CN103150678A - Method and device for discovering inter-user potential focus relationships on microblogs - Google Patents
Method and device for discovering inter-user potential focus relationships on microblogs Download PDFInfo
- Publication number
- CN103150678A CN103150678A CN2013100775246A CN201310077524A CN103150678A CN 103150678 A CN103150678 A CN 103150678A CN 2013100775246 A CN2013100775246 A CN 2013100775246A CN 201310077524 A CN201310077524 A CN 201310077524A CN 103150678 A CN103150678 A CN 103150678A
- Authority
- CN
- China
- Prior art keywords
- user
- matrix
- close attention
- prime
- sigma
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method for discovering inter-user potential focus relationships on microblogs. The method comprises the following steps of: building a user focus relationship matrix according to a user set and inter-user focus relationships; calculating two non-negative decomposed matrixes of the user focus relationship matrix; and obtaining a potential focus relationship matrix according to the product of the two non-negative matrixes and the user focus relationship matrix. According to method provided by the invention, the potential focus relationships are discovered by combining the inter-user focus relationships and inter-user interbehavior information in the microblogs, and thus, result errors of discovering the inter-user potential focus relationships can be reduced.
Description
Technical field
The present invention relates to the network data excavation field, particularly discover method and the device of potential concern relation between the user in a kind of microblogging.
Background technology
Microblogging is a kind of internet, applications service, it utilizes wireless network, cable network, the communication technology etc. to carry out instant messaging, allow the user that oneself latest tendency and idea are sent to mobile phone and personalized web site group with the note form, and be not only to send to the individual.Different from general social networks is, microblogging limits the message-length of each transmission, thereby has reduced the requirement to user language layout tissue, the content of speech phrase also facilitate upgrade in time oneself personal information of user.In microblogging, the relation between the user can be divided into two kinds, and a kind of is explicit relation, i.e. concern relation between the good friend, and a kind of is implicit relationship, as forwarding messages between the good friend, replys message etc.At present, how by the message in the microblogging that obtains and the concern relation between the user, finding concern relation potential between the user, has been one of very interested problem of researchers.
Between existing discovery user, mostly the method for potential concern relation is to obtain by the linking relationship between the user (namely paying close attention to relation) user's concern graph of a relation, then according to the structure of figure, adopts potential good friend's relation of the mode analysis user of calculating optimal path.In addition, Non-negative Matrix Factorization is also popular a kind of technology in Data Mining at present, calculates by Non-negative Matrix Factorization that between the user, potential concern concerns that efficient is higher and is suitable for processing mass data.Yet, the discover method of these potential concern relations only considers to connect between the user in microblogging the structural information on limit usually, it is explicit concern relation, and do not utilize mutual-action behavior between the user (or claiming interbehavior), make by prior art and find that the resultant error of potential concern relation between the user is larger.
Summary of the invention
According to one embodiment of the invention, the discover method of potential concern relation between the user is provided in a kind of microblogging, comprising:
Step 1), build the user and pay close attention to relational matrix R according to paying close attention to set of relations between user collection and user, wherein the user that concentrates of user comprises targeted customer and candidate user, each the element R in matrix R
ijRepresent that i user is to j user's attention rate;
Step 2), two nonnegative decomposition matrix A and the B of compute matrix R, wherein, matrix A and matrix B are used for portraying concern relation and the interbehavior impact on potential concern relation between the user between the user;
Step 3), the product of matrix A and matrix B and matrix R are subtracted each other obtain matrix
, from matrix
In choose all elements of the position of all targeted customer's corresponding row and all candidate user respective column, obtain targeted customer's potential concern relational matrix R
*, matrix R wherein
*Each element
Represent that the individual targeted customer of i ' is to the potential attention rate of the individual candidate user of j '.
In one embodiment, in step 1), R when i user pays close attention to j user
ij=1, otherwise R
ij=0.
In one embodiment, the user that the user concentrates also comprises intermediate user, and wherein, intermediate user comprises the user that the targeted customer pays close attention to and the user who pays close attention to the targeted customer; Candidate user comprises the user of the intermediate user concern that intermediate user is concentrated and the user who pays close attention to intermediate user.
In a further embodiment, step 2), two nonnegative decomposition matrix A of compute matrix R and B comprise:
Step 21), initialization matrix A and matrix B be any nonnegative value matrix, the multiply each other matrix dimensionality of gained of matrix A and matrix B is identical with the dimension of matrix R;
Step 22), upgrade matrix A and matrix B, wherein,
Utilize each element in following formula renewal matrix A:
Non-negative parameter lambda
1, λ
2And λ
3Represent respectively smoothing factor, structure regularization coefficient and mutual regularization coefficient, a
ukElement in representing matrix A, the number of the row of n representing matrix A, K represents the number of intermediate user,
(k, k') expression intermediate user k and k ' are with respect to the interbehavior similarity of user i, W
u(k, k') expression intermediate user k and k ' are with respect to the structural similarity of user u, and W is by W
uThe intermediate user structural similarity matrix that (k, k') consists of;
Utilize each element in following formula renewal matrix B:
b
kiElement in representing matrix B, the number of the row of m representing matrix B, W
k(i, j) be user i and j with respect to the structural similarity of intermediate user k, W ' is by W
kThe candidate user structural similarity matrix that (i, j) consists of;
Step 23), the matrix A after calculate upgrading and upgrade after the product R'=AB of matrix B, utilize the error between following formula compute matrix R ' and matrix R:
Wherein, r
uiElement in representing matrix R,
Element in representing matrix R ',
If error rmse less than predetermined threshold, exports matrix A and matrix B after upgrading;
Otherwise, adjust parameter lambda
1, λ
2And λ
3, return to step 22).
In a further embodiment, when user i is candidate user and intermediate user k and k ' when having forwarded the message of user i issue,
(k, k') is 1, otherwise is 0.When user u is the targeted customer, and intermediate user k and k ' be when paying close attention to simultaneously user u or user u and paying close attention to simultaneously intermediate user k and k ', W
u(k, k') is 1, otherwise is 0.When user i and j are candidate user, and user i and j be when paying close attention to simultaneously intermediate user k or intermediate user k and paying close attention to simultaneously user i and j, W
k(i, j) is 1, otherwise is 0.
In one embodiment, described method also comprises:
Step 4) is with matrix R
*Every delegation sort by element value is descending, obtain each targeted customer, by the potential concern user list of potential attention rate size sequence.
According to one embodiment of present invention, also provide in a kind of microblogging the discovery device of potential concern relation between the user, comprising: input media and calculation element.Wherein input media is used for paying close attention to interbehavior collection between set of relations and interior user of a period of time between input user collection, user, and the user that the user concentrates comprises targeted customer and candidate user; Calculation element is used for paying close attention to relational matrix R according to paying close attention to set of relations structure user between user's collection and user, according to two nonnegative decomposition matrix A of interbehavior collection compute matrix R and B between the user in a period of time, subtract each other according to product and the matrix R of matrix A and matrix B the matrix that obtains
, from matrix
In choose all elements of the position of all targeted customer's corresponding row and all candidate user respective column, obtain targeted customer's potential concern relational matrix R
*
In one embodiment, described calculation element also is used for matrix R
*Every delegation sort by element value is descending, obtain each targeted customer, by the potential concern user list of potential attention rate size sequence.
Adopt the present invention can reach following beneficial effect:
The present invention considers to connect between the user between the structural information on limit and user interbehavior information and finds potential concern relation between the user, makes the resultant error rate of finding potential concern relation between the user lower.
Description of drawings
Fig. 1 is the process flow diagram of the discover method of potential concern relation between the user in microblogging according to an embodiment of the invention;
Fig. 2 is that the user pays close attention to the schematic diagram of relational matrix according to an embodiment of the invention;
Fig. 3 is the schematic diagram of user interactions behavioural matrix according to an embodiment of the invention;
Fig. 4 is that the user pays close attention to graph of a relation according to an embodiment of the invention;
Fig. 5 is the schematic diagram of user's similarity matrix according to an embodiment of the invention;
Fig. 6 be used for to calculate the process flow diagram of method that the user pays close attention to two nonnegative decomposition matrixes of relational matrix according to an embodiment of the invention.
Embodiment
The present invention is described in detail below in conjunction with the drawings and specific embodiments.
According to one embodiment of present invention, provide in a kind of microblogging the discover method of potential concern relation between the user.The method is applicable to provide the microblogging of basic function, and has considered to connect between the user limit (namely paying close attention to behavior) and interbehavior.
At first the basic function that microblogging is provided is described: the concern relation between the user comprises to be paid close attention to and is concerned; Message function comprises transmission, comment and the forwarding of message.Make the behavior be difficult to follow the trail of because the user sends message by personal letter usually, and user comment often reflect the interest of user session topic and can not accurately reflect relation between the user.Thereby, compare transmission and comment, this interbehavior of forwarding messages can embody the relation (for example, identical hobby etc. being arranged between the user) between the user more exactly.As not specifying, interbehavior of the present invention mainly refers to the forwarding of message between the user.Fig. 1 shows in microblogging the discover method of potential concern relation embodiment between the user, specifically comprises following five steps:
Step S101, set up the user and pay close attention to relational matrix R and user interactions behavioural matrix R
f
The input of this step can comprise: pay close attention in set of relations and a period of time Δ t interbehavior collection between the user between user collection, user, in one embodiment, between the user, interbehavior collective has showed the behavior of forwarding messages between the user in a period of time Δ t.
Can build the user according to concern set of relations between user's collection of inputting and user and pay close attention to relational matrix R.Fig. 2 shows the example that the user pays close attention to relational matrix R, in matrix R, and R
ijBe the element in matrix, be used for describing i user to j user's attention rate.In one embodiment, R
ij=1 i user of expression (the capable node i in matrix) pays close attention to j user (the row node j in matrix), R
ijI user of=0 expression do not pay close attention to j user.Pay close attention to relational matrix R according to the user, can obtain the user and pay close attention to graph of a relation.It is digraph that this user pays close attention to graph of a relation, and the node in figure represents the user, and the limit represents the concern relation between the user, the direction indication user's on limit concern direction.Hereinafter, the user is sometimes also referred to as node.
Except the user pays close attention to relational matrix, according to interbehavior collection between the user that the user collects and a period of time Δ t is interior of input, can also set up the user interactions behavioural matrix R in a period of time Δ t
fFig. 3 shows user interactions behavioural matrix R
fAn example, in this embodiment,
I user of=1 expression forwarded j user's message in time period Δ t,
I user of=0 expression do not forward j user's message in time period Δ t.This time period Δ t can be a week or one month etc.
Should be understood that and abovely only described to be exemplified as purpose a kind of construction method that the user pays close attention to relational matrix and user interactions behavioural matrix, the construction method of these two kinds of matrixes is not limited to this.
Step S102, the user that the user is paid close attention in relational matrix classify.
In one embodiment, the user can be divided three classes: targeted customer, intermediate user and candidate user.Can set up respectively three set according to this sorting technique: targeted customer's collection, intermediate user collection and candidate user collection.
Wherein, the targeted customer represents user specified, its potential concern relation of needs discovery; Intermediate user is the user that the targeted customer that concentrates of targeted customer pays close attention to, and the user who pays close attention to the targeted customer; Candidate user refers to the user of the intermediate user concern that intermediate user is concentrated, and the user who pays close attention to intermediate user.Fig. 4 has described the example that the user who comprises these three kinds of user's set pays close attention to graph of a relation, and wherein, the node that comprises in subgraph 41 represents the targeted customer, and the node that comprises in subgraph 42 represents intermediate user, and the node that comprises in subgraph 43 represents candidate user.
Step S103, calculate user's similarity according to the classification results of user in step S102.Wherein, user's similarity comprises the structural similarity of intermediate user, the interbehavior similarity of intermediate user and the structural similarity of candidate user.The structural similarity of intermediate user has embodied the intermediate user similarity degree in the concern behavior between any two; The interbehavior similarity of intermediate user has embodied the similarity degree of intermediate user on interbehavior (forwarding messages behavior); And the structural similarity of candidate user has embodied the similarity degree of candidate user in the concern behavior.
according to one embodiment of the invention, can adopt following method to calculate the structural similarity of intermediate user: to concentrate from the targeted customer and choose (choose arbitrarily or choose in certain sequence) targeted customer u, concentrate from middle user and choose arbitrarily two intermediate user k and k ', pay close attention to relational matrix R with reference to the user, if targeted customer u has paid close attention to intermediate user k and k ' simultaneously, perhaps intermediate user k and k ' have paid close attention to targeted customer u simultaneously, intermediate user k and k ' are 1 with respect to the structural similarity of targeted customer u, otherwise intermediate user k and k ' are 0 with respect to the structural similarity of targeted customer u.Calculating intermediate user concentrates all intermediate users to concentrate all targeted customers' structural similarity with respect to the targeted customer, obtain intermediate user structural similarity matrix, matrix form as shown in Figure 5, the row of this intermediate user structural similarity matrix is the targeted customer, and this matrix column is intermediate user pair.
according to one embodiment of the invention, can adopt following method calculated candidate user's structural similarity: concentrate from middle user and choose (choose arbitrarily or choose in certain sequence) intermediate user k, concentrate from candidate user and choose arbitrarily two candidate user i and j, pay close attention to relational matrix R with reference to the user, if this intermediate user k has paid close attention to candidate user i and j simultaneously, or candidate user i and j have paid close attention to intermediate user k simultaneously, candidate user i and j are 1 with respect to the structural similarity of intermediate user k, otherwise candidate user i and j are 0 with respect to the structural similarity of intermediate user k.Calculated candidate user concentrates all candidate user to concentrate the structural similarity of all intermediate users with respect to intermediate user, obtain candidate user structural similarity matrix, matrix form as shown in Figure 5, the row of this candidate user structural similarity matrix is intermediate user, and described matrix column is candidate user pair.
According to one embodiment of the invention, can adopt following method to calculate the interbehavior similarity of intermediate user: concentrate arbitrarily or choose a candidate user i with a definite sequence from candidate user, concentrating from middle user and choose any two intermediate user k and k '.With reference to user interactions behavioural matrix R
fIf intermediate user k and k ' have forwarded the message of candidate user i issue within a period of time, intermediate user k and k ' are 1 with respect to the interbehavior similarity of candidate user i, otherwise intermediate user k and k ' are 0 with respect to the interbehavior similarity of candidate user i.Calculating intermediate user concentrates all intermediate users to concentrate the interbehavior similarity of all candidate user with respect to candidate user, obtain the interbehavior similarity matrix of intermediate user, described matrix form as shown in Figure 5, the row of this intermediate user interbehavior similarity matrix is candidate user, and this matrix column is intermediate user pair.
Here only exemplarily provided the method for calculating user's similarity, should be understood that other methods that are suitable for measuring the user behavior similarity also are applicable to this.For example, the mode that can take the forwarding situation of some (as the num bar) message that two intermediate users issue within a period of time candidate user to add up one by one.Particularly, to every message of candidate user, if intermediate user has forwarded this message, adding up value is 1, otherwise is 0.The intermediate user that to obtain like this two length be num forwards vector to the message of candidate user, calculates the interbehavior similarity of intermediate user according to the similarity (as the cosine similarity) of vector.
Step S104, the user's similarity that obtains according to step S103 are calculated two nonnegative decomposition matrix A and B that the user pays close attention to relational matrix R.Fig. 6 shows the process flow diagram of an embodiment of the method, wherein matrix A and B are the matrixes of the representative of consumer feature of two low-dimensionals, be called user-factor matrix (user-factor matrix), construct them and can be used for portraying some hidden feature of user to paying close attention to the impact of relation between the user.Be different from existing just based on the matrix decomposition of structural information between the user (pay close attention to or be concerned), here through decomposing matrix A and the B that obtains, also consider interbehavior between the user (as forwarding messages) to the impact of potential concern relation between the user, more met the generation mechanism of paying close attention to relation in microblogging between the user.In one embodiment, can by the multiplication iteration more new formula obtain matrix A and B, concrete steps are as follows:
The first step: initialization matrix A and matrix B are any nonnegative value matrix, make the multiply each other dimension of matrix of gained of matrix A and matrix B identical with the dimension that the user pays close attention to relational matrix R.Simultaneously, the non-negative parameter lambda of initialization
1, λ
2And λ
3, wherein, λ
1Smoothing factor, λ
2Structure regularization coefficient, and λ
3It is mutual regularization coefficient.
Second step: upgrade matrix A, comprise following two steps:
A), appoint from matrix A and get an element a
uk, utilize formula (1) to calculate the updating value a' of this element
ukIn formula (1), the number of the row of n representing matrix A, K represents the number of intermediate user.
(k, k') expression intermediate user k and k ' be with respect to the interbehavior similarity of user i, when user i does not belong to the candidate user collection,
W
u(k, k') expression intermediate user k and k ' be with respect to the structural similarity of user u, when user u does not belong to the targeted customer and collects, and W
u(k, k')=0, W is by W
uThe intermediate user structural similarity matrix that (k, k') consists of.
Formula (1)
B), repeating step a), until in matrix A, each element calculates corresponding updating value, utilize all elements updating value of resulting matrix A to upgrade matrix A.
The 3rd step: upgrade matrix B, comprise following two steps:
C), appoint from matrix B and get an element b
ki, utilize formula (2) to calculate the updating value b' of described element
kiIn formula (2), the number of the row of m representing matrix B.
(k, k') expression intermediate user k and k ' be with respect to the interbehavior similarity of user i, when user i does not belong to the candidate user collection,
W
k(i, j) be user i and j with respect to the structural similarity of intermediate user k, when user i and j do not belong to the candidate user collection, W
k(i, j)=0.W ' is by W
kThe candidate user structural similarity matrix that (i, j) consists of.
D), repeating step c), until in matrix B, each element calculates corresponding updating value, utilize all elements updating value of resulting matrix B to upgrade matrix B.
The 4th step: according to upgrading the matrix A obtain and upgrading the matrix B that obtains, calculate the matrix A after upgrading and upgrade after the product R'=AB of matrix B, then pay close attention to error between relational matrix R according to formula (3) compute matrix R ' and user:
Wherein, r
uiThe expression user pays close attention to the element in relational matrix R,
Element in representing matrix R '.If errors rmse meets the demands (for example, less than a certain predetermined threshold), output matrix A and matrix B, otherwise, adjust parameter lambda
1, λ
2And λ
3, and return to second step and continue to upgrade matrix A and matrix B, until error rmse meets the demands.
The process that should be understood that above-mentioned renewal matrix B also can be carried out before upgrading matrix A.
In another embodiment, also can based on the subtraction iteration of Gradient Descent more new formula upgrade matrix A and B.
Step S105, two nonnegative decomposition matrix A that obtain according to step S104 and B, the calculating user pays close attention to and concerns estimated matrix R ', this user is paid close attention to concern that estimated matrix and user pay close attention to relational matrix R and subtract each other, concentrate all targeted customers to concentrate the potential concern user list of candidate user corresponding to candidate user according to gained Output matrix targeted customer.In one embodiment, obtain the method for this potential concern user list specific as follows:
Matrix A and the matrix B of step S104 being calculated gained multiply each other, and obtain the user and pay close attention to and concern estimated matrix R ', and described estimated matrix R ' pays close attention to relational matrix R with the user and subtracts each other, and obtains matrix
Choose matrix
In, the targeted customer concentrates all targeted customers' corresponding row and candidate user to concentrate all elements of the position of all candidate user respective column, obtains targeted customer's potential concern relational matrix R
*Targeted customer's potential concern relational matrix R wherein
*Each element
Represent that the individual targeted customer of i ' is to the potential attention rate (or claiming to pay close attention to possibility) of the individual candidate user of j '.Sort by element value is descending to the u in targeted customer's potential concern relational matrix is capable, obtain the potential concern user's of the capable corresponding targeted customer u of u sequence, the candidate user of the row correspondence of greatest member value is the potential concern of the maximum possible user of targeted customer u, the candidate user of the row correspondence of second largest element value is the potential concern user of second largest possibility of targeted customer u, by that analogy.Aforesaid operations is carried out in all provisional capitals to targeted customer's potential concern relational matrix, can obtain the targeted customer and concentrate each targeted customer by the potential concern user list of paying close attention to the sequence of possibility size.
It will be understood by those skilled in the art that above-mentioned steps S101 also can carry out after step S102.
The present invention adopts respectively in prior art and microblogging described herein the discover method of potential concern relation between the user, is carrying out many experiments on true Twitter data set, and experiment parameter is as follows:
Testing user used, to pay close attention to relational matrix R be random 10,000 users that select and pay close attention to the user and the matrix of 10,000 row 10,000 row that the bean vermicelli user obtains in this half middle of the month from November 19,1 day to 2012 October in 2012 that utilize Twitter API to gather.User interactions behavioural matrix R
fBasis 10,000 row 10,000 column matrix of interior user's forwarding messages situation generation during this period of time.The Threshold of error rmse is 0.00001, smoothing factor λ
1Being initialized as 0.001, is 0.01 through adjusted value, structure regularization coefficient lambda
2Being initialized as 0.001, is 0.001 through adjusted value, mutual regularization coefficient lambda
3Be initialized as 0.001 through being adjusted into 0.005, the intermediate user number is 4, and matrix A is the matrix of 10,000 row 4 row, and matrix B is the matrix of 4 row 10,000 row.
Obtain following result through experiment: adopting the error rmse value of prior art is 0.102, and between the user, the error rmse value of the resulting potential concern relational result of the discover method of potential concern relation is only 0.033 and adopt in microblogging provided by the invention.Thereby adopt in microblogging provided by the invention that between the user, the discover method of potential concern relation reaches 70% left and right than the error that prior art reduces, greatly reduced the resultant error of the potential concern relation between the user of finding.
According to one embodiment of present invention, in microblogging provided by the invention between the user discover method of potential relation can be used in the disparate networks service with microblogging characteristics, social networks such as Twitter, Sina's microblogging and Tengxun's microblogging.
According to one embodiment of present invention, also provide in a kind of microblogging the discovery device of potential concern relation between the user, comprise input media, calculation element.
Input media is used for paying close attention to interbehavior collection between set of relations and interior user of a period of time between input user collection, user, and wherein, the user that the user concentrates is divided into targeted customer, intermediate user and candidate user.
Calculation element, be used for paying close attention to relational matrix R according to paying close attention to set of relations structure user between user's collection and user, according in a period of time between the user interbehavior collection calculate two nonnegative decomposition matrix A and the B that the user pays close attention to relational matrix R, pay close attention to relational matrix R according to the product of matrix A and matrix B and user and subtract each other the matrix that obtains
, from matrix
In choose all elements of the position of all targeted customer's corresponding row and all candidate user respective column, obtain targeted customer's potential concern relational matrix R
*, matrix R wherein
*Each element
Represent that the individual targeted customer of i ' is to the potential attention rate of the individual candidate user of j '.
In one embodiment, calculation element also is used for matrix R
*Every delegation sort by element value is descending, obtain each targeted customer, by the potential concern user list of potential attention rate size sequence.
Should be noted that and understand, in the situation that do not break away from the desired the spirit and scope of the present invention of accompanying claim, can make to the present invention of foregoing detailed description various modifications and improvement.Therefore, the scope of claimed technical scheme is not subjected to the restriction of given any specific exemplary teachings.
Claims (16)
1. the discover method of potential concern relation between the user in a microblogging comprises:
Step 1), build the user and pay close attention to relational matrix R according to paying close attention to set of relations between user collection and user, wherein the user that concentrates of user comprises targeted customer and candidate user, each the element R in matrix R
ijRepresent that i user is to j user's attention rate;
Step 2), two nonnegative decomposition matrix A and the B of compute matrix R, wherein, matrix A and matrix B are used for portraying concern relation and the interbehavior impact on potential concern relation between the user between the user;
Step 3), the product of matrix A and matrix B and matrix R are subtracted each other obtain matrix
, from matrix
In choose all elements of the position of all targeted customer's corresponding row and all candidate user respective column, obtain targeted customer's potential concern relational matrix R
*, matrix R wherein
*Each element
Represent that the individual targeted customer of i ' is to the potential attention rate of the individual candidate user of j '.
2. method according to claim 1, wherein, in step 1) when i user pays close attention to j user R
ij=1, otherwise R
ij=0.
3. method according to claim 1 and 2, the user that in step 1), the user concentrates also comprises intermediate user, wherein
Intermediate user comprises the user that the targeted customer pays close attention to and the user who pays close attention to the targeted customer;
Candidate user comprises the user of the intermediate user concern that intermediate user is concentrated and the user who pays close attention to intermediate user.
4. method according to claim 3, wherein, step 2), two nonnegative decomposition matrix A of compute matrix R and B comprise:
Step 21), initialization matrix A and matrix B be any nonnegative value matrix, the multiply each other matrix dimensionality of gained of matrix A and matrix B is identical with the dimension of matrix R;
Step 22), upgrade matrix A and matrix B, wherein,
Utilize each element in following formula renewal matrix A:
Non-negative parameter lambda
1, λ
2And λ
3Represent respectively smoothing factor, structure regularization coefficient and mutual regularization coefficient, a
ukElement in representing matrix A, the number of the row of n representing matrix A, K represents the number of intermediate user,
(k, k') expression intermediate user k and k ' are with respect to the interbehavior similarity of user i, W
u(k, k') expression intermediate user k and k ' are with respect to the structural similarity of user u, and W is by W
uThe intermediate user structural similarity matrix that (k, k') consists of;
Utilize each element in following formula renewal matrix B:
b
kiElement in representing matrix B, the number of the row of m representing matrix B, W
k(i, j) be user i and j with respect to the structural similarity of intermediate user k, W ' is by W
kThe candidate user structural similarity matrix that (i, j) consists of;
Step 23), the matrix A after calculate upgrading and upgrade after the product R'=AB of matrix B, utilize the error between following formula compute matrix R ' and matrix R:
If error rmse less than predetermined threshold, exports matrix A and matrix B after upgrading;
Otherwise, adjust parameter lambda
1, λ
2And λ
3, return to step 22).
5. method according to claim 4, wherein, when user i is candidate user and intermediate user k and k ' when having forwarded the message of user i issue,
(k, k') is 1, otherwise is 0.
6. method according to claim 4, wherein, when user u is the targeted customer, and intermediate user k and k ' be when paying close attention to simultaneously user u or user u and paying close attention to simultaneously intermediate user k and k ', W
u(k, k') is 1, otherwise is 0.
7. method according to claim 4, wherein, when user i and j are candidate user, and user i and j be when paying close attention to simultaneously intermediate user k or intermediate user k and paying close attention to simultaneously user i and j, W
k(i, j) is 1, otherwise is 0.
8. method according to claim 1 and 2 also comprises:
Step 4) is with matrix R
*Every delegation sort by element value is descending, obtain each targeted customer, by the potential concern user list of potential attention rate size sequence.
9. the discovering device of potential concern relation between the user in a microblogging comprises:
Input media is used for paying close attention to interbehavior collection between set of relations and interior user of a period of time between input user collection, user, and wherein, the user that the user concentrates comprises targeted customer and candidate user;
Calculation element, be used for paying close attention to relational matrix R according to paying close attention to set of relations structure user between user's collection and user, according to two nonnegative decomposition matrix A of interbehavior collection compute matrix R and B between the user in a period of time, subtract each other according to product and the matrix R of matrix A and matrix B the matrix that obtains
, from matrix
In choose all elements of the position of all targeted customer's corresponding row and all candidate user respective column, obtain targeted customer's potential concern relational matrix R
*
10. equipment according to claim 9 wherein builds the user and pays close attention to relational matrix R and comprise according to paying close attention to set of relations between user's collection and user:
R when i user pays close attention to j user
ij=1, otherwise R
ij=0.
11. according to claim 9 or 10 described equipment, wherein the user that concentrates of user also comprises intermediate user, wherein
Intermediate user comprises the user that the targeted customer pays close attention to and the user who pays close attention to the targeted customer;
Candidate user comprises the user of the intermediate user concern that intermediate user is concentrated and the user who pays close attention to intermediate user.
12. equipment according to claim 11, wherein according in a period of time between the user two nonnegative decomposition matrixes of interbehavior collection compute matrix R comprise:
Initialization matrix A and matrix B are any nonnegative value matrix, and the multiply each other matrix dimensionality of gained of matrix A and matrix B is identical with the dimension of matrix R; Upgrade each element in matrix A and matrix B; Matrix A after calculating is upgraded and the product R'=AB of the matrix B after renewal; If the error rmse of matrix R ' and matrix R less than predetermined threshold, exports matrix A and matrix B after upgrading, otherwise continue to upgrade each element in matrix A and matrix B;
Wherein,
Utilize each element in following formula renewal matrix A:
Non-negative parameter lambda
1, λ
2And λ
3Represent respectively smoothing factor, structure regularization coefficient and mutual regularization coefficient, a
ukElement in representing matrix A, the number of the row of n representing matrix A, K represents the number of intermediate user,
(k, k') expression intermediate user k and k ' are with respect to the interbehavior similarity of user i, W
u(k, k') expression intermediate user k and k ' are with respect to the structural similarity of user u, and W is by W
uThe intermediate user structural similarity matrix that (k, k') consists of;
Utilize each element in following formula renewal matrix B:
Element in bki representing matrix B, the number of the row of m representing matrix B, Wk (i, j) be user i and j with respect to the structural similarity of intermediate user k, W ' is the candidate user structural similarity matrix that is made of Wk (i, j); And
Utilize following formula calculation error rmse:
14. equipment according to claim 12, wherein, when user u is the targeted customer, and intermediate user k and k ' be when paying close attention to simultaneously user u or user u and paying close attention to simultaneously intermediate user k and k ', W
u(k, k') is 1, otherwise is 0.
15. equipment according to claim 12, wherein, when user i and j are candidate user, and user i and j be when paying close attention to simultaneously intermediate user k or intermediate user k and paying close attention to simultaneously user i and j, W
k(i, j) is 1, otherwise is 0.
16. according to claim 9 or 10 described equipment, described calculation element also are used for matrix R
*Every delegation sort by element value is descending, obtain each targeted customer, by the potential concern user list of potential attention rate size sequence.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310077524.6A CN103150678B (en) | 2013-03-12 | 2013-03-12 | Method and device for discovering inter-user potential focus relationships on microblogs |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310077524.6A CN103150678B (en) | 2013-03-12 | 2013-03-12 | Method and device for discovering inter-user potential focus relationships on microblogs |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103150678A true CN103150678A (en) | 2013-06-12 |
CN103150678B CN103150678B (en) | 2014-12-10 |
Family
ID=48548734
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310077524.6A Active CN103150678B (en) | 2013-03-12 | 2013-03-12 | Method and device for discovering inter-user potential focus relationships on microblogs |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103150678B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103744994A (en) * | 2014-01-22 | 2014-04-23 | 中国科学院信息工程研究所 | Communication-network-oriented user behavior pattern mining method and system |
CN103746895A (en) * | 2013-10-29 | 2014-04-23 | 广州华多网络科技有限公司 | Method and system for processing concern extent among communication users |
CN104702488A (en) * | 2013-12-10 | 2015-06-10 | 上海由你网络科技有限公司 | System and method for recommending friends |
CN105812280A (en) * | 2016-05-05 | 2016-07-27 | 四川九洲电器集团有限责任公司 | Classification method and electronic equipment |
CN105824921A (en) * | 2016-03-16 | 2016-08-03 | 广州彩瞳网络技术有限公司 | User social relation recognition device and method |
CN107122455A (en) * | 2017-04-26 | 2017-09-01 | 中国人民解放军国防科学技术大学 | A kind of network user's enhancing method for expressing based on microblogging |
CN109783740A (en) * | 2019-01-24 | 2019-05-21 | 北京字节跳动网络技术有限公司 | Pay close attention to the sort method and device of the page |
CN114996561A (en) * | 2021-03-02 | 2022-09-02 | 腾讯科技(深圳)有限公司 | Information recommendation method and device based on artificial intelligence |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102722566A (en) * | 2012-06-04 | 2012-10-10 | 上海电力学院 | Method for inquiring potential friends in social network |
-
2013
- 2013-03-12 CN CN201310077524.6A patent/CN103150678B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102722566A (en) * | 2012-06-04 | 2012-10-10 | 上海电力学院 | Method for inquiring potential friends in social network |
Non-Patent Citations (5)
Title |
---|
张怡文: "社交网络中分类推荐用户的设计与实现", 《贵州师范大学学报(自然科学版)》, vol. 31, no. 1, 28 February 2013 (2013-02-28), pages 94 - 97 * |
张聪: "微博用户关注推荐及排名策略研究", 《郑州大学硕士学位论文》, 31 December 2012 (2012-12-31), pages 33 - 38 * |
王连喜等: "微博用户关系挖掘研究综述", 《情报杂志》, vol. 31, no. 12, 31 December 2012 (2012-12-31) * |
程学旗,郭嘉丰,靳小龙: "网络信息的检索与挖掘回顾", 《中文信息学报》, vol. 25, no. 6, 30 November 2011 (2011-11-30), pages 111 - 117 * |
范敏敏: "非负矩阵分解与聚类方法在个性化推荐***中的应用研究", 《华东交通大学全日制专业学位硕士论文》, 31 December 2012 (2012-12-31), pages 27 - 32 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103746895A (en) * | 2013-10-29 | 2014-04-23 | 广州华多网络科技有限公司 | Method and system for processing concern extent among communication users |
CN104702488A (en) * | 2013-12-10 | 2015-06-10 | 上海由你网络科技有限公司 | System and method for recommending friends |
CN103744994B (en) * | 2014-01-22 | 2016-11-16 | 中国科学院信息工程研究所 | A kind of user's behavior pattern mining method and system towards communication network |
CN103744994A (en) * | 2014-01-22 | 2014-04-23 | 中国科学院信息工程研究所 | Communication-network-oriented user behavior pattern mining method and system |
CN105824921A (en) * | 2016-03-16 | 2016-08-03 | 广州彩瞳网络技术有限公司 | User social relation recognition device and method |
US10860595B2 (en) | 2016-03-16 | 2020-12-08 | Guangzhou Uc Network Technology Co., Ltd. | User social-relationship identification apparatus, method, and terminal device |
CN105812280B (en) * | 2016-05-05 | 2019-06-04 | 四川九洲电器集团有限责任公司 | A kind of classification method and electronic equipment |
CN105812280A (en) * | 2016-05-05 | 2016-07-27 | 四川九洲电器集团有限责任公司 | Classification method and electronic equipment |
CN107122455A (en) * | 2017-04-26 | 2017-09-01 | 中国人民解放军国防科学技术大学 | A kind of network user's enhancing method for expressing based on microblogging |
CN107122455B (en) * | 2017-04-26 | 2019-12-31 | 中国人民解放军国防科学技术大学 | Network user enhanced representation method based on microblog |
CN109783740A (en) * | 2019-01-24 | 2019-05-21 | 北京字节跳动网络技术有限公司 | Pay close attention to the sort method and device of the page |
CN114996561A (en) * | 2021-03-02 | 2022-09-02 | 腾讯科技(深圳)有限公司 | Information recommendation method and device based on artificial intelligence |
CN114996561B (en) * | 2021-03-02 | 2024-03-29 | 腾讯科技(深圳)有限公司 | Information recommendation method and device based on artificial intelligence |
Also Published As
Publication number | Publication date |
---|---|
CN103150678B (en) | 2014-12-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103150678B (en) | Method and device for discovering inter-user potential focus relationships on microblogs | |
CN103793489B (en) | Method for discovering topics of communities in on-line social network | |
CN104281882A (en) | Method and system for predicting social network information popularity on basis of user characteristics | |
CN102262681A (en) | Method for identifying key blog sets in blog information spreading | |
CN103179198B (en) | Based on the topic influence individual method for digging of many relational networks | |
CN104537096A (en) | Microblog message influence measuring method based on microblog message propagation tree | |
CN105550275A (en) | Microblog forwarding quantity prediction method | |
CN103294833A (en) | Junk user discovering method based on user following relationships | |
Dupuis | Chain rules for quantum Rényi entropies | |
CN104376083A (en) | Graph recommendation method based on concern relations and multiple user behaviors | |
Zhang et al. | On predicting Twitter trend: factors and models | |
Wang et al. | Predicting user activity level in point processes with mass transport equation | |
Wu et al. | Reveal quantum correlation in complementary bases | |
Raayatpanah et al. | Minimum cost multiple multicast network coding with quantized rates | |
CN104035978A (en) | Association discovering method and system | |
Zhao et al. | A Retweet Number Prediction Model Based on Followers' Retweet Intention and Influence | |
Zhang | Graph sampling by lagged random walk | |
Zhang et al. | Role of subgraphs in epidemics over finite-size networks under the scaled SIS process | |
CN107346517A (en) | User-interaction parameter acquisition methods and acquisition device in customer relationship network | |
Smailovic et al. | Calculating user's social influence through the SmartSocial Platform | |
Ebihara et al. | Dominant pole analysis of stable time‐delay positive systems | |
Bourguignon et al. | A new geometric INAR (1) process based on counting series with deflation or inflation of zeros | |
Bonmariage et al. | A survey of optimal network congestion control for unicast and multicast transmission | |
Jiang et al. | Efficiency improvements in social network communication via MapReduce | |
Pourheidari et al. | Recommender system based on extracted data from different social media. a study of twitter and linkedin |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20130612 Assignee: Branch DNT data Polytron Technologies Inc Assignor: Institute of Computing Technology, Chinese Academy of Sciences Contract record no.: 2018110000033 Denomination of invention: Method and device for discovering inter-user potential focus relationships on microblogs Granted publication date: 20141210 License type: Common License Record date: 20180807 |
|
EE01 | Entry into force of recordation of patent licensing contract |