CN106874355A - The collaborative filtering method of social networks and user's similarity is incorporated simultaneously - Google Patents
The collaborative filtering method of social networks and user's similarity is incorporated simultaneously Download PDFInfo
- Publication number
- CN106874355A CN106874355A CN201611237097.3A CN201611237097A CN106874355A CN 106874355 A CN106874355 A CN 106874355A CN 201611237097 A CN201611237097 A CN 201611237097A CN 106874355 A CN106874355 A CN 106874355A
- Authority
- CN
- China
- Prior art keywords
- user
- similarity
- matrix
- article
- characteristic 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000001914 filtration Methods 0.000 title claims abstract description 24
- 239000011159 matrix material Substances 0.000 claims abstract description 131
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 26
- 238000010606 normalization Methods 0.000 claims abstract description 20
- 238000005457 optimization Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 11
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 2
- 238000011478 gradient descent method Methods 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 229940050561 matrix product Drugs 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012552 review Methods 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
- G06F16/60—Information retrieval; Database structures therefor; File system structures therefor of audio data
- G06F16/63—Querying
- G06F16/635—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/73—Querying
- G06F16/735—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Abstract
It is an object of the invention to solve the problems of prior art, find a kind of while incorporating the collaborative filtering method of good friend's feature and similar users feature, improve the degree of accuracy of collaborative filtering method, including:Consumer articles rating matrix is obtained, the similarity between user is calculated by Pearson correlation coefficient, set up user's similarity matrix;User's customer relationship matrix is obtained, original user customer relationship matrix is normalized, the user's customer relationship matrix after being normalized;Similar users feature and good friend's user characteristics are incorporated the user characteristics of probability matrix decomposition model according to the user's customer relationship matrix after user's similarity matrix and normalization, prediction scoring of the user to article is obtained according to probability matrix decomposition model, according to prediction scoring for user recommends article.Advantageous Effects:So that recommendation results keep friendly user characteristics and the influence of similar users feature well, the degree of accuracy of collaborative filtering method is improved.
Description
Technical field
It is the present invention relates to collaborative filtering method more particularly to a kind of while incorporating the collaboration of social networks and user's similarity
Filter method.
Background technology
In recent years, with internet product problem of information overload it is more serious, all in the urgent need to carrying in many products
For the function of personalized recommendation.However, traditional recommended technology only considered two kinds of entities i.e. " user " and " article ", and ignore
The influence of social networks between good friend to recommendation results.Therefore, the commending system for incorporating social networks gradually attracts attention,
Current scientific research and the engineering practice work for having had many proves that the introducing of social networks can effectively improve the standard of commending system
True property and and personalization level.
At present, the proposed algorithm for incorporating the recommendation method use of social networks is largely divided into two major classes:(1) based on internal memory
Personalized social recommendation algorithm:Social networks are dissolved among traditional proposed algorithm based on internal memory, such as based on use
The arest neighbors proposed algorithm at family or the arest neighbors proposed algorithm based on article add social networks.(2) individual character based on model
Change social algorithm:Social networks are dissolved among traditional proposed algorithm based on model, such as decompose mould in probability matrix
Social networks are added in type.
Above-mentioned two class incorporates the recommendation method of social networks, can only respectively incorporate social networks or user's similarity letter
Breath, it is impossible to while being modeled to social networks and user's similarity information, and among real internet product application scenarios, use
On the one hand the interest preference at family can be influenceed by good friend, i.e. the influence of social networks.On the other hand its interest be intended to phase again
With the user of interest preference, i.e., influenceed by similarity user high.So patent of the present invention proposes that one kind incorporates social activity simultaneously
The collaborative filtering method and system of relation and user's similarity, while social networks and user's similarity information are dissolved into recommendation
Among algorithm, the degree of accuracy of collaborative filtering method is improved.
The content of the invention
It is an object of the invention to solve the problems of prior art, one kind is found while incorporating social networks and use
The collaborative filtering method of family similarity, improves the degree of accuracy of collaborative filtering method.
In order to realize the purpose, the present invention incorporates the collaborative filtering method of social networks and user's similarity simultaneously, together
When incorporate the collaborative filtering method of social networks and user's similarity, including following steps:
Step one:User-article rating matrix is obtained, the similarity between user is calculated by Pearson correlation coefficient,
Set up user's similarity matrix;
Step 2:User-user relational matrix is obtained, original user-customer relationship matrix is normalized, obtained
User-user relational matrix after to normalization;
Step 3:According to the user-user relational matrix after user's similarity matrix and normalization by similar users feature
In incorporating the prediction scoring formula of probability matrix decomposition model with good friend's user characteristics, according to incorporating similar users feature and good friend
The probability matrix decomposition model of user characteristics obtains prediction scoring of the user to article, according to prediction scoring for user recommends thing
Product.
Further, in step one, the computing formula of similarity is between user:
Wherein, SiwIt is the similarity of user i and article w, I is the public scoring set of user i and user w, RijIt is user
Scorings of the i to article j,It is the scoring average of user i, RwjScoring for user w to article j,For the scoring of user w is equal
Value.
Further, in step 2, it is to the process that original user-customer relationship matrix is normalized:By normalizing
Change so thatFiRepresent good friend's set of user i, TivIt is the i-th row v in original user-customer relationship matrix
Relations of the element representation user i of row to user v.
Further, in step one, similarity threshold is set, selects high with user's similarity according to user's similarity matrix
Gather as similarity user high in user's set of similarity threshold.
Further, in step one, the threshold value Y of similarity number of users high is set, is selected according to user's similarity matrix
Gather as similarity user high with Y user of user's similarity highest.
Further, Y value is equal to good friend's quantity of user.
Further, in incorporating the probability matrix decomposition model of similar users feature and good friend's user characteristics:
The conditional probability of the hidden factor matrix U of user is:
The conditional probability of the hidden factor matrix V of article is:
Score in predicting formula is:
Wherein VjIt is the characteristic vector of article j, UiIt is the characteristic vector of user i, UυIt is the characteristic vector of user υ, UwIt is use
The characteristic vector of family w, VjIt is the characteristic vector of article j, FiFor the good friend of user i gathers, NiIt is the similarity user high of user i
Set, SiwRepresent the similarity between user i and user w, TivIt is the relation value after the normalization of user i and article w,
It is the inner product of the characteristic vector of the characteristic vector and article j of user i,For the characteristic vector of user w and the feature of article j to
The inner product of amount,It is the inner product of the characteristic vector of the characteristic vector and article j of user υ, λTAnd λSIt is adjustable weight parameter, 0
< λTThe < λ of < 1,0S< 1, and λT+λS≤1.Because score in predicting formula is the weighting of three parts, λTOr λSSize control
The weights influence of which part is larger in last score in predicting formula, such as λTIt is good in that explanation score in predicting formula than larger
Friend's scoring influence is larger;λSIt is larger, then illustrate that similarity user high is larger on scoring influence.
Further, user is obtained according to the probability matrix decomposition model for incorporating similar users feature and good friend's user characteristics
The step of prediction to article is scored includes:
Conditional probability according to prediction scoring formula, the conditional probability of the hidden factor matrix of user and the hidden factor matrix of article is carried
Go out the posterior probability for needing maximized hidden factor matrix:
Above-mentioned probability function is maximized, is equivalent to minimize following loss object function,
Wherein, U is the hidden factor matrix of user, and V is the hidden factor matrix of article, and R is user-article rating matrix, and T is normalizing
User-user relational matrix after change, S is user's similarity matrix, RijScoring for user i to article j, UiIt is user i's
Characteristic vector, UυIt is the characteristic vector of user υ, UwIt is the characteristic vector of user w, VjIt is the characteristic vector of article j, FiIt is user i
Good friend set, NiFor the similarity user high of user i gathers, SiwIt is the relation value after the normalization of user i and article w,It is the inner product of the characteristic vector of the characteristic vector and article j of user i,It is the characteristic vector and article j of user w
The inner product of characteristic vector,It is the inner product of the characteristic vector of the characteristic vector and article j of user υ,Represent Frobenius
Norm;
Using stochastic gradient descent method optimization loss object function, obtain the hidden factor matrix U of user and article it is hidden because
Submatrix V, and usePrediction scoring.
Further, the step of according to prediction scoring for user's recommendation article, includes:Scored according to prediction, to preceding pre- test and appraisal
Article N high is recommended before point.
Following Advantageous Effects can be obtained by implementing the present invention:Similar users feature and good friend's user characteristics are melted
Enter the prediction scoring formula of probability matrix decomposition model, according to the probability matrix for incorporating similar users feature and good friend's user characteristics
Decomposition model obtains prediction scoring of the user to article, according to prediction scoring for user recommends article, probability matrix decomposition model
On the basis of introduce social networks and user's similarity information, and then improve probability matrix decomposition model, and acquisition preferably push away
Recommend effect.
Brief description of the drawings
Fig. 1 is the flow chart of the collaborative filtering method for incorporating social networks and user's similarity;
Fig. 2 is canonical matrix decomposition model;
Fig. 3 is while incorporating the probability matrix decomposition model of social networks and user's similarity;
Fig. 4 is while incorporating the block diagram of the collaborative filtering system of social networks and user's similarity.
Specific embodiment
For the ease of the understanding of those skilled in the art, the present invention is made further with reference to specific embodiment
It is bright:
As shown in figure 1, the present invention incorporates the collaborative filtering method of good friend's feature and similar users feature simultaneously, including it is following
Several steps:
Step one:User-article rating matrix is obtained, the similarity between user is calculated by Pearson correlation coefficient,
Set up user's similarity matrix.Similarity user intersection high is obtained by user's similarity matrix.
User's similarity matrix sets up process:Obtain user-article rating matrixRijFor i-th in matrix R
OK, the scorings of the element representation user i to article j of jth row, general user's article score value standard be 5 points of systems, ten point system,
100 points system etc., by Pearson correlation coefficient calculate user between similarity, set up user similarity matrix S, computing formula
For:
Wherein, SiwIt is used to represent the similarity of user i and article w for the element of i rows w row in matrix S, I is user i and use
The public scoring set of family w, RijScoring for user i to article j,It is the scoring average of user i, RwjIt is user w to article j
Scoring,It is the scoring average of user w.
Similarity user set high can typically be obtained using three kinds of modes:
First method, sets similarity threshold, and similarity is similarity user high set higher than all users of threshold value,
When such as threshold value is 0.5, all users of the similarity higher than 0.5 pick out and do similarity user set high.
Second method, sets the threshold value of similarity number of users high, such as when threshold value is 10, select similarity highest
10 users do similarity user set high.
The third method, the relation quantity of correspondence user good friend, such as one user has 20 good friends, then we are corresponding
, choose 20 similarity users high and gather as similarity user high, then just can be with the fair ratio of the set of friend relation
Compared with.
Step 2:Original user-customer relationship matrix is normalized, the user-user relation after being normalized
Matrix;
Normalize user-user relational matrix detailed process of setting up be:Use original user-customer relationship matrixHere the i-th row in matrix T, the element T of v rowivRepresent relations of the user i to user v, general most common user
Relation score value standard is 01 binary numerical value, i.e., T when there is friend relation between useriv=1, otherwise Tiv=0, by normalizing
Change so thatWherein FiRepresent good friend's set of user i.
By taking the user-user relational matrix that score value standard is 01 binary numerical value as an example, the user-user before normalization
Relational matrix is as follows:
Normalized process is by customer relationship matrix, numerical value per a line and to become 1, and the matrix after normalization is such as
Under.
FiRepresent good friend's set of user i, in above-mentioned table, good friend's intersection of user i be expert at by user i in it is non-zero
Numerical value answers the user in row.Good friend's intersection such as user 5 is user 1, user 2 and user 3, and user 5 and the good friend of user 1
Coefficient values are 1/3, and user 5 is 1/3 with the friend relation numerical value of user 1, and user 5 is 1/3 with the friend relation numerical value of user 1, its
Be 1.
Step 3:According to the user-user relational matrix after user's similarity matrix and normalization by similar users feature
In incorporating the prediction scoring formula of probability matrix decomposition model with good friend's user characteristics, according to incorporating similar users feature and good friend
The probability matrix decomposition model of user characteristics obtains prediction scoring of the user to article, according to prediction scoring for user recommends thing
Product.
Its step includes:
1) incorporate social networks and user's similarity information simultaneously on the basis of probability matrix decomposition model, propose a kind of
New score in predicting formula, formation incorporates the probability matrix decomposition model of social networks and user's similarity.
Such as Fig. 2, in the probability matrix decomposition model of standard, the conditional probability of the hidden factor matrix U of user isThe hidden ratio characteristics vector of i-th user is Ui, the condition of the hidden factor matrix of article is general
Rate isThe hidden ratio characteristics vector of j-th article is Vj, scorings of the user i to article j
Predictor formula is the inner product of the hidden factor vector of user and the hidden factor vector of article Represent UiClothes
From average for 0 variance isMultiple normal distribution,Represent VjAverage is obeyed for 0 variance is's
Multiple normal distribution, m represents the line number of homography, and n represents the columns of homography, similarly, same in the follow-up formula of the application
The meaning of type formula is not being elaborated.
Such as Fig. 3, in incorporating the probability matrix decomposition model of user characteristics and similar users feature at the same time, user it is hidden because
The conditional probability of submatrix isThe conditional probability of the hidden factor matrix of article isAnd user i is to the score in predicting of article jComputing formula be:
Wherein σV、σUIt is variance, VjIt is the characteristic vector of article j, UiIt is the characteristic vector of user i, UυIt is the feature of user υ
Vector, UwIt is the characteristic vector of user w, I is the public scoring set of user i and user w, FiFor the good friend of user i gathers, Ni
For the similarity user high of user i gathers, SiwRepresent the similarity between user i and user w, TivIt is user i's and article w
Relation value after normalization,It is the inner product of the characteristic vector of the characteristic vector and article j of user i,It is the spy of user w
The inner product of vector and the characteristic vector of article j is levied,It is the inner product of the characteristic vector of the characteristic vector and article j of user υ, λT
And λSIt is adjustable weight parameter, 0 < λTThe < λ of < 1,0S< 1, and λT+λS≤1.As preferred, λT+λS≥0.5
(2) the hidden factor of the score in predicting formula, the conditional probability of the hidden factor matrix of user and article proposed according to step 1
The conditional probability of matrix, proposes to need the posterior probability of maximized hidden factor matrix as follows:
Above-mentioned probability function is maximized, is equivalent to minimize following loss object function,
Wherein U is the hidden factor matrix of user, and V is the hidden factor matrix of article, and R is user-article rating matrix, and T is normalizing
User-user relational matrix after change, S is user's similarity matrix, σ, σV、σU、σSIt is variance, RijIt is user i to article j's
Scoring, λU, λV, λTAnd λSIt is adjustable weight parameter, UiIt is the characteristic vector of user i, UυIt is the characteristic vector of user υ, UwIt is use
The characteristic vector of family w, VjThe public scoring set of user i and user w, F are represented for the characteristic vector I of article jiIt is good for user i
Friend's set, NiFor the similarity user high of user i gathers, SiwIt is the relation value after the normalization of user i and article w.For
The inner product of the characteristic vector of user i and the characteristic vector of article j,It is the characteristic vector and the characteristic vector of article j of user w
Inner product,It is the inner product of the characteristic vector of the characteristic vector and article j of user υ,Represent Frobenius norms, mesh
Scalar functions show user characteristics while being influenceed by its good friend's feature and similar users feature, IijIndicator function is represented, if with
Family i is commented too with to article j, and its value is equal to 1, is otherwise 0, in order to prevent over-fitting.Subscript T represents the Inner of its both sides matrix
Product.
(3) using stochastic gradient descent method minimize step 2) in optimization object function, obtain the hidden factor of user
Matrix U and the hidden factor matrix V of article, and useUnknown scoring is predicted, and then is each user-customized recommended
Top-N article.
In the present invention, without front and rear logical relation before step one and step 2, that is to say, that both can first carry out step
One, it is also possible to first carry out step 2.
As shown in figure 4, the present invention have also been devised a kind of use while incorporating the collaboration of good friend's feature and similar users feature
The collaborative filtering system of filter method, including:
User's similarity matrix sets up module:User-article rating matrix is obtained, is calculated by Pearson correlation coefficient and used
Similarity between family, sets up user's similarity matrix;
User-user relational matrix normalizes module:User-user relational matrix is obtained, to original user-customer relationship
Matrix is normalized, the user-user relational matrix after being normalized;
Article recommending module:According to the user-user relational matrix after user's similarity matrix and normalization by similar use
Family feature and good friend's user characteristics incorporate the user characteristics of probability matrix decomposition model, according to incorporating similar users feature and good friend
The probability matrix decomposition model of user characteristics obtains prediction scoring of the user to article, according to prediction scoring for user recommends thing
Product.
It is described further so that film is recommended and music is recommended as an example below:
Film is recommended
1) user's similarity matrix is set up:From original user-film score data, calculated by Pearson correlation coefficient
Similarity between user, sets up user's similarity matrix;Similarity between user is calculated by Pearson correlation coefficient, is set up
User similarity matrix S, Pearson correlation coefficient computing formula is
WhereinIt is scoring averages of the user i to film,It is the scoring average to film of user w, set I represents user i and user
The common electrical film review of w point set.
2) normalized user-user relational matrix is set up:Use original user-customer relationship matrixHere
I-th row in matrix T, the element T of v rowivRelations of the user i to user v is represented, by normalization so that
Wherein FiRepresent good friend's set of user i.
3) personalized recommendation generation:After the normalization that similarity information and step 2 are obtained between the user that step 1 is obtained
User's friend relation, while being dissolved among user characteristics, further improves normal probability matrix decomposition model, by boarding steps
Degree declines the object function for solving and being proposed in the present invention, is top-N personalized film recommendation results of user's generation.
Music is recommended
1) user's similarity matrix is set up:From original user-music score data, Pierre is crossed all
Inferior coefficient correlation calculates the similarity between user, sets up user's similarity matrix;It is related by Pearson came
Coefficient calculates the similarity between user, sets up user similarity matrix S, and Pearson correlation coefficient computing formula isWhereinIt is that user i is commented music
Divide average,It is the scoring average to music of user w, set I represents the common music scoring set of user i and user w.
2) normalized user-user relational matrix is set up:Use original user-customer relationship matrixHere
I-th row in matrix T, the element T of v rowivRelations of the user i to user v is represented, by normalization so that
Wherein FiRepresent good friend's set of user i.
3) personalized recommendation generation:After the normalization that the similar users characteristic information and step 2 that step 1 is obtained are obtained
User's friend relation, while being dissolved among user characteristics, further improves normal probability matrix decomposition model, by boarding steps
Degree declines the object function for solving and being proposed in the present invention, is top-N personalized music recommendation results of user's generation.
Specific embodiment of the invention is the foregoing is only, but technical characteristic of the invention is not limited thereto, Ren Heben
The technical staff in field in the field of the invention, all cover among the scope of the claims of the invention by the change or modification made.
Claims (9)
1. the collaborative filtering method of social networks and user's similarity is incorporated simultaneously, it is characterised in that including following steps:
Step one:User-article rating matrix is obtained, the similarity between user is calculated by Pearson correlation coefficient, set up
User's similarity matrix;
Step 2:User-user relational matrix is obtained, original user-customer relationship matrix is normalized, returned
User-user relational matrix after one change;
Step 3:Similar users feature is become reconciled according to the user-user relational matrix after user's similarity matrix and normalization
Friendly user characteristics is incorporated in the prediction scoring formula of probability matrix decomposition model, according to incorporating similar users feature and good friend user
The probability matrix decomposition model of feature obtains prediction scoring of the user to article, according to prediction scoring for user recommends article.
2. the collaborative filtering method of social networks and user's similarity is incorporated simultaneously as claimed in claim 1, it is characterised in that
In step one, the computing formula of similarity is between user:
Wherein, SiwIt is the similarity of user i and article w, I is the public scoring set of user i and user w, RijIt is user i to thing
The scoring of product j,It is the scoring average of user i, RwjScoring for user w to article j,It is the scoring average of user w.
3. the collaborative filtering method of social networks and user's similarity is incorporated simultaneously as claimed in claim 1, it is characterised in that
In step 2, it is to the process that original user-customer relationship matrix is normalized:By normalization so thatFiRepresent good friend's set of user i, TivIt is the unit of the i-th row v row in original user-customer relationship matrix
Element represents relations of the user i to user v.
4. the collaborative filtering method of social networks and user's similarity is incorporated simultaneously as claimed in claim 1, it is characterised in that
In step one, similarity threshold is set, the use according to the selection of user's similarity matrix with user's similarity higher than similarity threshold
Family set is gathered as similarity user high.
5. the collaborative filtering method of social networks and user's similarity is incorporated simultaneously as claimed in claim 1, it is characterised in that
In step one, the threshold value Y of similarity number of users high is set, according to the selection of user's similarity matrix and user's similarity highest
Y user gathers as similarity user high.
6. the collaborative filtering method of social networks and user's similarity is incorporated simultaneously as claimed in claim 5, it is characterised in that
Y value is equal to good friend's quantity of user.
7. the collaboration of good friend's user characteristics and similar users feature is incorporated while as described in any one in claim 1~6
Filter method, it is characterised in that in incorporating the probability matrix decomposition model of similar users feature and good friend's user characteristics:
The conditional probability of the hidden factor matrix U of user is:
The conditional probability of the hidden factor matrix V of article is:
Score in predicting formula is:
Wherein VjIt is the characteristic vector of article j, UiIt is the characteristic vector of user i, UυIt is the characteristic vector of user υ, UwIt is user w's
Characteristic vector, VjIt is the characteristic vector of article j, FiFor the good friend of user i gathers, NiFor the similarity user high of user i gathers,
SiwRepresent the similarity between user i and user w, TivIt is the relation value after the normalization of user i and article w,It is user
The inner product of the characteristic vector of i and the characteristic vector of article j,It is the characteristic vector and the characteristic vector of article j of user w
Inner product,It is the inner product of the characteristic vector of the characteristic vector and article j of user υ, λTAnd λSIt is adjustable weight parameter, 0 < λT<
1,0 < λS< 1, and λT+λS≤1。
8. the collaborative filtering method of social networks and user's similarity is incorporated simultaneously as claimed in claim 7, it is characterised in that
Probability matrix decomposition model according to similar users feature and good friend's user characteristics is incorporated obtains prediction scoring of the user to article
The step of include:
Conditional probability according to prediction scoring formula, the conditional probability of the hidden factor matrix of user and the hidden factor matrix of article proposes need
Want the posterior probability of maximized hidden factor matrix:
Above-mentioned probability function is maximized, is equivalent to minimize following loss object function,
Wherein, U is the hidden factor matrix of user, and V is the hidden factor matrix of article, and R is user-article rating matrix, after T is for normalization
User-user relational matrix, S be user's similarity matrix, RijScoring for user i to article j, UiIt is the feature of user i
Vector, UvIt is the characteristic vector of user v, UwIt is the characteristic vector of user w, VjIt is the characteristic vector of article j, FiIt is good for user i
Friend's set, NiFor the similarity user high of user i gathers, SiwIt is the relation value after the normalization of user i and article w,For
The inner product of the characteristic vector of user i and the characteristic vector of article j,It is the characteristic vector and the characteristic vector of article j of user w
Inner product,It is the inner product of the characteristic vector of the characteristic vector and article j of user υ,Represent Frobenius norms;
Using the method optimization loss object function of stochastic gradient descent, the hidden factor matrix U of user and the hidden factor square of article are obtained
Battle array V, and usePrediction scoring.
9. the collaborative filtering side of social networks and user's similarity is incorporated while as described in any one in claim 1~6
Method, it is characterised in that the step of being scored as user's recommendation article according to prediction includes:Scored according to prediction, to preceding prediction scoring
Article preceding N high is recommended.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611237097.3A CN106874355A (en) | 2016-12-28 | 2016-12-28 | The collaborative filtering method of social networks and user's similarity is incorporated simultaneously |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611237097.3A CN106874355A (en) | 2016-12-28 | 2016-12-28 | The collaborative filtering method of social networks and user's similarity is incorporated simultaneously |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106874355A true CN106874355A (en) | 2017-06-20 |
Family
ID=59164402
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611237097.3A Pending CN106874355A (en) | 2016-12-28 | 2016-12-28 | The collaborative filtering method of social networks and user's similarity is incorporated simultaneously |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106874355A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106886559A (en) * | 2016-12-28 | 2017-06-23 | 浙江浙大网新集团有限公司 | The collaborative filtering method of good friend's feature and similar users feature is incorporated simultaneously |
CN107862012A (en) * | 2017-10-30 | 2018-03-30 | 江苏大学 | A kind of information resources auto recommending method for group of college students |
CN108509560A (en) * | 2018-03-23 | 2018-09-07 | 广州杰赛科技股份有限公司 | User's similarity preparation method and device, equipment, storage medium |
CN108563690A (en) * | 2018-03-15 | 2018-09-21 | 中山大学 | A kind of collaborative filtering recommending method based on object-oriented cluster |
CN110020883A (en) * | 2018-12-12 | 2019-07-16 | 阿里巴巴集团控股有限公司 | The method and device that unknown scoring in a kind of pair of rating matrix is predicted |
CN110413878A (en) * | 2019-07-04 | 2019-11-05 | 四川金赞科技有限公司 | User based on adaptive elastomeric network-commodity preference prediction meanss and method |
CN110489656A (en) * | 2019-07-02 | 2019-11-22 | 华南师范大学 | Item recommendation method, system and storage medium based on schema merging |
CN111737576A (en) * | 2020-06-22 | 2020-10-02 | 中国银行股份有限公司 | Application function personalized recommendation method and device |
CN111858972A (en) * | 2020-07-28 | 2020-10-30 | 山东大学 | Movie recommendation method based on family knowledge graph |
CN112948701A (en) * | 2021-04-16 | 2021-06-11 | 泰康保险集团股份有限公司 | Information recommendation device, method, equipment and storage medium |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102789499A (en) * | 2012-07-16 | 2012-11-21 | 浙江大学 | Collaborative filtering method on basis of scene implicit relation among articles |
CN103246672A (en) * | 2012-02-09 | 2013-08-14 | 中国科学技术大学 | Method and device for performing personalized recommendation on users |
CN105160539A (en) * | 2015-06-17 | 2015-12-16 | 南京邮电大学 | Probability matrix decomposition recommendation method |
CN105761151A (en) * | 2016-02-02 | 2016-07-13 | 浙江工业大学 | Mobile socialization recommendation method based on joint probability matrix decomposition |
US20160217374A1 (en) * | 2015-01-27 | 2016-07-28 | Milq Inc. | Method and system utilizing collaborative filtering |
CN106126549A (en) * | 2016-06-16 | 2016-11-16 | 传化公路港物流有限公司 | A kind of community's trust recommendation method decomposed based on probability matrix and system thereof |
CN106886559A (en) * | 2016-12-28 | 2017-06-23 | 浙江浙大网新集团有限公司 | The collaborative filtering method of good friend's feature and similar users feature is incorporated simultaneously |
-
2016
- 2016-12-28 CN CN201611237097.3A patent/CN106874355A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103246672A (en) * | 2012-02-09 | 2013-08-14 | 中国科学技术大学 | Method and device for performing personalized recommendation on users |
CN102789499A (en) * | 2012-07-16 | 2012-11-21 | 浙江大学 | Collaborative filtering method on basis of scene implicit relation among articles |
US20160217374A1 (en) * | 2015-01-27 | 2016-07-28 | Milq Inc. | Method and system utilizing collaborative filtering |
CN105160539A (en) * | 2015-06-17 | 2015-12-16 | 南京邮电大学 | Probability matrix decomposition recommendation method |
CN105761151A (en) * | 2016-02-02 | 2016-07-13 | 浙江工业大学 | Mobile socialization recommendation method based on joint probability matrix decomposition |
CN106126549A (en) * | 2016-06-16 | 2016-11-16 | 传化公路港物流有限公司 | A kind of community's trust recommendation method decomposed based on probability matrix and system thereof |
CN106886559A (en) * | 2016-12-28 | 2017-06-23 | 浙江浙大网新集团有限公司 | The collaborative filtering method of good friend's feature and similar users feature is incorporated simultaneously |
Non-Patent Citations (3)
Title |
---|
C. WANG 等: "A Modified PMF Model Incorporating Implicit Item Associations" * |
MA H 等: "Learning to recommend with social trust ensemble", pages 4 * |
薛毅: "基于用户信任影响力的社会化推荐算法研究", pages 7 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106886559A (en) * | 2016-12-28 | 2017-06-23 | 浙江浙大网新集团有限公司 | The collaborative filtering method of good friend's feature and similar users feature is incorporated simultaneously |
CN107862012A (en) * | 2017-10-30 | 2018-03-30 | 江苏大学 | A kind of information resources auto recommending method for group of college students |
CN108563690A (en) * | 2018-03-15 | 2018-09-21 | 中山大学 | A kind of collaborative filtering recommending method based on object-oriented cluster |
CN108563690B (en) * | 2018-03-15 | 2022-01-21 | 中山大学 | Collaborative filtering recommendation method based on object-oriented clustering |
CN108509560A (en) * | 2018-03-23 | 2018-09-07 | 广州杰赛科技股份有限公司 | User's similarity preparation method and device, equipment, storage medium |
CN110020883A (en) * | 2018-12-12 | 2019-07-16 | 阿里巴巴集团控股有限公司 | The method and device that unknown scoring in a kind of pair of rating matrix is predicted |
CN110489656A (en) * | 2019-07-02 | 2019-11-22 | 华南师范大学 | Item recommendation method, system and storage medium based on schema merging |
CN110413878A (en) * | 2019-07-04 | 2019-11-05 | 四川金赞科技有限公司 | User based on adaptive elastomeric network-commodity preference prediction meanss and method |
CN110413878B (en) * | 2019-07-04 | 2022-04-15 | 五五海淘(上海)科技股份有限公司 | User-commodity preference prediction device and method based on adaptive elastic network |
CN111737576A (en) * | 2020-06-22 | 2020-10-02 | 中国银行股份有限公司 | Application function personalized recommendation method and device |
CN111737576B (en) * | 2020-06-22 | 2023-09-19 | 中国银行股份有限公司 | Application function personalized recommendation method and device |
CN111858972A (en) * | 2020-07-28 | 2020-10-30 | 山东大学 | Movie recommendation method based on family knowledge graph |
CN111858972B (en) * | 2020-07-28 | 2023-01-31 | 山东大学 | Movie recommendation method based on family knowledge graph |
CN112948701A (en) * | 2021-04-16 | 2021-06-11 | 泰康保险集团股份有限公司 | Information recommendation device, method, equipment and storage medium |
CN112948701B (en) * | 2021-04-16 | 2023-10-20 | 泰康保险集团股份有限公司 | Information recommendation device, method, equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106874355A (en) | The collaborative filtering method of social networks and user's similarity is incorporated simultaneously | |
CN111428147B (en) | Social recommendation method of heterogeneous graph volume network combining social and interest information | |
CN109785062B (en) | Hybrid neural network recommendation system based on collaborative filtering model | |
CN110502704B (en) | Group recommendation method and system based on attention mechanism | |
CN106886559A (en) | The collaborative filtering method of good friend's feature and similar users feature is incorporated simultaneously | |
CN108665323B (en) | Integration method for financial product recommendation system | |
CN107220365A (en) | Accurate commending system and method based on collaborative filtering and correlation rule parallel processing | |
CN104462383B (en) | A kind of film based on a variety of behavior feedbacks of user recommends method | |
Stankovic et al. | The digital competitiveness of European countries: A multiple-criteria approach. | |
CN112765480B (en) | Information pushing method and device and computer readable storage medium | |
CN106127506B (en) | recommendation method for solving cold start problem of commodity based on active learning | |
CN110008397A (en) | A kind of recommended models training method and device | |
CN113761359B (en) | Data packet recommendation method, device, electronic equipment and storage medium | |
Ahmed et al. | Discovering diverse, high quality design ideas from a large corpus | |
CN107656918B (en) | Obtain the method and device of target user | |
CN107944485A (en) | The commending system and method, personalized recommendation system found based on cluster group | |
CN106919699A (en) | A kind of recommendation method for personalized information towards large-scale consumer | |
CN109597899A (en) | The optimization method of media personalized recommendation system | |
CN108241619A (en) | A kind of recommendation method based on the more interest of user | |
CN112699310A (en) | Cold start cross-domain hybrid recommendation method and system based on deep neural network | |
US20220075804A1 (en) | Method and device for providing guide information for enhancement of artist's reputation | |
CN104035978B (en) | Combo discovering method and system | |
CN108984551A (en) | A kind of recommended method and system based on the multi-class soft cluster of joint | |
CN112560105B (en) | Joint modeling method and device for protecting multi-party data privacy | |
Carvalho et al. | Generation of coalition structures to provide proper groups' formation in group recommender systems |
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 |