有限公司 | A kind of trip based reminding method and terminal device |
| CN109104620A (en) * | 2018-07-26 | 2018-12-28 | 腾讯科技（深圳）有限公司 | A kind of short video recommendation method, device and readable medium |
Non-Patent Citations (3)
| Title |
|---|
| RAN DUAN 等: "Recoverable recommended keypoint-aware visual tracking using coupled-layer appearance modelling", 《2016 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS》 * |
| 刘恒友: "基于时间效应的推荐算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 高睿: "基于深度神经网络的视频个性化推荐系统研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111782925A (en) * | 2019-04-04 | 2020-10-16 | 阿里巴巴集团控股有限公司 | Item recommendation method, device, equipment, system and readable storage medium |
| CN111782925B (en) * | 2019-04-04 | 2024-05-14 | 阿里巴巴集团控股有限公司 | Item recommendation method, device, equipment, system and readable storage medium |
| CN112000872A (en) * | 2019-05-27 | 2020-11-27 | 北京地平线机器人技术研发有限公司 | Recommendation method based on user vector, training method of model and training device |
| CN110971973A (en) * | 2019-12-03 | 2020-04-07 | 北京奇艺世纪科技有限公司 | Video pushing method and device and electronic equipment |
| CN111125428A (en) * | 2019-12-17 | 2020-05-08 | 东北大学 | Time-dependent movie recommendation method based on score prediction function fitting structure |
| CN111813992A (en) * | 2020-07-14 | 2020-10-23 | 四川长虹电器股份有限公司 | Sorting system and method for movie recommendation candidate set |
| CN112528071A (en) * | 2020-10-30 | 2021-03-19 | 百果园技术(新加坡)有限公司 | Video data sorting method and device, computer equipment and storage medium |
| WO2022089467A1 (en) * | 2020-10-30 | 2022-05-05 | 百果园技术(新加坡)有限公司 | Video data sorting method and apparatus, computer device, and storage medium |
| CN113065067A (en) * | 2021-03-31 | 2021-07-02 | 达而观信息科技（上海）有限公司 | Article recommendation method and device, computer equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109558514B (en) | 2023-04-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109558514A (en) | Video recommendation method, its device, information processing equipment and storage medium | |
| CN107888950B (en) | A kind of method and system for recommending video | |
| CN108629665B (en) | Personalized commodity recommendation method and system | |
| CN110704674B (en) | Video playing integrity prediction method and device | |
| CN111797321B (en) | Personalized knowledge recommendation method and system for different scenes | |
| CN111080413A (en) | E-commerce platform commodity recommendation method and device, server and storage medium | |
| CN111461841B (en) | Article recommendation method, device, server and storage medium | |
| CN108431833A (en) | End-to-end depth collaborative filtering | |
| US20160212494A1 (en) | Video recommendation method and device | |
| CN109460514A (en) | Method and apparatus for pushed information | |
| US20230066853A1 (en) | Method and apparatus for training information prediction models, method and apparatus for predicting information, and storage medium and device thereof | |
| EP4181026A1 (en) | Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium | |
| CN108205768A (en) | Database building method and data recommendation method and device, equipment and storage medium | |
| CN101482884A (en) | Cooperation recommending system based on user predilection grade distribution | |
| CN109800328A (en) | Video recommendation method, its device, information processing equipment and storage medium | |
| CN110019943A (en) | Video recommendation method, device, electronic equipment and storage medium | |
| CN110490683B (en) | Offline collaborative multi-model hybrid recommendation method and system | |
| CN112395496A (en) | Information recommendation method and device, electronic equipment and storage medium | |
| KR101151176B1 (en) | The Content Recommendation Method by Using Menu Preferences | |
| CN114168843A (en) | Search word recommendation method, device and storage medium | |
| CN107506479B (en) | A kind of object recommendation method and apparatus | |
| CA3111094C (en) | Noise contrastive estimation for collaborative filtering | |
| CN117455687A (en) | Recommendation method and device for financial products, storage medium and electronic equipment | |
| CN108647227A (en) | A kind of recommendation method and device | |
| CN112989174A (en) | Information recommendation method and device, medium and equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |