CN104935963B - A kind of video recommendation method based on timing driving - Google Patents
A kind of video recommendation method based on timing driving Download PDFInfo
- Publication number
- CN104935963B CN104935963B CN201510290170.2A CN201510290170A CN104935963B CN 104935963 B CN104935963 B CN 104935963B CN 201510290170 A CN201510290170 A CN 201510290170A CN 104935963 B CN104935963 B CN 104935963B
- Authority
- CN
- China
- Prior art keywords
- user
- mrow
- interest
- project
- video
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
- H04N21/25—Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
- H04N21/258—Client or end-user data management, e.g. managing client capabilities, user preferences or demographics, processing of multiple end-users preferences to derive collaborative data
- H04N21/25866—Management of end-user data
- H04N21/25891—Management of end-user data being end-user preferences
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
- H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
- H04N21/4667—Processing of monitored end-user data, e.g. trend analysis based on the log file of viewer selections
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
- H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
- H04N21/4668—Learning process for intelligent management, e.g. learning user preferences for recommending movies for recommending content, e.g. movies
Abstract
The present invention relates to a kind of video recommendation method based on timing driving.This method includes：1) the interest graded by third party's data analysis user to video, and user interest gradient curve is obtained, the time point that the singular point of user interest gradient curve is migrated as user interest；2) the nearest interest transit time point of user is determined, the later user items scoring of the interest transit time point nearest to user is acquired, so as to establish the user items rating matrix for meeting user's current interest in seclected time window；3) the user items rating matrix is based on, the individualized video that user is carried out by using random walk model is recommended.The present invention considers the interest migration problem during individualized video is recommended, and has merged time window method and the trust metric model progress individualized video recommendation based on Random Walker, improves the accuracy and efficiency of video recommendations.
Description
Technical field
The invention belongs to video technique, video recommendations technical field, and in particular to a kind of regarding based on timing driving
Frequency recommends method.
Background technology
With the development of the communication technology, network availability bandwidth increases sharply, and viewing video is even mobile whole for pc user
End subscriber has all been a lax usual thing, and this also brings the development like a raging fire of internet video industry, and video pushes away
The system of recommending is the important component in internet video service.Initial video recommendation system is according to the similitude between film
Recommended, identical recommendation results can be provided for two users for watching same portion's film.But recently as individual character
Change successful application of the recommendation method on search engine and shopping website, video recommendations field is also increasingly paid attention to special for user
The individualized video of sign is recommended.Individualized video is recommended to consider the viewing history, social networks and currently playing shadow of user
The information such as piece, recommend the possible film interested of the user.It is contemplated that personalized recommendation method will also possess in video field
Wide application prospect and considerable profit are expected.
Individualized video recommends method to be roughly divided into content filtering method (Content-based Recommendation)
With collaborative filtering method (Collaborative Filtering Recommendation) two classes.Content filtering method according to
The film information that family is liked, other similar in terms of content videos, such as party A-subscriber are recommended to like to user《Bullet is allowed to fly》, that
Content filtering system can recommend Jiang Wen as director other works such as《It is one step away》.Content filtering method as can be seen here
Essence be calculate film similarity, generally, content filtering method first define a series of labels be used for characterize one regard
Frequency feature and user preferences, so as to which video similarity problem is converted into vector distance computational problem.Collaborative filtering method utilizes
User calculates user's similitude or item similarity to the history score data of project, and then is recommended according to similitude.
Collaborative filtering method records scoring of all users to all items, forms user-project rating matrix of a M*N dimension, its
Middle M is number of users, and N is item number, a unit R in matrixm,nScorings of the user m to project n is represented, per a line in matrix
A user vector is represented, the similarity between user vector, namely the phase between user can be obtained by similarity calculating method
Like degree；Each row in matrix represent a project, it is same can obtain project with similarity calculating method between similarity.
According to the similarity between user, the user project for recommending the user for there are similar interests with him to be liked can be given, it is this to push away
Recommend and be called the collaborative filtering (UserCF) based on user；According to the similarity between project, a user can to recommend to like with him
The similar project of joyous project, this recommendation are called project-based collaborative filtering (ItemCF).
No matter information filtering algorithm or collaborative filtering, be all the angle from user interest, for user recommend
Its film that may like, but in actual life, what the interest of user often gradually changed over time, this phenomenon
It is called interest migration.If to user do not account for the interest migration of user during individualized video recommendation, just mean
The system of wearing may only recommend the film liked before him to user, but the interest nearest with it is not inconsistent, and this reduces recommendation
Accuracy.
The content of the invention
The present invention is directed to the user interest migration problem in individualized video commending system, there is provided one kind is based on time series data
The video recommendation method of excavation, it is possible to increase the accuracy and efficiency of video recommendations.
The technical solution adopted by the present invention is as follows：
A kind of video recommendation method based on timing driving, its step include：
1) the interest graded by third party's data analysis user to video, and user interest gradient curve is obtained,
The time point that the singular point of user interest gradient curve is migrated as user interest；
2) the nearest interest transit time point of user is determined, the later user of the interest transit time point nearest to user-
Project scoring is acquired, so as to establish the user for meeting user's current interest-project rating matrix in seclected time window；
3) user-project rating matrix is based on, the individualized video of user is carried out by using random walk model
Recommend.
Further, step 1) third party's data are preferably bean cotyledon scoring, and the bean cotyledon by analyzing user, which scores, to be gone through
History draws user interest gradient curve.
Further, the method that step 1) obtains the user interest gradient curve is：Each attribute of video is formed
One attribute vector；At each time point, by each attribute of video, according to current time and before, the user of time scores
Record is weighted, and obtains a user interest vector；It is emerging according to the sequence of user at user's multiple time points on a timeline
Inclination amount, analyze the user interest gradient curve changed over time.Each attribute of the video includes：Title, performer,
Director, age, brief introduction etc..
Further, the step 2) user-project scoring includes explicit scoring and implicit scores；The explicit scoring is
When user watches video, user is to the scoring for the video progress being currently viewed；The implicit scores are record users
Watch, redirect, collecting, sharing action, the fancy grade according to the different action prediction users of user to video.
Further, step 2) is by being quantified to user-project scoring and normalizing process, when obtaining interest migration
Between later user-project rating matrix, every a line in matrix represents a user characteristics, and each row represent a project spy
Sign.
Further, step 3) calculates user u according to user-project rating matrix first and is in project j's in kth step
Transition probability, user u is then calculated to project j global probability, then calculates global probability of all users to all items,
So as to obtain user-entry sorting matrixIndividualized video recommendation is carried out according to the matrix.
The present invention key point be：By third party's data to user modeling, user interest migration is found；By determining to use
The nearest interest transit time point in family, establishes user-project rating matrix；Collaborative filtering is replaced by using random walk model
Method carries out individualized video recommendation.
The present invention is directed to the user interest migration problem in individualized video commending system, is used using third party's data analysis
Family interest graded, user interest transit time is found, so that it is determined that suitable training time window size；In time window
It is interior, trust metric model is created based on Random Walker random algorithms, so as to realize that the individualized video for some user pushes away
Recommend.This method considers the interest migration problem during individualized video is recommended, and has merged time window method and be based on
Random Walker trust metric model carries out individualized video recommendation, improves the accuracy and efficiency of video recommendations.
Brief description of the drawings
Fig. 1 is the video recommendation method overview flow chart based on timing driving.
Fig. 2 is the user interest migration detection curve based on interest gradient analysis.
Embodiment
In order to facilitate the understanding of the purposes, features and advantages of the present invention, below by specific embodiment and
Accompanying drawing, the present invention will be further described.
The present invention has found user interest migration using third party's data (for example being scored using bean cotyledon), when determining training data
Between window size, so as to shield the influence of user interest migration, and then entered using the trust metric model based on Random Walker
Row individualized video is recommended.Fig. 1 is the overview flow chart of the inventive method.
1. the user interest migration based on third party's data is found
As it was previously stated, usage time windowhood method of the present invention overcomes user interest migration influence to caused by recommendation results.
The time window of fixed size is unable to the interest migration of accurate response user, therefore, in order to obtain suitable time window size,
The present invention uses third party's data analysis user interest graded, estimating subscriber's interest transit time, so that it is determined that time window
The size of mouth.
Recommended due to of the invention for video resource, and in the network environment of China, bean cotyledon scoring is undoubtedly
It is best suitable for the third party website of objective evaluation user video interest.Therefore, the bean cotyledon scoring history of present invention analysis user, draws
User interest gradient curve, and wherein nearest singular point is migrated to the mark originated as user's the last time interest, with this
To determine the size of time window in video recommendations model.
1) bean cotyledon user data models
Bean cotyledon is that platform is shared and commented on to the film of largest domestic, and the registration amount more than ten million user causes bean cotyledon scoring to exist
Video service industry enjoys very big influence power.User on bean cotyledon website scores using very full marks system, therefore eliminate number
According to normalization process, directly history scoring record of the user on bean cotyledon can be brought as training data.
2) user interest models
In order to which film is described, the present invention using the title of film, performer, director, age, brief introduction etc. as one to
Amount, a film is represented with the attribute vector.Such as in table 1 film a cans with [Name a, Actor a Actor b,
Director a, 2014, Keyword a Keyword d] as attribute vector represent.
Scoring record of film is watched it according to some user, the attribute of each film can be added by scoring
Power, so as to obtain the hobby characteristic vector of the user.For example certain user A watched 4 films altogether, respectively film a, film b,
Film c, film d, and scorings of the A to four films records as shown in table 1.
The user's viewing history of table 1 and scoring record sheet
So, the interest vector can of the user is expressed as I (t)=[name a*9+name b*5+name c*9+
name d*9,actor a*24+actor b*9+actor c*14+actor d*18,director a*18+director b*
9+director c*9,2014*27+2010*5,keyword a*18+keyword b*9+keyword c*9+keyword e*
9+keyword f*9+keyword g*5+keyword h*5]=[name a, actor a, director a, 2014,
Keyword a] (it is assumed that each single item only takes a value by weight sequencing in interest vector, weights identical takes a value at random).
Over time, the scoring record of user can be more and more, therefore at each time point, we can root
Score and record according to current time and the before history of time, obtain a user interest vector, accordingly, we can be somebody's turn to do
The interest vector at user's multiple time points on a timeline, each interest vector represent user and regarded in corresponding time point viewing
The preference of frequency.
3) user interest migration is found
According to above-mentioned sequence of user interest vector, the user interest gradient changed over time can be analyzed, is looked for
The time point migrated to the singular point of gradient curve as interest.The point is the mark of user interest migration, so as to be follow-up
Collaborative filtering video recommendations algorithm provide time window size foundation.
Some time point t interest gradient calculation formula is：
D (t)=I (t)-I (t-1)
Wherein, D (t) is the user interest gradient of t；I (t) is the user interest vector of t；When I (t-1) is t
Carve the user interest vector of previous moment.In order to facilitate calculating, it is specified that the rule in subtraction is, it is if identical subtract each other
0, it is 1 if difference is subtracted each other, the otherness between two vectors is embodied with this.Such as [name a, actor a,
Director a, 2014, keyword a]-[name b, actor a, director c, 2014, keyword a]=[1,0,
1,0,0].As can be seen here, the modulus value of user interest gradient vector is bigger, and it is more obvious just to represent interests change.
Fig. 2 show interest gradient, and wherein transverse axis is the time, and the longitudinal axis is the modulus value of interest gradient.From the figure
In it can be seen that, there occurs one respectively 2 months 2009, in December, 2009, in June, 2012 and in June, 2014 for the interest of user
Secondary interest migration.
2. user-project rating matrix
Migrated according to above-mentioned interest and find result, it may be determined that the nearest interest transit time point of user.According to the time
It is nonsensical that user items scoring record before point, which does video recommendations, because that does not meet the newest interest of user.Cause
This present invention user later to interest transit time point-project scoring is acquired, and each user is best suited so as to obtain
The user of current interest-project rating matrix.
1) user-project scoring collection
In video recommendation system, can there are explicit scoring and two kinds of users of implicit scores-project scoring acquisition mode.Its
In, explicit scoring is when user watches video, it is allowed to which user is scored the video being currently viewed (generally 5 points
Full marks or 10 points of full marks systems)；Implicit scores are to record the viewing of user, the action such as redirect, collect, sharing, according to user's
Fancy grade of the different action prediction users to the video.The user of the present invention-project scoring takes into account explicit scoring and implicitly commented
It is divided to two kinds of scoring acquisition modes, so as to more all-sidedly and accurately react fancy grade of the user to each video.
2) user-project scoring normalization
In the present invention, explicit scoring refers to score made during user's viewing video watching film, for side
Continue after an action of the bowels and calculate, all explicit scorings are normalized into full marks 1 divides.Implicit scores include the viewing of user, redirect, collect, dividing
The action such as enjoy, assume that score value corresponding to the action of user is as shown in table 2 in the present invention.
The user concealed scoring of table 2 quantifies and normalization table
| Action | Share | Collection | Watch and watch ending | Watch and redirect halfway |
| Scoring | 10 | 9 | 8 | 5 |
| Normalization scoring | 1 | 0.9 | 0.8 | 0.5 |
By above-mentioned quantization and normalization process, the later user of interest transit time-project rating matrix can be obtained,
Every a line in matrix represents a user characteristics, and each row represent an item characteristic.
3. the video recommendations based on trust metric model
As it was previously stated, according to the scoring of each user and viewing behavior, can obtain user to the explicit of film or
Implicit marking, so as to build user-project matrix in seclected time window.In general, user-project matrix is sparse square
Battle array, in order to solve the problems, such as that user-project matrix is openness, the present invention is entered using random walk model instead of collaborative filtering method
Row individualized video is recommended.
1) Random Walker algorithms
Random Walker algorithms use Markov model, the migration since a point, until complete collection of traversal
Close., all can be with probability 1-Pr (X in any one pointu,k+1|Xu,k) migration is to some neighbor node, with probability P r (Xu,k+1|
Xu,k) any point (including itself) in random skip to set.A migration is often carried out, reaches the probability of some node only
By being currently located Determines.Wherein, Pr (Xu,k+1|Xu,k) be random walk transition probability parameter, Xu,k+1Represent in u situations
Under, the state at k+1 time points；Xu,kRepresent in the case of u, the state at k time points.
The migration each time of random walk can obtain a probability distribution, and the distribution features each point quilt in set
The probability having access to.The process of whole random walk is exactly by the use of above-mentioned probability distribution as the input of next migration and iterated
Process, when meeting certain condition, the distribution probability can tend to restrain, so as to obtain a stable probability distribution.At random
Migration model is used widely in fields such as data minings, for example classical PageRank algorithms are exactly one of random walk
Example.
2) individualized video based on random walk model is recommended
Assuming that I represents the Item Sets of video presentation in individualized video commending system, m is the size of Item Sets, then can be with
Obtain the oriented weighted graph of m node, way interior joint nodeiRepresent project i, side edgei,jWeight Pi,jRepresent from project i
It is transferred to project j probability Pi,j=Pr (Xu,k+1=j | Xu,k=i)
According to above-mentioned transition probability, the item advance probability matrix P of a m*m rank can be obtained, each row generation in matrix
Table is from m item advance to project j transition probability vector.If U represents user's collection of commending system, n is the scale of user's collection,
Initial user-project rating matrix R of n*m ranks, each of which row generation can be so established according to the history of user scoring record
The scoring vector of table user, each element RuiHistory score datas of the user u to project i is represented, sets Rui=Pr (Xu,0=
i).User u then can be calculated in transition probability of the kth step in project j according to user-project rating matrix：
The α factors are the probability that user continues next step under current state in above formula, and the parameter can control
The step-length of random walk, rule of thumb, the numerical value should increase with the reduction of training set ratio.It can thus be concluded that u couples of user
Project j global probability：
Wherein c represents constant, and global probability of all users to all items can be calculated according to above formula, so as to
To user-entry sorting matrix
Interest level of the user to any one project can be estimated by the matrix, so as to recommend for individualized video
Reference frame is provided.
The above embodiments are merely illustrative of the technical solutions of the present invention rather than is limited, the ordinary skill of this area
Technical scheme can be modified by personnel or equivalent substitution, without departing from the spirit and scope of the present invention, this
The protection domain of invention should be to be defined described in claims.
Claims (7)
1. a kind of video recommendation method based on timing driving, its step include：
1) the interest graded by third party's data analysis user to video, and user interest gradient curve is obtained, it will use
The time point that the singular point of family interest gradient curve migrates as user interest；
Obtaining the method for the user interest gradient curve is：Each attribute of video is formed into an attribute vector；Each
At time point, by each attribute of video, according to current time and before, user's scoring record of time is weighted, and obtains one
Individual user interest vector；According to the sequence of user interest vector at user's multiple time points on a timeline, analysis obtains at any time
Between the user interest gradient curve that changes；The transverse axis of the user interest gradient curve is the time, and the longitudinal axis is the mould of interest gradient
Value；
The calculation formula of some time point t interest gradient is：
D (t)=I (t)-I (t-1),
Wherein, D (t) is the user interest gradient of t；I (t) is the user interest vector of t；I (t-1) is before t
The user interest vector at one moment；Rule in regulation subtraction is, is 0 if identical subtract each other, if difference is subtracted each other
For 1；
2) the nearest interest transit time point of user, the later user-project of the interest transit time point nearest to user are determined
Scoring is acquired, so as to establish the user for meeting user's current interest-project rating matrix in seclected time window；It is described
After time window is the nearest interest transit time point of user；
3) user-project rating matrix is based on, the individualized video that user is carried out using random walk model is recommended.
2. the method as described in claim 1, it is characterised in that：Step 1) third party's data score for bean cotyledon, by dividing
The bean cotyledon scoring history for analysing user draws user interest gradient curve.
3. the method as described in claim 1, it is characterised in that each attribute of the video includes：Title, performer, director,
Age and brief introduction.
4. method as claimed in claim 1 or 2, it is characterised in that the step 2) user-project scoring includes explicit scoring
And implicit scores；The explicit scoring is the scoring that user is carried out to the video being currently viewed when user watches video；
The implicit scores are to record the viewing of user, redirect, collect, sharing action, according to the different action prediction users couple of user
The fancy grade of video.
5. method as claimed in claim 4, it is characterised in that step 2) is by being quantified and being returned to user-project scoring
One changes process, obtains the later user of interest transit time-project rating matrix, and every a line in matrix represents a user spy
Sign, each row represent an item characteristic.
6. method as claimed in claim 1 or 2, it is characterised in that step 3) calculates according to user-project rating matrix first
Go out the transition probability that user u is in project j in kth step, then calculate global probability of the user u to project j, then calculate all
User is to the global probability of all items, so as to obtain user-entry sorting matrixAnd personalization is carried out according to the matrix and regarded
Frequency is recommended.
7. method as claimed in claim 6, it is characterised in that the specific method of step 3) is：
Assuming that I represents the Item Sets of video presentation in individualized video commending system, m is the size of Item Sets, has obtained m first
The oriented weighted graph of individual node, way interior joint nodeiRepresent project i, side edgei,jWeight Pi,jRepresentative is transferred to from project i
Project j probability Pi,j=Pr (Xu,k+1=j | Xu,k=i)；Wherein Xu,k+1Represent in user u in the state at k+1 time points, Xu,k
Represent the state at k time points in user u；
According to above-mentioned transition probability, the item advance probability matrix P of a m*m rank is obtained, each row are represented from m item in matrix
Mesh is transferred to project j transition probability vector；If U represents user's collection of commending system, n is the scale of user's collection, according to user
History scoring record establish initial user-project rating matrix R of n*m ranks, each of which row represents the scoring vector of user,
Each element RuiHistory score datas of the user u to project i is represented, sets Rui=Pr (Xu,0=i), wherein Xu,0Represent user u
Init state, according to user-project rating matrix calculate user u kth step in project j transition probability：
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mi>Pr</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>X</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>=</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>&alpha;</mi>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>m</mi>
</munderover>
<mi>Pr</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>X</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>k</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>=</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<msub>
<mi>P</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mo>=</mo>
<msup>
<mi>&alpha;</mi>
<mi>k</mi>
</msup>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>m</mi>
</munderover>
<msubsup>
<mi>R</mi>
<mrow>
<mi>u</mi>
<mi>i</mi>
</mrow>
<mo>*</mo>
</msubsup>
<msubsup>
<mi>P</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
<mi>k</mi>
</msubsup>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mo>=</mo>
<msup>
<mi>&alpha;</mi>
<mi>k</mi>
</msup>
<msubsup>
<mover>
<mi>R</mi>
<mo>&OverBar;</mo>
</mover>
<mrow>
<mi>u</mi>
<mo>.</mo>
</mrow>
<mo>*</mo>
</msubsup>
<mo>&CenterDot;</mo>
<msubsup>
<mover>
<mi>P</mi>
<mo>&OverBar;</mo>
</mover>
<mrow>
<mo>-</mo>
<mi>j</mi>
</mrow>
<mi>k</mi>
</msubsup>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>,</mo>
</mrow>
Wherein, the α factors are the probability that user continues next step under current state；It can thus be concluded that user u is to project j
Global probability：
<mrow>
<mi>Pr</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>X</mi>
<mi>u</mi>
</msub>
<mo>=</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
</munder>
<mi>Pr</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>X</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>=</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>x</mi>
</munderover>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>></mo>
<mn>1</mn>
</mrow>
<mi>m</mi>
</munderover>
<mi>Pr</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>X</mi>
<mrow>
<mi>u</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>=</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>=</mo>
<mi>c</mi>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>x</mi>
</munderover>
<msup>
<mi>&alpha;</mi>
<mi>k</mi>
</msup>
<msubsup>
<mover>
<mi>R</mi>
<mo>&OverBar;</mo>
</mover>
<mrow>
<mi>u</mi>
<mo>-</mo>
</mrow>
<mo>*</mo>
</msubsup>
<mo>&CenterDot;</mo>
<msubsup>
<mover>
<mi>P</mi>
<mo>&OverBar;</mo>
</mover>
<mrow>
<mo>-</mo>
<mi>j</mi>
</mrow>
<mi>k</mi>
</msubsup>
<mo>,</mo>
</mrow>
Wherein c represents constant, global probability of all users to all items is calculated according to above formula, so as to obtain user-project
Ordinal matrix
<mrow>
<mover>
<mi>R</mi>
<mo>~</mo>
</mover>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>&alpha;</mi>
</munderover>
<msup>
<mi>&alpha;</mi>
<mi>k</mi>
</msup>
<msup>
<mi>RP</mi>
<mi>k</mi>
</msup>
<mo>=</mo>
<mi>R</mi>
<mi>&alpha;</mi>
<mi>P</mi>
<msup>
<mrow>
<mo>(</mo>
<mi>I</mi>
<mo>-</mo>
<mi>&alpha;</mi>
<mi>P</mi>
<mo>)</mo>
</mrow>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<mo>,</mo>
</mrow>
Interest level according to Matrix Estimation user to any one project, so as to realize that the individualized video of user pushes away
Recommend.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510290170.2A CN104935963B (en) | 2015-05-29 | 2015-05-29 | A kind of video recommendation method based on timing driving |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510290170.2A CN104935963B (en) | 2015-05-29 | 2015-05-29 | A kind of video recommendation method based on timing driving |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104935963A CN104935963A (en) | 2015-09-23 |
| CN104935963B true CN104935963B (en) | 2018-03-16 |
Family
ID=54122874
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510290170.2A Active CN104935963B (en) | 2015-05-29 | 2015-05-29 | A kind of video recommendation method based on timing driving |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104935963B (en) |
Families Citing this family (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105447087B (en) * | 2015-11-06 | 2021-08-24 | 腾讯科技（深圳）有限公司 | Video recommendation method and device |
| CN106919582A (en) * | 2015-12-24 | 2017-07-04 | 阿里巴巴集团控股有限公司 | The association of network articles and related information statistical method and device |
| CN105740473B (en) * | 2016-03-14 | 2021-03-02 | 腾讯科技（深圳）有限公司 | User generated content display method and device |
| CN105913296B (en) * | 2016-04-01 | 2020-01-03 | 北京理工大学 | Personalized recommendation method based on graph |
| CN107544981B (en) * | 2016-06-25 | 2021-06-01 | 华为技术有限公司 | Content recommendation method and device |
| CN106372101B (en) * | 2016-07-20 | 2019-09-17 | Tcl集团股份有限公司 | A kind of video recommendation method and device |
| CN106168980B (en) * | 2016-07-26 | 2020-07-28 | 阿里巴巴（中国）有限公司 | Multimedia resource recommendation sequencing method and device |
| CN106250499B (en) * | 2016-08-02 | 2020-07-14 | 阿里巴巴（中国）有限公司 | Video pair mining method and device |
| CN107220329B (en) * | 2017-05-24 | 2021-01-19 | 苏州百智通信息技术有限公司 | Video resource evaluation method and device |
| CN108257035A (en) * | 2017-12-30 | 2018-07-06 | 黄敬易 | User's similarity calculating method, device, server and readable storage medium storing program for executing |
| CN108170868A (en) * | 2018-02-09 | 2018-06-15 | 宁夏灵智科技有限公司 | Video recommendation method and device |
| CN108415987B (en) * | 2018-02-12 | 2021-05-11 | 大连理工大学 | Cold start solution for movie recommendation |
| CN110390415A (en) | 2018-04-18 | 2019-10-29 | 北京嘀嘀无限科技发展有限公司 | A kind of method and system carrying out trip mode recommendation based on user's trip big data |
| CN108628999B (en) * | 2018-05-02 | 2022-11-11 | 南京大学 | Video recommendation method based on explicit and implicit information |
| CN109189988B (en) * | 2018-09-18 | 2021-06-22 | 北京邮电大学 | Video recommendation method |
| CN109389168B (en) * | 2018-09-29 | 2021-07-13 | 国信优易数据股份有限公司 | Project recommendation model training method, project recommendation method and device |
| CN109587527B (en) * | 2018-11-09 | 2021-06-04 | 青岛聚看云科技有限公司 | Personalized video recommendation method and device |
| CN109543111B (en) * | 2018-11-28 | 2021-09-21 | 广州虎牙信息科技有限公司 | Recommendation information screening method and device, storage medium and server |
| CN111292151B (en) * | 2018-12-07 | 2023-07-25 | 中移互联网有限公司 | Object recommendation method and device, storage medium and object recommendation equipment |
| CN109800328B (en) * | 2019-01-08 | 2021-08-20 | 青岛聚看云科技有限公司 | Video recommendation method, device thereof, information processing equipment and storage medium |
| CN110175286B (en) * | 2019-05-17 | 2021-05-11 | 山东师范大学 | Product recommendation method and system combining pairwise optimization and matrix decomposition |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101551825A (en) * | 2009-05-15 | 2009-10-07 | 中国科学技术大学 | Personalized film recommendation system and method based on attribute description |
| CN104063481A (en) * | 2014-07-02 | 2014-09-24 | 山东大学 | Film individuation recommendation method based on user real-time interest vectors |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2727335A4 (en) * | 2011-06-28 | 2015-01-14 | Tata Consultancy Services Ltd | A method and system for context-aware recommendation |
-
2015
- 2015-05-29 CN CN201510290170.2A patent/CN104935963B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101551825A (en) * | 2009-05-15 | 2009-10-07 | 中国科学技术大学 | Personalized film recommendation system and method based on attribute description |
| CN104063481A (en) * | 2014-07-02 | 2014-09-24 | 山东大学 | Film individuation recommendation method based on user real-time interest vectors |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104935963A (en) | 2015-09-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104935963B (en) | A kind of video recommendation method based on timing driving | |
| CN108256093B (en) | Collaborative filtering recommendation algorithm based on multiple interests and interest changes of users | |
| CN103514304B (en) | Project recommendation method and device | |
| CN106802956B (en) | Movie recommendation method based on weighted heterogeneous information network | |
| TWI636416B (en) | Method and system for multi-phase ranking for content personalization | |
| US9171078B2 (en) | Automatic recommendation of vertical search engines | |
| CN104063481B (en) | A kind of film personalized recommendation method based on the real-time interest vector of user | |
| CN103533390B (en) | The method and system of television program recommendations are carried out based on social network information | |
| EP2344958B1 (en) | Image-based semantic distance | |
| CN110879864B (en) | Context recommendation method based on graph neural network and attention mechanism | |
| CN108920527A (en) | A kind of personalized recommendation method of knowledge based map | |
| US20090006368A1 (en) | Automatic Video Recommendation | |
| CN108334592B (en) | Personalized recommendation method based on combination of content and collaborative filtering | |
| CN103955535A (en) | Individualized recommending method and system based on element path | |
| CN106599226A (en) | Content recommendation method and content recommendation system | |
| CN107301247B (en) | Method and device for establishing click rate estimation model, terminal and storage medium | |
| Benouaret et al. | A package recommendation framework for trip planning activities | |
| KR20150023432A (en) | Method and apparatus for inferring user demographics | |
| CN103150667B (en) | A kind of personalized recommendation method based on body construction | |
| CN104899246A (en) | Collaborative filtering recommendation method of user rating neighborhood information based on fuzzy mechanism | |
| CN112966091B (en) | Knowledge map recommendation system fusing entity information and heat | |
| CN110083764A (en) | A kind of collaborative filtering cold start-up way to solve the problem | |
| CN105426550A (en) | Collaborative filtering tag recommendation method and system based on user quality model | |
| Liu et al. | Using collaborative filtering algorithms combined with Doc2Vec for movie recommendation | |
| CN105843799A (en) | Academic paper label recommendation method based on multi-source heterogeneous information graph model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |