CN104063481B - A kind of film personalized recommendation method based on the real-time interest vector of user - Google Patents
A kind of film personalized recommendation method based on the real-time interest vector of user Download PDFInfo
- Publication number
- CN104063481B CN104063481B CN201410313446.XA CN201410313446A CN104063481B CN 104063481 B CN104063481 B CN 104063481B CN 201410313446 A CN201410313446 A CN 201410313446A CN 104063481 B CN104063481 B CN 104063481B
- Authority
- CN
- China
- Prior art keywords
- msub
- mrow
- user
- film
- vector
- 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.)
- Expired - Fee Related
Links
- 239000013598 vector Substances 0.000 title claims abstract description 132
- 238000000034 method Methods 0.000 title claims abstract description 25
- 239000011159 matrix material Substances 0.000 claims abstract description 27
- 238000001914 filtration Methods 0.000 claims abstract description 13
- 239000012141 concentrate Substances 0.000 claims description 7
- 238000012804 iterative process Methods 0.000 claims description 5
- 238000000354 decomposition reaction Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 3
- 230000001550 time effect Effects 0.000 claims description 3
- 238000000205 computational method Methods 0.000 claims 1
- 230000005611 electricity Effects 0.000 claims 1
- 230000008859 change Effects 0.000 abstract description 8
- 230000008569 process Effects 0.000 abstract description 4
- 206010068052 Mosaicism Diseases 0.000 abstract description 3
- 210000003765 sex chromosome Anatomy 0.000 abstract description 3
- 238000004364 calculation method Methods 0.000 description 3
- 230000002349 favourable effect Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses the film personalized recommendation method of a kind of comprehensive movie contents and the real-time score information of user, mainly solves the problems, such as interests change and Deta sparseness that conventional recommendation algorithm can not reflect user in time.In order to solve Sparse sex chromosome mosaicism, invention introduces user interest vector.Start with from movie features vector, handle to obtain the interest characteristics vector of user in an iterative manner by the rating matrix of user, according to obtained user characteristics vector structure user's similar matrix, finally complete to recommend according to traditional collaborative filtering score in predicting formula.For the situation of user interest change, time factor has been incorporated again in user interest vector process is built so that the scoring behavior weight closer to current time is bigger, can more show the interest of user.
Description
Technical field
The present invention be the real-time score data of synthetic user and the characteristic attribute of film caused by recommendation method, mainly solve
Deta sparseness and the problem of can not reflect user interest change in time in traditional Collaborative Filtering Recommendation Algorithm, belongs to multimedia
Field of information processing.
Background technology
As internet has brought us into the epoch of information explosion, movie contents were also rapidly increasing, surrounded in recent years
The website and application that movie contents correlation is reached the standard grade are even more countless.For a user, in face of so abundant movie resource,
Want therefrom to choose really desired content extremely difficult.Commending system turns into the effective way for solving these problems, and recommends
Algorithm is to influence the key that commending system function is realized.The proposed algorithm of main flow mainly includes content-based recommendation calculation at present
Method, the proposed algorithm of collaborative filtering and mixing proposed algorithm etc..
Content-based recommendation main thought is exactly the film that the film that favorable comment was carried out to user is similar as recommendation
As a result, it is simple, effective the advantages of the algorithm, shortcoming is that recommendation results lack novelty, it is impossible to find that new sense is emerging for user
The article of interest.
And Collaborative Filtering Recommendation Algorithm then calculates the similitude between different user, Ran Houli using the historical information of user
Fancy grade of the targeted customer to particular film is predicted to the evaluation of film with the neighbours higher with targeted customer's similitude.With
Similarity between family is that the rating matrix of film is obtained according to user.Assuming that user is to the same as the scoring on some films
It is more similar, then it represents that the interest of two users is also more similar.Although Collaborative Filtering Recommendation System is widely used,
It is also to face the problem of many real.For example in internet video business, because the consciousness of user's marking is not very strong, make
Into in video recommendations there is obvious Sparse sex chromosome mosaicism, and collaborative filtering recommending give a mark it is sparse in the case of, property
Can be poor.An also problem, traditional collaborative filtering do not take into account that the hobby of user can dynamically become with the time
Change, the film for causing to recommend may not be user's current interest.
The content of the invention
The present invention is to solve Deta sparseness in traditional Collaborative Filtering Recommendation Algorithm and can not reflect that user is emerging in time
The problem of interest change, and the generation of the characteristic attribute of a kind of real-time score data of the synthetic user proposed and film is real based on user
When interest vector recommendation method.
In order to realize object of the invention it is proposed that a kind of film personalized recommendation side based on the real-time interest vector of user
Method, the feature of the rating matrix of user and film is combined, incorporates time factor, obtain that the real-time interest of user can be reflected
Characteristic vector.Obtained user interest vector can both solve Sparse sex chromosome mosaicism, also reflect the interest of user in real time
Change.The feature of all films in recommended models is passed through VSM (Vector by the method for the present invention when selecting neighbor user
Space Model) characteristic vector form represent；The interest vector model of user is directed to again, time factor is incorporated, by user's
The characteristic vector of rating matrix and film is modeled in an iterative manner；The interest vector of user is processed into and movie features
The similar form of vector；The similitude between user is finally calculated according to the user interest of output vector, used so as to obtain target
The neighbor user at family.The present invention is used as power when predicting user's scoring, using the similitude of neighbours and the interest vector of targeted customer
Weight, using the score in predicting mechanism of the collaborative filtering based on user of classics, the corresponding scoring of targeted customer is predicted, finally
It is ranked up by the prediction scoring of all films that do not score of the targeted customer to obtaining, draws topN recommendation lists.It is based on
The proposed algorithm framework of user interest vector is as shown in Figure 1.
The technical solution adopted by the present invention is：
A kind of film personalized recommendation method based on the real-time interest vector of user, comprises the following steps：
Step 1：The characteristic attribute of every film is modeled according to the metamessage of film, obtains the characteristic vector of film,
Film tiCharacteristic vector be expressed asWhereinObtained by following formula：
The attribute p for representing film metamessage is included even in the filmj, then weights corresponding to the attributeAdd 1, if in the film
Not comprising attribute pj, then weights corresponding to the attributeIt is zero, T={ t1, t2..., tnRepresent film set；
Step 2：The real-time interest vector of user is obtained by the real-time score information and the movie features vector of user, used
Family set expression is U={ u1, u2..., un, user uiReal-time interest vector be expressed as
WhereinRepresent user uiTo attribute pjFavorable rating, user uiReal-time interest vectorIt is by uiScored film
Characteristic vectorWeighted sum, weight is real-time fancy grade of the user to film；
Step 3：The score information of the user interest vector sum user obtained with reference to the step 2 is to movie features vector
It is updated, generates more rational movie features vector.
Step 4：Iterative process is formed by the step 2 and the step 3, the real-time interest vector of user is carried out
Renewal, obtains the real-time interest vector of more accurate user；
Step 5：The similarity matrix of user is established by the more accurate real-time interest vector of user, and then obtained
The neighbor user of targeted customer, corresponding scoring is beaten according to the score in predicting formula of the collaborative filtering based on user, finally
Complete topN recommendation lists.
Preferably, in step 1, dimensionality reduction is carried out to the characteristic vector of film using singular value decomposition method.
Preferably, the vector of user interest described in step 2It can be obtained by following step：
A) the scoring film collection of each user is obtained from rating matrix
B) characteristic vector that user's scoring film concentrates film is obtained from film-eigenvectors matrix
C) user uiReal-time interest vectorIt is expressed asIn each film characteristic vectorWeighted sum, weight
It is exactly real-time fancy grade of the user to film, specific formula is：
WhereinRepresent attribute pjReverse features frequency, by total number of users mesh in system divided by like attribute pjUse
The number at family, then obtained business is taken the logarithm to obtain,It is user uiTo film tjScoring,It is user uiBe averaged and comment
Point,Represent film tjTo user uiTime-based weighting function；
Above formula is decomposed into the real-time interest vector of userMiddle attribute pjWeightCalculation formula be：
WhereinRepresent attribute pjIn film tjWeight in characteristic vector,Represent vectorBy maximum
Weight, allUserNumber are total number of users mesh in system, and preferedUserNumber is to like attribute pjUser number
Mesh；
In above-mentioned formulaIt is film tjTo user uiTime-based weighting function, scoring the time from current
Nearlyer weight is bigger, otherwise weight is smaller, is represented by below equation：
Wherein dnowThe current time is represented,Represent user uiTo film tjThe scoring time, when β ∈ (0,1) are referred to as
Between Effects of Factors parameter, the bigger time factors of βTo be appraised point of time effects are bigger, i.e., user comments undue the nearest period
The weight of film is bigger, and the weight for the film that scores in the past is smaller, can be by adjusting β value come excellent according to different commending systems
Change recommendation results.
Preferably, more rational movie features vector can be obtained by following step described in step 3：
A) obtained from rating matrix to film tjUndue user is commented to collect
B) interest vector that scoring user concentrates user is obtained from user-interest vector matrix
C) film t is obtained with reference to the rating matrix of userjMore rational characteristic vector,It is film tjRenewal obtains
Characteristic vector in pjThe weight of feature, formula are as follows：
Wherein, uiIt is to film tjUndue user is commented,It is user uiInterest vector in feature pjWeight,Effect is to make normalized.
Preferably, the iterations of iterative process described in step 4 is adjusted according to different commending systems, with excellent
Change recommendation results.
The beneficial effects of the present invention are：It is truer that user according to the content information of film is obtained by the thought of iteration
Interest vector, so being built upon to the interest modeling of user in the hobby feature to film, to user modeling granularity more
Adduction is managed, while also solves the problems, such as Deta sparseness, improves the accuracy of recommendation.Establishing the user characteristics vector
While add time factor so that recommendation results more conform to the recent hobby of user.
Brief description of the drawings
Fig. 1 is the film personalized recommendation method frame diagram based on the real-time interest vector of user of the present invention；
Fig. 2 is the iterative process of user interest vector sum movie features vector in the inventive method；
Fig. 3 is recommendation precision of the inventive method under neighbor user quantity different situations.
Embodiment
The present invention is described further with specific embodiment below in conjunction with the accompanying drawings.
Before the specific steps of method are introduced, several related definitions are first provided.
Define 1. attribute pj：The metamessage of film in database, such as the school or type of director, playwright, screenwriter, protagonist and film
Deng.
Define 2. film tiCharacteristic vectorBy film tiMetamessage pass through model it is public
The vector that formula obtains,InRepresent the attribute p in film metamessagejImportance to describing the film.
Define the real-time interest vector of 3. usersRepresent user to film tiAttribute pj
Real-time favorable rating, numerical value more it is big then represent user's proximal segment time to respective attributes pjFavorable rating it is bigger.
Define 4. time factorsRepresent film tjTo user uiTime-based weight factor, scoring the time from
Current nearlyer weight is bigger, otherwise weight is smaller.
Define 5. Reverse features frequenciesIt is attribute pjThe measurement of general importance.By total number of users mesh in system divided by
Like attribute pjUser number, then obtained business is taken the logarithm to obtain.
The present invention is to obtain the real-time interest vector of user using the metamessage of film and the score information of user, is used
Family similar matrix, for the recommendation of film.It is as follows to implement step：
Step 1：The characteristic vector of film is modeled according to the metamessage of film.By the set of film in the present invention
It is expressed as T={ t1, t2..., tn, film tiCharacteristic vector be expressed asSpecifically modeled
Cheng Shi：For example, when certain star occurs in film, its corresponding element is 1；Otherwise it is 0.In particular cases, director is also in play
In serve as performer, then its corresponding attribute adds one again.
Film tiCharacteristic vectorIn weightsIt can be obtained by following formula：
Step 2：The real-time interest vector of user is obtained by the real-time score information and the movie features vector of user.
User's set expression is U={ u in the present invention1, u2..., un, user uiReal-time interest vector be expressed asWhereinRepresent user uiTo attribute pjFavorable rating, attribute pjStill represent film
Metamessage.
The real-time interest vector of user is described below in detail establishes process.
1) the scoring film collection of each user is obtained from rating matrix
2) characteristic vector that user's scoring film concentrates film is obtained from film-eigenvectors matrix
3) user uiReal-time interest vectorSubstantially it isIn each film characteristic vectorWeighted sum, power
It is exactly real-time fancy grade of the user to film again.The real-time interest vector formula of user is represented by：
WhereinRepresent attribute pjReverse features frequency,It is user uiTo film tjScoring,It is user ui
Average score,Represent film tjTo user uiTime-based weighting function.
Formula (2) is decomposed into the characteristic vector of userIn characteristic attribute pjWeightCalculation formula be：
WhereinRepresent characteristic attribute pjIn film tjWeight in characteristic vector,Represent vectorIn
Weight limit.
Step 3：Movie features vector is updated with reference to the user interest vector obtained in the step 2, generated
More rational movie features vector.Detailed process is as follows：
1) obtained from rating matrix to film tjUndue user is commented to collect
2) interest vector that scoring user concentrates user is obtained from user-interest vector matrix
3) film t is obtained with reference to the rating matrix of userjMore rational characteristic vector.It is film tjRenewal obtains
Characteristic vector in pjThe weight of feature, formula are as follows：
Wherein, uiIt is to film tjUndue user is commented,It is user uiInterest vector in feature pjWeight.Together
Sample,Effect is to make normalized.
Step 4：The real-time interest vector of user is updated by way of iteration.Grade form of the user to film
Understand favorable rating of the user to film.Same portion's film will obtain different scorings in the crowd of different hobbies.Such as
One romance movie score in the crowd for liking seeing romance movie is higher, and score may be compared with the crowd for liking seeing action movie
It is low.Therefore by the more newly-generated more accurate rational movie features vector of the step 3.For the interest vector model of user,
It can be updated again by the more rational movie features vector described in the step 3 by the step 2.Finally may be used
To obtain the iteration renewal process shown in accompanying drawing 2.
Step 5：The similarity matrix of user is established by the real-time interest vector of the user, and then obtains targeted customer's
Neighbor user, scored accordingly according to the score in predicting formula of the collaborative filtering based on user, be finally completed topN
Recommendation list.
In proposed algorithm based on the real-time interest vector of user, the similarity between user does not use classical collaborative filtering
In the similarity based on scoring, but use the similarity of the real-time interest vector of user obtained in the step 2.The present invention
In user's similarity calculated using cosine similarity formula.User uiWith user ujBetween similarity can be represented with following formula：
Wherein l is the length of the interest vector of user.WithCharacteristic attribute p is represented respectivelymIn user uiAnd ujIt is emerging
Weighted value in inclination amount.
For targeted customer ui, setRepresent user uiNeighbor user set, user niIt is target
User uiOne of neighbours.Represent user uiAverage score,Represent user uiTo film tjPrediction scoring can
Obtained by following formula：
To user uiThe prediction scoring of all films that do not score is ranked up, and is finally given N number of scoring highest film and is
Recommendation results, and consequently recommended result is shown to user.
Wherein, the movie features vector obtained in the step 1 obtains according to whole movie collection, and vector can relate to
And to substantial amounts of attribute, cause the characteristic vector dimension of film can be especially high.It is too high to algorithm computational efficiency in order to reduce dimension
Influence, the present invention carries out dimensionality reduction using SVD (Singular Value Decomposition) to the characteristic vector of film.
Time factor in the step 3There can be following formula to be calculated：
Above formula is a nonlinear function, and main prominent user comments the weight of undue film the nearest period.Wherein dnow
The current time is represented,Represent user uiTo film tjThe scoring time.β ∈ (0,1) are referred to as time factor affecting parameters, β
Bigger time factorTo be appraised point of time effects are bigger, i.e., user comments the weight of undue film bigger the nearest period,
The weight of scoring film was smaller in the past.According to different commending systems, recommendation results can be optimized by adjusting β value.
Below by embodiment, and with reference to accompanying drawing 1, technical scheme is described in further detail.
The MovieLens data sets provided herein using Univ Minnesota-Twin Cities USA's GroupLens research projects
(http://grouplens.org), the test set is frequent accepted standard data set in commending system research.Make herein
With ml-100k data set versions in the data set, 100000 of 1682 films are commented including 943 anonymous
Divide and score the time, time span is 7 months.Including cinematic genre information, and the director on film, playwright, screenwriter, act the leading role information
The film source address then provided according to data set crawls.
The present invention uses evaluation criterias of the mean absolute error MAE (Mean Absolute Error) as proposed algorithm.
Mean absolute error MAE weighs prediction by calculating the mean error between user's scoring of prediction and user's scoring of reality
The accuracy of scoring.MAE is smaller, and proposed algorithm quality is higher.
Wherein, N represents score in predicting quantity,WithUser u is represented respectivelyiTo film tjPrediction scoring and true
Real scoring.
1. according to the film source address provided in MovieLens data sets, crawl and the member letter such as directed, write a play, acted the leading role
Breath.And then by traveling through the metamessage of all films, the attribute library that dimension is n is established, the storehouse includes going out in data set film
The all properties now crossed.
For the foundation of certain movie features vector, the attribute in the metamessage of the film is traveled through, according in the step 1
Formula (1) obtain n dimension characteristic vector.1682 movie features vectors in final MovieLens have been built into one
1682*N movie features matrix M.Then dimension-reduction treatment is carried out to M using SVD matrix decompositions technology, it is special obtains final film
Levy vector ft。
2. with targeted customer uiExemplified by, establish the real-time interest vector of the user.U is obtained from rating matrixiScoring film
CollectionFilm concentrates the characteristic vector of film to be obtained by step 1.Time factor affecting parameters β is set
Appropriate value, user u is obtained according to formula (2) in the step 2iReal-time interest vector.
3. with film tjExemplified by, renewal obtains more rational movie features vector.Obtained from rating matrix to film tj
Undue user is commented to collectUser concentrates the characteristic vector of user to be obtained by step 2.According to described
Formula (4) obtains film t in step 3jMore rational characteristic vector.
4. setting iterations is c, according to accompanying drawing 2 to targeted customer uiReal-time interest vector be updated.
5. user's similarity matrix S can be obtained according to the formula (5) of the step 5.To targeted customer uiWith other use
The sequence of the similarity data at family, n user is as neighbor user before obtaining the similitude highest of targeted customer.Last basis
Formula (6) in the step 5 obtains prediction scoring of the targeted customer to the film that do not score, so as to obtain prediction scoring highest
Preceding k portions film recommend user ui。
When embodiment is set with iterations difference, the mean absolute error situation of the present invention is observed.Specific experiment knot
Fruit sees accompanying drawing 3.
Claims (5)
1. a kind of film personalized recommendation method based on the real-time interest vector of user, it is characterised in that this method includes following step
Suddenly：
Step 1：The characteristic attribute of every film is modeled according to the metamessage of film, obtains the characteristic vector of film, electricity
Shadow tiCharacteristic vector be expressed asWhereinObtained by following formula：I.e.
If the attribute p for representing film metamessage is included in the filmj, then weights corresponding to the attributeAdd 1, if in the film not
Include attribute pj, then weights corresponding to the attributeIt is zero, T={ t1,t2,…,tnRepresent film set；
Step 2：The real-time interest vector of user, user are obtained by the real-time score information and the movie features vector of user
Set expression is U={ u1,u2,…,un, user uiReal-time interest vector be expressed as WhereinRepresent user uiTo attribute pjFavorable rating, user uiReal-time interest vectorIt is by uiScored film feature to
AmountWeighted sum, weight is real-time fancy grade of the user to film, user uiReal-time interest vectorComputational methods
For：
<mrow>
<msub>
<mi>f</mi>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
</msub>
<mo>=</mo>
<mfenced open = "[" close = "]">
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>idf</mi>
<msub>
<mi>t</mi>
<mn>1</mn>
</msub>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>idf</mi>
<msub>
<mi>t</mi>
<mn>2</mn>
</msub>
</msub>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>...</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
<msub>
<mi>&Sigma;</mi>
<mrow>
<msub>
<mi>t</mi>
<mi>j</mi>
</msub>
<mo>&Element;</mo>
<msub>
<mi>T</mi>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
</msub>
</mrow>
</msub>
<msub>
<mi>f</mi>
<msub>
<mi>t</mi>
<mi>j</mi>
</msub>
</msub>
<mo>*</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>r</mi>
<mrow>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>t</mi>
<mi>j</mi>
</msub>
</mrow>
</msub>
<mo>-</mo>
<mover>
<msub>
<mi>r</mi>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
<mo>*</mo>
<msub>
<mi>WD</mi>
<mrow>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>t</mi>
<mi>j</mi>
</msub>
</mrow>
</msub>
</mrow>
WhereinRepresent attribute pjReverse features frequency, by total number of users mesh in system divided by like attribute pjUser
Number, then obtained business is taken the logarithm to obtain,It is each user's scoring film collection,It is user uiIt is right
Film tjScoring,It is user uiAverage score,Represent film tjTo user uiTime-based weight letter
Number；
Step 3：The score information of the user interest vector sum user obtained with reference to the step 2 is carried out to movie features vector
Renewal, the movie features vector after generation renewal；
Step 4：Iterative process is formed by the step 2 and the step 3, the real-time interest vector of user is updated,
The real-time interest vector of user after being updated；
Step 5：The similarity matrix of user is established by the real-time interest vector of user after the renewal, and then obtains target use
The neighbor user at family, corresponding scoring is beaten according to the score in predicting formula of the collaborative filtering based on user, is finally completed
TopN recommendation lists, wherein, user uiWith user ujBetween similarity can be represented with following formula：
<mrow>
<mi>s</mi>
<mi>i</mi>
<mi>m</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>u</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>m</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>l</mi>
</msubsup>
<msub>
<mi>w</mi>
<msub>
<mi>u</mi>
<msub>
<mi>i</mi>
<mi>m</mi>
</msub>
</msub>
</msub>
<mo>*</mo>
<msub>
<mi>w</mi>
<msub>
<mi>u</mi>
<msub>
<mi>j</mi>
<mi>m</mi>
</msub>
</msub>
</msub>
</mrow>
<mrow>
<msqrt>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>m</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>l</mi>
</msubsup>
<msup>
<msub>
<mi>w</mi>
<msub>
<mi>u</mi>
<msub>
<mi>i</mi>
<mi>m</mi>
</msub>
</msub>
</msub>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
<msqrt>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>m</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>l</mi>
</msubsup>
<msup>
<msub>
<mi>w</mi>
<msub>
<mi>u</mi>
<msub>
<mi>i</mi>
<mi>m</mi>
</msub>
</msub>
</msub>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
</mrow>
</mfrac>
</mrow>
Wherein l is the length of the interest vector of user,WithCharacteristic attribute p is represented respectivelymIn user uiAnd ujInterest to
Weighted value in amount；Score in predicting formula is：
Wherein gatherRepresent user uiNeighbor user set, user niIt is targeted customer uiTherein one
Individual neighbours.
2. the film personalized recommendation method according to claim 1 based on the real-time interest vector of user, it is characterised in that：
In step 1, dimensionality reduction is carried out to the characteristic vector of film using singular value decomposition method.
3. the film personalized recommendation method according to claim 1 based on the real-time interest vector of user, it is characterised in that：
User u in step 2iReal-time interest vectorCalculating can be analyzed to the real-time interest vector of userMiddle attribute pjPower
WeightCalculating, formula is：
<mrow>
<msub>
<mi>w</mi>
<msub>
<mi>u</mi>
<msub>
<mi>i</mi>
<mi>j</mi>
</msub>
</msub>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<mi>a</mi>
<mi>l</mi>
<mi>l</mi>
<mi>U</mi>
<mi>s</mi>
<mi>e</mi>
<mi>r</mi>
<mi>N</mi>
<mi>u</mi>
<mi>m</mi>
<mi>b</mi>
<mi>e</mi>
<mi>r</mi>
</mrow>
<mrow>
<mi>p</mi>
<mi>r</mi>
<mi>e</mi>
<mi>f</mi>
<mi>e</mi>
<mi>r</mi>
<mi>e</mi>
<mi>d</mi>
<mi>U</mi>
<mi>s</mi>
<mi>e</mi>
<mi>r</mi>
<mi>N</mi>
<mi>u</mi>
<mi>m</mi>
<mi>b</mi>
<mi>e</mi>
<mi>r</mi>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<msub>
<mi>t</mi>
<mi>j</mi>
</msub>
<mo>&Element;</mo>
<msub>
<mi>T</mi>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
</msub>
</mrow>
</msub>
<msub>
<mi>w</mi>
<msub>
<mi>t</mi>
<msub>
<mi>i</mi>
<mi>j</mi>
</msub>
</msub>
</msub>
<mo>*</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>r</mi>
<mrow>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>t</mi>
<mi>j</mi>
</msub>
</mrow>
</msub>
<mo>-</mo>
<mover>
<msub>
<mi>r</mi>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
<mo>*</mo>
<msub>
<mi>WD</mi>
<mrow>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>t</mi>
<mi>j</mi>
</msub>
</mrow>
</msub>
</mrow>
<mrow>
<mi>M</mi>
<mi>a</mi>
<mi>x</mi>
<mrow>
<mo>(</mo>
<mo>|</mo>
<msub>
<mi>w</mi>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
</msub>
<mo>|</mo>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
WhereinRepresent attribute pjIn film tjWeight in characteristic vector,Represent vectorMiddle weight limit,
AllUserNumber is total number of users mesh in system, and preferedUserNumber is to like attribute pjUser number；
In above-mentioned formulaIt is film tjTo user uiTime-based weighting function, scoring the time from currently nearer
Weight is bigger, otherwise weight is smaller, is represented with below equation：
<mrow>
<msub>
<mi>WD</mi>
<mrow>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>t</mi>
<mi>j</mi>
</msub>
</mrow>
</msub>
<mo>=</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<mi>&beta;</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>&beta;</mi>
<mo>*</mo>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>d</mi>
<mrow>
<mi>n</mi>
<mi>o</mi>
<mi>w</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>d</mi>
<mrow>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>t</mi>
<mi>j</mi>
</msub>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</msup>
</mrow>
Wherein dnowThe current time is represented,Represent user uiTo film tjThe scoring time, β ∈ (0,1) be referred to as the time because
Sub- affecting parameters, the bigger time factors of βTo be appraised point of time effects are bigger, i.e., user comments undue film the nearest period
Weight it is bigger, in the past score film weight it is smaller, according to different commending systems, can be pushed away by adjusting β value to optimize
Recommend result.
4. the film personalized recommendation method according to claim 1 based on the real-time interest vector of user, it is characterised in that：
Movie features vector after being updated described in step 3 can be obtained by following step：
A) obtained from rating matrix to film tjUndue user is commented to collect
B) interest vector that scoring user concentrates user is obtained in the matrix formed from user interest vector
C) the rating matrix more New cinema t of user is combinedjCharacteristic vector,It is film tjUpdate p in obtained characteristic vectorj
The weight of feature, formula are as follows：
<mrow>
<msubsup>
<mi>w</mi>
<msub>
<mi>t</mi>
<msub>
<mi>i</mi>
<mi>j</mi>
</msub>
</msub>
<mo>&prime;</mo>
</msubsup>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
<mo>&Element;</mo>
<msub>
<mi>U</mi>
<msub>
<mi>t</mi>
<mi>j</mi>
</msub>
</msub>
</mrow>
</msub>
<msub>
<mi>w</mi>
<msub>
<mi>u</mi>
<msub>
<mi>i</mi>
<mi>j</mi>
</msub>
</msub>
</msub>
<mo>*</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>r</mi>
<mrow>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>t</mi>
<mi>j</mi>
</msub>
</mrow>
</msub>
<mo>-</mo>
<mover>
<msub>
<mi>r</mi>
<msub>
<mi>u</mi>
<mi>i</mi>
</msub>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>M</mi>
<mi>a</mi>
<mi>x</mi>
<mrow>
<mo>(</mo>
<mo>|</mo>
<msub>
<mi>w</mi>
<msub>
<mi>t</mi>
<mi>j</mi>
</msub>
</msub>
<mo>|</mo>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
Wherein, uiIt is to film tjUndue user is commented,It is user uiInterest vector in feature pjWeight,Effect is to make normalized.
5. the film personalized recommendation method according to claim 1 based on the real-time interest vector of user, it is characterised in that：
The iterations of iterative process described in step 4 is adjusted according to different commending systems, to optimize recommendation results.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410313446.XA CN104063481B (en) | 2014-07-02 | 2014-07-02 | A kind of film personalized recommendation method based on the real-time interest vector of user |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410313446.XA CN104063481B (en) | 2014-07-02 | 2014-07-02 | A kind of film personalized recommendation method based on the real-time interest vector of user |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104063481A CN104063481A (en) | 2014-09-24 |
| CN104063481B true CN104063481B (en) | 2017-11-14 |
Family
ID=51551195
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410313446.XA Expired - Fee Related CN104063481B (en) | 2014-07-02 | 2014-07-02 | A kind of film personalized recommendation method based on the real-time interest vector of user |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104063481B (en) |
Families Citing this family (36)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104281956B (en) * | 2014-10-27 | 2018-09-07 | 南京信息工程大学 | The dynamic recommendation method for adapting to user interest variation based on temporal information |
| CN104281718B (en) * | 2014-11-04 | 2018-03-02 | 深圳市英威诺科技有限公司 | A kind of method that intelligent recommendation is excavated based on user group's behavioral data |
| CN104834969A (en) * | 2015-05-05 | 2015-08-12 | 东南大学 | Film evaluation prediction method and system |
| CN104935963B (en) * | 2015-05-29 | 2018-03-16 | 中国科学院信息工程研究所 | A kind of video recommendation method based on timing driving |
| CN104965889B (en) * | 2015-06-17 | 2017-06-13 | 腾讯科技（深圳）有限公司 | Content recommendation method and device |
| CN105205130A (en) * | 2015-09-15 | 2015-12-30 | 广东工业大学 | Method of improving accuracy of recommendation system |
| CN106570008B (en) * | 2015-10-09 | 2020-03-27 | 阿里巴巴集团控股有限公司 | Recommendation method and device |
| CN105338408B (en) * | 2015-12-02 | 2018-11-13 | 南京理工大学 | Video recommendation method based on time factor |
| CN105868334B (en) * | 2016-03-28 | 2020-10-30 | 云南财经大学 | Feature incremental type-based personalized movie recommendation method and system |
| CN105930425A (en) * | 2016-04-18 | 2016-09-07 | 乐视控股（北京）有限公司 | Personalized video recommendation method and apparatus |
| CN106095974B (en) * | 2016-06-20 | 2019-12-24 | 上海理工大学 | Recommendation system score prediction and recommendation algorithm based on network structure similarity |
| CN106372235A (en) * | 2016-09-12 | 2017-02-01 | 中国联合网络通信集团有限公司 | Movie recommendation method and system |
| CN106484777B (en) * | 2016-09-12 | 2020-09-08 | 腾讯科技（深圳）有限公司 | Multimedia data processing method and device |
| CN106649540B (en) * | 2016-10-26 | 2022-04-01 | Tcl科技集团股份有限公司 | Video recommendation method and system |
| CN106709037B (en) * | 2016-12-29 | 2019-11-08 | 武汉大学 | A kind of film recommended method based on Heterogeneous Information network |
| CN106951462B (en) * | 2017-02-27 | 2021-06-25 | 四川大学 | Movie recommendation method based on Time-Trust similarity |
| CN108932645A (en) * | 2017-05-26 | 2018-12-04 | 车伯乐（北京）信息科技有限公司 | A kind of user selects vehicle system |
| CN108932644A (en) * | 2017-05-26 | 2018-12-04 | 车伯乐（北京）信息科技有限公司 | A kind of user selects vehicle method, apparatus, equipment and computer-readable medium |
| CN107368519A (en) * | 2017-06-05 | 2017-11-21 | 桂林电子科技大学 | A kind of cooperative processing method and system for agreeing with user interest change |
| CN107368540A (en) * | 2017-06-26 | 2017-11-21 | 北京理工大学 | The film that multi-model based on user's self-similarity is combined recommends method |
| CN108959332B (en) * | 2017-08-03 | 2021-09-14 | 中国人民解放军国防科学技术大学 | Content recommendation method and system using high-dimensional auxiliary information correction |
| CN110019943B (en) * | 2017-09-11 | 2021-09-14 | 中国移动通信集团浙江有限公司 | Video recommendation method and device, electronic equipment and storage medium |
| CN109978580A (en) * | 2017-12-28 | 2019-07-05 | 北京京东尚科信息技术有限公司 | Object recommendation method, apparatus and computer readable storage medium |
| CN108415987B (en) * | 2018-02-12 | 2021-05-11 | 大连理工大学 | Cold start solution for movie recommendation |
| CN108460145B (en) * | 2018-03-15 | 2020-07-03 | 北京邮电大学 | Collaborative filtering recommendation method based on mixed interest similarity |
| CN108615177B (en) * | 2018-04-09 | 2021-09-03 | 武汉理工大学 | Electronic terminal personalized recommendation method based on weighting extraction interestingness |
| CN108763362B (en) * | 2018-05-17 | 2020-10-09 | 浙江工业大学 | Local model weighted fusion Top-N movie recommendation method based on random anchor point pair selection |
| CN108804683B (en) * | 2018-06-13 | 2021-11-23 | 重庆理工大学 | Movie recommendation method combining matrix decomposition and collaborative filtering algorithm |
| CN110737822B (en) * | 2018-07-03 | 2022-07-26 | 百度在线网络技术（北京）有限公司 | User interest mining method, device, equipment and storage medium |
| CN109299992A (en) * | 2018-07-16 | 2019-02-01 | 深圳市谷熊网络科技有限公司 | A kind of method, apparatus and terminal to user's recommendation information |
| CN109241203B (en) * | 2018-09-27 | 2021-08-31 | 天津理工大学 | Clustering method for user preference and distance weighting by fusing time factors |
| CN109710835B (en) * | 2018-11-15 | 2020-12-29 | 中国人民解放军国防科技大学 | Heterogeneous information network recommendation method with time weight |
| CN109740064B (en) * | 2019-01-18 | 2021-08-17 | 北京化工大学 | CF recommendation method for fusing matrix decomposition and mining user project information |
| CN109933726B (en) * | 2019-03-22 | 2022-04-12 | 江西理工大学 | Collaborative filtering movie recommendation method based on user average weighted interest vector clustering |
| CN110532330B (en) * | 2019-09-03 | 2022-06-03 | 四川长虹电器股份有限公司 | Hive-based collaborative filtering recommendation method |
| CN117235366B (en) * | 2023-09-19 | 2024-06-18 | 北京学说科技有限公司 | Collaborative recommendation method and system based on content relevance |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2002010954A2 (en) * | 2000-07-27 | 2002-02-07 | Polygnostics Limited | Collaborative filtering |
| CN102495864A (en) * | 2011-11-25 | 2012-06-13 | 清华大学 | Collaborative filtering recommending method and system based on grading |
| CN103761237A (en) * | 2013-12-04 | 2014-04-30 | 南京邮电大学 | Collaborative filtering recommending method based on characteristics and credibility of users |
-
2014
- 2014-07-02 CN CN201410313446.XA patent/CN104063481B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2002010954A2 (en) * | 2000-07-27 | 2002-02-07 | Polygnostics Limited | Collaborative filtering |
| CN102495864A (en) * | 2011-11-25 | 2012-06-13 | 清华大学 | Collaborative filtering recommending method and system based on grading |
| CN103761237A (en) * | 2013-12-04 | 2014-04-30 | 南京邮电大学 | Collaborative filtering recommending method based on characteristics and credibility of users |
Non-Patent Citations (1)
| Title |
|---|
| 个性化推荐系统的研究进展;刘建国，周涛，汪秉宏;《自然科学进展》;20090131;第19卷(第1期);1-15 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104063481A (en) | 2014-09-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104063481B (en) | A kind of film personalized recommendation method based on the real-time interest vector of user | |
| Rafailidis et al. | Modeling users preference dynamics and side information in recommender systems | |
| CN104935963B (en) | A kind of video recommendation method based on timing driving | |
| CN102332006B (en) | A kind of information push control method and device | |
| CN108334592B (en) | Personalized recommendation method based on combination of content and collaborative filtering | |
| CN109241203B (en) | Clustering method for user preference and distance weighting by fusing time factors | |
| CN110162693A (en) | A kind of method and server of information recommendation | |
| CN106682114A (en) | Personalized recommending method fused with user trust relationships and comment information | |
| CN106294758A (en) | Collaborative recommendation method based on the change of user cognition degree | |
| CN107220365A (en) | Accurate commending system and method based on collaborative filtering and correlation rule parallel processing | |
| CN109190030B (en) | Implicit feedback recommendation method fusing node2vec and deep neural network | |
| CN104462383A (en) | Movie recommendation method based on feedback of users' various behaviors | |
| Zhou et al. | A personalized recommendation algorithm based on approximating the singular value decomposition (ApproSVD) | |
| CN106846029B (en) | Collaborative filtering recommendation algorithm based on genetic algorithm and novel similarity calculation strategy | |
| Wu et al. | Optimization matrix factorization recommendation algorithm based on rating centrality | |
| CN104008204A (en) | Dynamic multi-dimensional context awareness film recommending system and achieving method thereof | |
| CN110059257B (en) | Project recommendation method based on score correction | |
| Bai et al. | Customer loyalty improves the effectiveness of recommender systems based on complex network | |
| Cao et al. | A Recommendation Approach Based on Product Attribute Reviews: Improved Collaborative Filtering Considering the Sentiment Polarity. | |
| Wang et al. | Research on BP neural network recommendation model fusing user reviews and ratings | |
| Samundeeswary et al. | Comparative study of recommender systems built using various methods of collaborative filtering algorithm | |
| CN108959506A (en) | A kind of resource recommendation method based on group interest | |
| CN108763515B (en) | Time-sensitive personalized recommendation method based on probability matrix decomposition | |
| CN106650972B (en) | Recommendation system score prediction method based on cloud model and oriented to social network | |
| Li et al. | Collaborative filtering recommendation algorithm based on user characteristics and user interests |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20171114 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |