CN104750731A - Method and device for obtaining complete user portrait - Google Patents
Method and device for obtaining complete user portrait Download PDFInfo
- Publication number
- CN104750731A CN104750731A CN201310745711.7A CN201310745711A CN104750731A CN 104750731 A CN104750731 A CN 104750731A CN 201310745711 A CN201310745711 A CN 201310745711A CN 104750731 A CN104750731 A CN 104750731A
- Authority
- CN
- China
- Prior art keywords
- parameter matrix
- matrix
- user
- portrait
- customer parameter
- 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
Landscapes
- Information Transfer Between Computers (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for obtaining a complete user portrait. The method includes: obtaining an incomplete user portrait matrix and randomly generating a user parameter matrix P and a label matrix Q; calculating portrait errors of the first part of users and updating the user parameter matrix and the label parameter matrix, wherein the first variable difference value of the first part of the users is larger than that of the first rest part of the users, the first rest part of the users are the users in addition to the first part of the users of multiple users, and the first variable difference value is the difference value between the first predicted value updated by the r-1 times updating of the users and the first predicated value updated by the r-2 times updating of the users; and after the R times updating of the user parameter matrix P and the label parameter matrix Q and according to the result of matrix decomposition, obtaining the complete user portrait matrix. The method and device for obtaining the complete user portrait can improve the convergence speed of the portrait errors, and reduces the calculating amount and the calculating time of a system.
Description
Technical field
The present invention relates to field of information processing, be specifically related to a kind of method and the device that obtain whole user portrait.
Background technology
User tag plays an important role in actual applications.Such as, know that the interest that label comprises user and the place often gone just can be made and market accurately, better service experience can be brought to user simultaneously.Usually, the label of user can by a matrix r
m × Nrepresent, wherein 1≤m≤M represents user index, i.e. a total M user, and 1≤n≤N represents user tag index, and namely each user one has N number of label.Usually, the user tag observed is all incomplete, namely there is the value of unknown label, represents with " " symbol.When there is a large number of users number and number of tags, matrix r
m × Nmay be extremely sparse, be filled with 0 a large amount of elements in matrix, namely this 0 element represents Unknown Label, is referred to as incomplete user and draws a portrait matrix.
In order to the label information of each user can be obtained, adopt at present the method for matrix decomposition, draw a portrait matrix according to the user of incompleteness, record complete user in advance and draw a portrait matrix.Specifically comprising of the method:
Step one: input incomplete portrait matrix r
m × N, and random initializtion customer parameter matrix P and tag parameter matrix Q.
Step 2: compute matrix r
m × Nthe portrait error e of middle nonzero element
m, n, computing formula is:
e
m,n=r
m,n-p
m Tq
n。
Step 3: according to portrait error update parameter P and Q, computing formula is:
p
m=p
m+γ(e
m,nq
n-λ
Pp
m),q
n=q
n+γ(e
m,np
m-λ
Qq
n)。
Wherein, γ is learning efficiency, is traditionally arranged to be γ=0.001.
Step 4: repetition step 2 and step 3 are until matrix r
m × Nin all nonzero elements scan one time, be referred to as once to circulate.
Step 5: repeat step 4 1≤t≤T circulation until customer parameter matrix P and tag parameter matrix Q and portrait error are roughly stabilized in a state, namely repeat step 4 until portrait error convergence.
Step 6: export whole user portrait matrix
But, find in enforcement the method: each circulation needs scanning t test matrix r
m × Nin all nonzero elements, each nonzero element needs the portrait error calculated on all class of subscribers, and it is many to reach the cycle index that portrait error convergence needs, and therefore the operand of the method is large, very consuming time.
Summary of the invention
The object of the invention is to a kind of method and the device that obtain whole user portrait, described method is the portrait error according to Part I user, upgrades customer parameter matrix and tag parameter matrix, thus can reduce operand.
A kind of method obtaining whole user portrait that first aspect present invention provides, comprising:
Obtain incomplete user and draw a portrait matrix, and the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q; Wherein, the user of described incompleteness draws a portrait matrix, multiple users that described customer parameter matrix P is corresponding identical with described tag parameter matrix Q, and described customer parameter matrix P and described tag parameter matrix Q is that the user of described incompleteness draws a portrait the split-matrix of matrix;
Drawing a portrait matrix according to the user of described incompleteness, upgrading described customer parameter matrix P and described tag parameter matrix Q, R for R time for making the update times needed for the first portrait error convergence; Described first portrait error is the difference that the user of described incompleteness draws a portrait in matrix between nonzero element and the first corresponding predicted value, and described first predicted value is that the merging entry of a matrix of described customer parameter matrix P and described tag parameter matrix Q is plain;
Wherein, upgrade described customer parameter matrix P and described tag parameter matrix Q the r time, r is more than or equal to 1 and is less than or equal to R, comprising:
As r=1, draw a portrait matrix, the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q according to the user of described incompleteness, the user calculating described incompleteness draws a portrait the first portrait error of nonzero element in matrix; According to described first portrait error, upgrade described customer parameter matrix P and described tag parameter matrix Q;
When r is more than or equal to 2, upgrades described customer parameter matrix P and described tag parameter matrix Q, comprising for described the r time:
Draw a portrait matrix, described customer parameter matrix P and described tag parameter matrix Q according to the user of described incompleteness, calculate the first portrait error of the Part I user in described multiple user; The first change difference of described Part I user is greater than the first change difference of the first remaining users, and described first remaining users is the user except described Part I user in described multiple user;
When r equals 2, described first change difference is the difference according to upgrading for the r-1 time between first predicted value of the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition and the first predicted value obtained according to initial described customer parameter matrix P and the described tag parameter matrix Q of stochastic generation;
When r is greater than 2, described first change difference be according to upgrade for the r-1 time the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition the first predicted value and according to the difference upgraded for the r-2 time between the first predicted value that the described customer parameter matrix P that obtains and described tag parameter matrix Q obtains;
According to the first portrait error of described Part I user, upgrade described customer parameter matrix P and described tag parameter matrix Q;
Upgrade after described customer parameter matrix P and described tag parameter matrix Q at the R time, draw a portrait matrix and the described customer parameter matrix P of the R time renewal and described tag parameter matrix Q according to the user of described incompleteness, obtain complete user and draw a portrait matrix.
In conjunction with first aspect present invention, in the first possibility implementation of first aspect, upgrade described customer parameter matrix P and described tag parameter matrix Q described the r time, r is more than or equal to 1 and is less than or equal to R, comprising:
Matrix, described customer parameter matrix P and described tag parameter matrix Q is being drawn a portrait according to the user of described incompleteness, before calculating the first portrait error of the Part I user in described multiple user, calculate the change difference of described multiple user, and the change difference of described multiple user is sorted, select the user of predetermined ratio as described Part I user, the change difference of the user of described predetermined ratio is all greater than the change difference of non-selected user in described multiple user.
May implementation in conjunction with the first of first aspect present invention or first aspect, may in implementation at the second of first aspect, after the R time renewal described customer parameter matrix P and described tag parameter matrix Q, described method also comprises:
Described customer parameter matrix P is carried out S deep decomposition, and obtains customer parameter matrix and the tag parameter matrix of S the level degree of depth; Described is for S time the number of times set;
Wherein, customer parameter matrix P described in the w time deep decomposition, w is more than or equal to 1 and is less than or equal to S, comprising:
Stochastic generation customer parameter matrix P
wwith tag parameter matrix Q
w; Wherein, described customer parameter matrix P
w, described tag parameter matrix Q
widentical multiple users corresponding to described customer parameter matrix P; Customer parameter matrix P
wwith tag parameter matrix Q
wfor the split-matrix of described customer parameter matrix P;
According to described customer parameter matrix P, upgrade described customer parameter matrix P Y time
wwith described tag parameter matrix Q
w, Y is for making the update times needed for the second portrait error convergence; Described second portrait error is the difference in described customer parameter matrix P between nonzero element and the second corresponding predicted value, and described second predicted value is described customer parameter matrix P
wwith described tag parameter matrix Q
wmerging entry of a matrix element;
Wherein, described customer parameter matrix P is upgraded the y time
wwith described tag parameter matrix Q
w, y is more than or equal to 1 and is less than or equal to Y, comprising:
As y=1, according to described customer parameter matrix P, the customer parameter matrix P that stochastic generation is initial
wwith tag parameter matrix Q
w, the user calculating described incompleteness draws a portrait the second portrait error of nonzero element in matrix;
According to described second portrait error, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
When y is more than or equal to 2, according to described customer parameter matrix P, described customer parameter matrix P
wwith described tag parameter matrix Q
w, calculate the second portrait error of the Part II user in described multiple user; The second change difference of described Part II user is greater than the second change difference of the second remaining users, and described second remaining users is the user except described Part II user in described multiple user;
When y equals 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wthe second predicted value obtained with according to the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q
wdifference between the second predicted value obtained;
When y is greater than 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P and tag parameter matrix Q that obtain
wobtain the second predicted value with upgrade according to the y-2 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wdifference between the second predicted value obtained;
According to the second portrait error of described Part II user nonzero element, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
The described customer parameter matrix P obtained is upgraded by Y time
wwith described tag parameter matrix Q
was the customer parameter matrix P of the w level degree of depth
wwith tag parameter matrix Q
w;
When described w is less than described S, upgrade Y time the described customer parameter matrix P obtained
was the w+1 time want the customer parameter matrix P of deep decomposition;
The described user according to described incompleteness draws a portrait matrix, and the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, obtains complete user and draws a portrait matrix, comprising:
Described customer parameter matrix P is being carried out S deep decomposition, and after the customer parameter matrix obtaining S the level degree of depth and tag parameter matrix, matrix is drawn a portrait according to the user of described incompleteness, the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, and the customer parameter matrix of S the level degree of depth obtained and tag parameter matrix, obtain complete user and draw a portrait matrix.
Second aspect present invention provides a kind of signal conditioning package, comprising:
Load module, draws a portrait matrix for obtaining incomplete user, and incomplete user described in stochastic generation draws a portrait the customer parameter matrix P of matrix and tag parameter matrix Q; Wherein, the user of described incompleteness draws a portrait multiple users corresponding to matrix, described customer parameter matrix P and described tag parameter matrix Q;
Matrix decomposition module, for: obtain incomplete user at described load module and draw a portrait matrix, and incomplete user described in stochastic generation draws a portrait after the customer parameter matrix P of matrix and tag parameter matrix Q, matrix is drawn a portrait according to the user of described incompleteness, R the described customer parameter matrix P of renewal and described tag parameter matrix Q, R are for making the update times needed for the first portrait error convergence; Described first portrait error is the difference that the user of described incompleteness draws a portrait in matrix between nonzero element and the first corresponding predicted value, and described first predicted value is that the merging entry of a matrix of described customer parameter matrix P and described tag parameter matrix Q is plain;
Wherein, upgrade described customer parameter matrix P and described tag parameter matrix Q the r time, r is more than or equal to 1 and is less than or equal to R, comprising:
As r=1, upgrade described customer parameter matrix P and described tag parameter matrix Q described the r time, comprise: draw a portrait matrix according to the user of described incompleteness, the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q, the user calculating described incompleteness draws a portrait the first portrait error of nonzero element in matrix; According to described first portrait error, upgrade described customer parameter matrix P and described tag parameter matrix Q;
When r is more than or equal to 2, upgrades described customer parameter matrix P and described tag parameter matrix Q, comprising for described the r time:
Draw a portrait matrix, described customer parameter matrix P and described tag parameter matrix Q according to the user of described incompleteness, calculate the first portrait error of the Part I user in described multiple user; The first change difference of described Part I user is greater than the first change difference of the first remaining users, and described first remaining users is the user except described Part I user in described multiple user;
When r equals 2, described first change difference is the difference according to upgrading for the r-1 time between first predicted value of the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition and the first predicted value obtained according to initial described customer parameter matrix P and the described tag parameter matrix Q of stochastic generation;
When r is greater than 2, described first change difference be according to upgrade for the r-1 time the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition the first predicted value and according to the difference upgraded for the r-2 time between the first predicted value that the described customer parameter matrix P that obtains and described tag parameter matrix Q obtains;
According to the first portrait error of described Part I user, upgrade described customer parameter matrix P and described tag parameter matrix Q;
Output module, for after the R time upgrades described customer parameter matrix P and described tag parameter matrix Q, draw a portrait matrix according to the user of described incompleteness, and the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, obtain complete user and draw a portrait matrix.
In conjunction with second aspect present invention, in the first possibility implementation of second aspect, described matrix decomposition module comprises dynamic dispatching module, described dynamic dispatching module is used for drawing a portrait matrix according to the user of described incompleteness, described customer parameter matrix P and described tag parameter matrix Q, before calculating the first portrait error of the Part I user in described multiple user, calculate the change difference of described multiple user, and the change difference of described multiple user is sorted, select the user of predetermined ratio as described Part I user, the change difference of the user of described predetermined ratio is all greater than the change difference of non-selected user in described multiple user.
In conjunction with second aspect present invention or the first possibility implementation of second aspect, in the second possibility implementation of second aspect, described matrix decomposition module also comprises deep decomposition module, described deep decomposition module is used for: after the R time upgrades described customer parameter matrix P and described tag parameter matrix Q, described customer parameter matrix P is carried out S deep decomposition, and obtains customer parameter matrix and the tag parameter matrix of S the level degree of depth; Described is for S time the number of times set;
Wherein, customer parameter matrix P described in the w time deep decomposition, w is more than or equal to 1 and is less than or equal to S, comprising:
Stochastic generation customer parameter matrix P
wwith tag parameter matrix Q
w; Wherein, described customer parameter matrix P
w, described tag parameter matrix Q
widentical multiple users corresponding to described customer parameter matrix P; Customer parameter matrix P
wwith tag parameter matrix Q
wfor the split-matrix of described customer parameter matrix P;
According to described customer parameter matrix P, upgrade described customer parameter matrix P Y time
wwith described tag parameter matrix Q
w, Y is for making the update times needed for the second portrait error convergence; Described second portrait error is the difference in described customer parameter matrix P between nonzero element and the second corresponding predicted value, and described second predicted value is described customer parameter matrix P
wwith described tag parameter matrix Q
wmerging entry of a matrix element;
Wherein, described customer parameter matrix P is upgraded the y time
wwith described tag parameter matrix Q
w, y is more than or equal to 1 and is less than or equal to Y, comprising:
As y=1, according to described customer parameter matrix P, the customer parameter matrix P that stochastic generation is initial
wwith tag parameter matrix Q
w, the user calculating described incompleteness draws a portrait the second portrait error of nonzero element in matrix;
According to described second portrait error, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
When y is more than or equal to 2, according to described customer parameter matrix P, described customer parameter matrix P
wwith described tag parameter matrix Q
w, calculate the second portrait error of the Part II user in described multiple user; The second change difference of described Part II user is greater than the second change difference of the second remaining users, and described second remaining users is the user except described Part II user in described multiple user;
When y equals 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wthe second predicted value obtained with according to the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q
wdifference between the second predicted value obtained;
When y is greater than 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P and tag parameter matrix Q that obtain
wobtain the second predicted value with upgrade according to the y-2 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wdifference between the second predicted value obtained;
According to the second portrait error of described Part II user nonzero element, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
The described customer parameter matrix P obtained is upgraded by Y time
wwith described tag parameter matrix Q
was the customer parameter matrix P of the w level degree of depth
wwith tag parameter matrix Q
w;
When described w is less than described S, upgrade Y time the described customer parameter matrix P obtained
was the w+1 time want the customer parameter matrix P of deep decomposition;
Described output module is also for carrying out S deep decomposition by described customer parameter matrix P, and after the customer parameter matrix obtaining S the level degree of depth and tag parameter matrix, matrix is drawn a portrait according to the user of described incompleteness, the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, and the customer parameter matrix of S the level degree of depth obtained and tag parameter matrix, obtain complete user and draw a portrait matrix.
In the present invention, the first change difference of the Part I user selected is greater than the first change difference of the first remaining users, according to the portrait error of Part I user, upgrade customer parameter matrix and tag parameter matrix, the speed of convergence of portrait error can be accelerated, reduce operand, thus reduce the operation time of system.
Accompanying drawing explanation
In order to be illustrated more clearly in the technical scheme in the embodiment of the present invention, below the accompanying drawing used required in describing embodiment is briefly described, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawing can also be obtained according to these accompanying drawings.
Fig. 1 is a kind of method flow schematic diagram obtaining whole user portrait that the embodiment of the present invention provides;
Fig. 2 is deep decomposition matrix procedures schematic diagram;
Fig. 3 is the structural representation of a kind of signal conditioning package that the embodiment of the present invention provides;
Fig. 4 is another structural representation of a kind of signal conditioning package that the embodiment of the present invention provides;
Fig. 5 is the structural representation of the another kind of signal conditioning package that the embodiment of the present invention provides.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, be clearly and completely described the technical scheme in the embodiment of the present invention, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art, not making the every other embodiment obtained under creative work prerequisite, belong to the scope of protection of the invention.
Below enumerate embodiment to be introduced the present invention.
As shown in Figure 1, the embodiment of the present invention provides a kind of method obtaining whole user portrait, comprising:
101, obtain incomplete user and draw a portrait matrix, and the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q; Wherein, the user of described incompleteness draws a portrait matrix, multiple users that described customer parameter matrix P is corresponding identical with described tag parameter matrix Q, and described customer parameter matrix P and described tag parameter matrix Q is that the user of described incompleteness draws a portrait the split-matrix of matrix.
102, drawing a portrait matrix according to the user of described incompleteness, upgrading described customer parameter matrix P and described tag parameter matrix Q, R for R time for making the update times needed for the first portrait error convergence; Described first portrait error is the difference that the user of described incompleteness draws a portrait in matrix between nonzero element and the first corresponding predicted value, and described first predicted value is that the merging entry of a matrix of described customer parameter matrix P and described tag parameter matrix Q is plain.
Wherein, 103, the r time upgrade described customer parameter matrix P and described tag parameter matrix Q, r is more than or equal to 1 and is less than or equal to R.Step 103 comprises:
As r=1, upgrade described customer parameter matrix P and described tag parameter matrix Q, comprising for described the r time:
Draw a portrait matrix according to the user of described incompleteness, the customer parameter matrix P that stochastic generation is initial and tag parameter matrix Q, the user calculating described incompleteness draws a portrait the first portrait error of nonzero element in matrix;
According to described first portrait error, upgrade described customer parameter matrix P and described tag parameter matrix Q;
When r is more than or equal to 2, upgrades described customer parameter matrix P and described tag parameter matrix Q, comprising for described the r time:
103a, to draw a portrait matrix, described customer parameter matrix P and described tag parameter matrix Q according to the user of described incompleteness, calculate the first portrait error of the Part I user in described multiple user; The first change difference of described Part I user is greater than the first change difference of the first remaining users, and described first remaining users is the user except described Part I user in described multiple user.
When r equals 2, described first change difference is the difference according to upgrading for the r-1 time between first predicted value of the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition and the first predicted value obtained according to initial described customer parameter matrix P and the described tag parameter matrix Q of stochastic generation;
When r is greater than 2, described first change difference be according to upgrade for the r-1 time the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition the first predicted value and according to the difference upgraded for the r-2 time between the first predicted value that the described customer parameter matrix P that obtains and described tag parameter matrix Q obtains.
103b, according to the described Part I user's first portrait error, upgrade described customer parameter matrix P and described tag parameter matrix Q;
104, after the R time upgrades described customer parameter matrix P and described tag parameter matrix Q, matrix is drawn a portrait according to the user of described incompleteness, and the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, obtain complete user and draw a portrait matrix.
In embodiments of the present invention, the first change difference of Part I user is greater than the first change difference of the first remaining users, according to the portrait error of Part I user, upgrade customer parameter matrix and tag parameter matrix, the speed of convergence of portrait error can be accelerated, reduce operand, thus reduce the operation time of system.
Preferably, step 103 can also comprise: drawing a portrait matrix, described customer parameter matrix P and described tag parameter matrix Q according to the user of described incompleteness, before calculating the first portrait error of the Part I user in described multiple user, calculate the change difference of described multiple user, and the change difference of described multiple user is sorted, select the user of predetermined ratio as described Part I user, the change difference of the user of described predetermined ratio is all greater than the change difference of non-selected user in described multiple user.Therefore, the embodiment of the present invention can accurately determine Part I user.
As shown in Figure 2, after the R time upgrades described customer parameter matrix P and described tag parameter matrix Q, described method can also comprise:
201, described customer parameter matrix P is carried out S deep decomposition, and obtain customer parameter matrix and the tag parameter matrix of S the level degree of depth; Described is for S time the number of times set;
Wherein, 202, customer parameter matrix P described in the w time deep decomposition, w is more than or equal to 1 and is less than or equal to S.Step 202, comprising:
202a, stochastic generation customer parameter matrix P
wwith tag parameter matrix Q
w; Wherein, described customer parameter matrix P
w, described tag parameter matrix Q
widentical multiple users corresponding to described customer parameter matrix P; Customer parameter matrix P
wwith tag parameter matrix Q
wfor the split-matrix of described customer parameter matrix P.
202b, according to described customer parameter matrix P, Y time upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w, Y is for making the update times needed for the second portrait error convergence; Described second portrait error is the difference in described customer parameter matrix P between nonzero element and the second corresponding predicted value, and described second predicted value is described customer parameter matrix P
wwith described tag parameter matrix Q
wmerging entry of a matrix element.
Wherein, described customer parameter matrix P is upgraded the y time
wwith described tag parameter matrix Q
w, y is more than or equal to 1 and is less than or equal to Y, comprising:
As y=1, upgrade described customer parameter matrix P the y time
wwith described tag parameter matrix Q
w, comprising: according to described customer parameter matrix P, the customer parameter matrix P that stochastic generation is initial
wwith tag parameter matrix Q
w, the user calculating described incompleteness draws a portrait the second portrait error of nonzero element in matrix;
According to described second portrait error, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
When y is more than or equal to 2, upgrade described customer parameter matrix P the y time
wwith described tag parameter matrix Q
w, comprising: according to described customer parameter matrix P, described customer parameter matrix P
wwith described tag parameter matrix Q
w, calculate the second portrait error of the Part II user in described multiple user; The second change difference of described Part II user is greater than the second change difference of the second remaining users, and described second remaining users is the user except described Part II user in described multiple user;
When y equals 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wthe second predicted value obtained with according to the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q
wdifference between the second predicted value obtained;
When y is greater than 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P and tag parameter matrix Q that obtain
wobtain the second predicted value with upgrade according to the y-2 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wdifference between the second predicted value obtained;
According to the second portrait error of described Part II user nonzero element, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
The described customer parameter matrix P obtained is upgraded by Y time
wwith described tag parameter matrix Q
was the customer parameter matrix P of the w level degree of depth
wwith tag parameter matrix Q
w;
When described w is less than described S, upgrade Y time the described customer parameter matrix P obtained
was the w+1 time want the customer parameter matrix P of deep decomposition;
203, described customer parameter matrix P is being carried out S deep decomposition, and after the customer parameter matrix obtaining S the level degree of depth and tag parameter matrix, matrix is drawn a portrait according to the user of described incompleteness, the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, and the customer parameter matrix of S the level degree of depth obtained and tag parameter matrix, obtain complete user and draw a portrait matrix.
Implementation step 201 to 203, after the R time upgrades described customer parameter matrix P and described tag parameter matrix Q, customer parameter matrix P is carried out further multi-level deep decomposition, merges the result of multi-level matrix decomposition, improve the precision of user's portrait.
Below enumerate specific embodiment, the present invention is described further.
The embodiment of the present invention provides a kind of method obtaining whole user portrait, comprising:
Step one: input incomplete portrait matrix r
m, N, and random initializtion customer parameter matrix P and tag parameter matrix Q.
Step 2: the change difference c of computational prediction value between twice circulation t and t-1
m, and it is sorted, select 20% maximum ratio user of change difference to carry out portrait error calculation, the concrete formula calculated is as follows:
c
m,n=|(p
m Tq
n)
t-(p
m Tq
n)
t-1|;
c
m=Σ
nc
m,n。
Step 3: the portrait error e calculating the certain customers' nonzero element selected
m, n, the concrete formula calculated is as follows:
e
m,n=r
m,n-p
m Tq
n。
Step 4: according to portrait error update parameter P and Q, the concrete formula calculated is as follows:
p
m=p
m+γ(e
m,nq
n-λ
Pp
m)。
Step 5: repeat step 2, three and four, step 2, three and four, for once to circulate, cycles through sequence c at every turn
muser's nonzero element that dynamic dispatching scans, but user's nonzero element of all selection parts calculates.
Step 6: repeat step 4 1≤t≤T circulation until customer parameter matrix P and tag parameter matrix Q and portrait error are roughly stabilized in a state, namely until portrait error convergence.
After execution step 6, deep decomposition will be carried out to customer parameter matrix P, draw a portrait precision with the user improving output matrix.
Step 7: to customer parameter matrix P as input matrix, repeats step one to step 6, obtains the portrait matrix P of matrix P
1, Q
1, P
2and Q
2, until P
sand Q
s, the degree of depth s of matrix decomposition is specified by user, usual s=2.
Step 8: the matrix decomposition result merging different levels, exports whole user portrait matrix
Below in conjunction with embody rule example, the present invention is further described.
Such as, the film matrix of user is drawn a portrait matrix, wherein the row representative of consumer of matrix as the user of incompleteness of input, and list shows that user likes film, and other correlation parameters that can provide are as { K=100, λ
p=λ
q=0.05, γ=0.001, s=0, ρ
m=0.2}, K represent class of subscriber number, λ
prepresent vector matrix P={p
madjustment parameter, λ
qrepresent vector matrix Q={q
nadjustment parameter, γ represents learning efficiency, and s represents the deep layer matrix decomposition number of plies, ρ
mrepresent and choose ratio.The present embodiment is only tested and is worked as ρ
mwhen=0.2, the acceleration effect of system operations.In the present embodiment, system, by according to the nonzero element in the matrix of input, calculates the gradient of portrait error, and is decomposed into matrix P and Q, and 20% user is carried out gradient updating.By circulating 100 times, speed is than fast about 2 times of existing matrix decomposition scheme.
Also such as, the music matrix of user draws a portrait matrix as the user of the incompleteness of input, and provides following parameters { K=100, λ
p=λ
q=0.05, γ=0.001, s=2, ρ
m=1}.The number of plies of deep layer matrix decomposition module is decided to be 2 by the present embodiment, circulates to provide user for 100 times and draw a portrait precision 0.81, lower than the error 0.85 of script matrix decomposition scheme, reflects that the precision that user draws a portrait significantly improves.
As shown in Figure 3, the invention provides a kind of signal conditioning package 301, comprising:
Load module 302, draws a portrait matrix for obtaining incomplete user, and incomplete user described in stochastic generation draws a portrait the customer parameter matrix P of matrix and tag parameter matrix Q; Wherein, the user of described incompleteness draws a portrait multiple users corresponding to matrix, described customer parameter matrix P and described tag parameter matrix Q;
Matrix decomposition module 303, matrix is drawn a portrait for obtaining incomplete user at described load module 302, and incomplete user described in stochastic generation draws a portrait after the customer parameter matrix P of matrix and tag parameter matrix Q, matrix is drawn a portrait according to the user of described incompleteness, R the described customer parameter matrix P of renewal and described tag parameter matrix Q, R are for making the update times needed for the first portrait error convergence; Described first portrait error is the difference that the user of described incompleteness draws a portrait in matrix between nonzero element and the first corresponding predicted value, and described first predicted value is that the merging entry of a matrix of described customer parameter matrix P and described tag parameter matrix Q is plain;
Wherein, upgrade described customer parameter matrix P and described tag parameter matrix Q the r time, r is more than or equal to 1 and is less than or equal to R, comprising:
As r=1, upgrade described customer parameter matrix P and described tag parameter matrix Q described the r time, comprise: draw a portrait matrix according to the user of described incompleteness, the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q, the user calculating described incompleteness draws a portrait the first portrait error of nonzero element in matrix; According to described first portrait error, upgrade described customer parameter matrix P and described tag parameter matrix Q;
When r is more than or equal to 2, upgrades described customer parameter matrix P and described tag parameter matrix Q, comprising for described the r time: upgrade described customer parameter matrix P and described tag parameter matrix Q, comprising for described the r time:
Draw a portrait matrix, described customer parameter matrix P and described tag parameter matrix Q according to the user of described incompleteness, calculate the first portrait error of the Part I user in described multiple user; The first change difference of described Part I user is greater than the first change difference of the first remaining users, and described first remaining users is the user except described Part I user in described multiple user;
When r equals 2, described first change difference is the difference according to upgrading for the r-1 time between first predicted value of the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition and the first predicted value obtained according to initial described customer parameter matrix P and the described tag parameter matrix Q of stochastic generation;
When r is greater than 2, described first change difference be according to upgrade for the r-1 time the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition the first predicted value and according to the difference upgraded for the r-2 time between the first predicted value that the described customer parameter matrix P that obtains and described tag parameter matrix Q obtains;
According to the first portrait error of described Part I user, upgrade described customer parameter matrix P and described tag parameter matrix Q;
Output module 304, after upgrading described customer parameter matrix P and described tag parameter matrix Q the R time at described matrix decomposition module 303, matrix is drawn a portrait according to the user of described incompleteness, and the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, obtain complete user and draw a portrait matrix.
Preferably, as shown in Figure 4, described matrix decomposition module 303 in signal conditioning package 301 of the present invention also comprises dynamic dispatching module 305, described dynamic dispatching module 305 is for drawing a portrait matrix according to the user of described incompleteness, described customer parameter matrix P and described tag parameter matrix Q, before calculating the first portrait error of the Part I user in described multiple user, calculate the change difference of described multiple user, and the change difference of described multiple user is sorted, select the user of predetermined ratio as described Part I user, the change difference of the user of described predetermined ratio is all greater than the change difference of non-selected user in described multiple user.
Preferably, described matrix decomposition module 303 also comprises deep decomposition module 306, described deep decomposition module 306 is for after the R time upgrades described customer parameter matrix P and described tag parameter matrix Q, described customer parameter matrix P is carried out S deep decomposition, and obtains customer parameter matrix and the tag parameter matrix of S the level degree of depth; Described is for S time the number of times set;
Wherein, customer parameter matrix P described in the w time deep decomposition, w is more than or equal to 1 and is less than or equal to S, comprising:
Stochastic generation customer parameter matrix P
wwith tag parameter matrix Q
w; Wherein, described customer parameter matrix P
w, described tag parameter matrix Q
widentical multiple users corresponding to described customer parameter matrix P; Customer parameter matrix P
wwith tag parameter matrix Q
wfor the split-matrix of described customer parameter matrix P;
According to described customer parameter matrix P, upgrade described customer parameter matrix P Y time
wwith described tag parameter matrix Q
w, Y is for making the update times needed for the second portrait error convergence; Described second portrait error is the difference in described customer parameter matrix P between nonzero element and the second corresponding predicted value, and described second predicted value is described customer parameter matrix P
wwith described tag parameter matrix Q
wmerging entry of a matrix element;
Wherein, described customer parameter matrix P is upgraded the y time
wwith described tag parameter matrix Q
w, y is more than or equal to 1 and is less than or equal to Y, comprising:
As y=1, according to described customer parameter matrix P, the customer parameter matrix P that stochastic generation is initial
wwith tag parameter matrix Q
w, the user calculating described incompleteness draws a portrait the second portrait error of nonzero element in matrix;
According to described second portrait error, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
When y is more than or equal to 2, according to described customer parameter matrix P, described customer parameter matrix P
wwith described tag parameter matrix Q
w, calculate the second portrait error of the Part II user in described multiple user; The second change difference of described Part II user is greater than the second change difference of the second remaining users, and described second remaining users is the user except described Part II user in described multiple user;
When y equals 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wthe second predicted value obtained with according to the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q
wdifference between the second predicted value obtained;
When y is greater than 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P and tag parameter matrix Q that obtain
wobtain the second predicted value with upgrade according to the y-2 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wdifference between the second predicted value obtained;
According to the second portrait error of described Part II user nonzero element, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
The described customer parameter matrix P obtained is upgraded by Y time
wwith described tag parameter matrix Q
was the customer parameter matrix P of the w level degree of depth
wwith tag parameter matrix Q
w;
When described w is less than described S, upgrade Y time the described customer parameter matrix P obtained
was the w+1 time want the customer parameter matrix P of deep decomposition;
Described output module 304 is also for carrying out S deep decomposition at described matrix decomposition module 303 by described customer parameter matrix P, and after the customer parameter matrix obtaining S the level degree of depth and tag parameter matrix, matrix is drawn a portrait according to the user of described incompleteness, the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, and the customer parameter matrix of S the level degree of depth obtained and tag parameter matrix, obtain complete user and draw a portrait matrix.
As shown in Figure 5, the invention provides a kind of signal conditioning package 401, comprising: input interface 402, processor 403 and output interface 404, described processor connects input interface and output interface respectively.
Described processor 403 draws a portrait matrix for being obtained incomplete user by input interface 402, and incomplete user described in stochastic generation draws a portrait the customer parameter matrix P of matrix and tag parameter matrix Q; Wherein, the user of described incompleteness draws a portrait multiple users corresponding to matrix, described customer parameter matrix P and described tag parameter matrix Q;
And for: obtain incomplete user at described load module and draw a portrait matrix, and incomplete user described in stochastic generation draws a portrait after the customer parameter matrix P of matrix and tag parameter matrix Q, matrix is drawn a portrait according to the user of described incompleteness, R the described customer parameter matrix P of renewal and described tag parameter matrix Q, R are for making the update times needed for the first portrait error convergence; Described first portrait error is the difference that the user of described incompleteness draws a portrait in matrix between nonzero element and the first corresponding predicted value, and described first predicted value is that the merging entry of a matrix of described customer parameter matrix P and described tag parameter matrix Q is plain;
Wherein, upgrade described customer parameter matrix P and described tag parameter matrix Q the r time, r is more than or equal to 1 and is less than or equal to R, comprising:
As r=1, upgrade described customer parameter matrix P and described tag parameter matrix Q described the r time, comprise: draw a portrait matrix according to the user of described incompleteness, the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q, the user calculating described incompleteness draws a portrait the first portrait error of nonzero element in matrix; According to described first portrait error, upgrade described customer parameter matrix P and described tag parameter matrix Q;
When r is more than or equal to 2, upgrades described customer parameter matrix P and described tag parameter matrix Q, comprising for described the r time: upgrade described customer parameter matrix P and described tag parameter matrix Q, comprising for described the r time:
Draw a portrait matrix, described customer parameter matrix P and described tag parameter matrix Q according to the user of described incompleteness, calculate the first portrait error of the Part I user in described multiple user; The first change difference of described Part I user is greater than the first change difference of the first remaining users, and described first remaining users is the user except described Part I user in described multiple user;
When r equals 2, described first change difference is the difference according to upgrading for the r-1 time between first predicted value of the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition and the first predicted value obtained according to initial described customer parameter matrix P and the described tag parameter matrix Q of stochastic generation;
When r is greater than 2, described first change difference be according to upgrade for the r-1 time the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition the first predicted value and according to the difference upgraded for the r-2 time between the first predicted value that the described customer parameter matrix P that obtains and described tag parameter matrix Q obtains;
According to the first portrait error of described Part I user, upgrade described customer parameter matrix P and described tag parameter matrix Q;
Described processor 403 is for after the R time upgrades described customer parameter matrix P and described tag parameter matrix Q, matrix is drawn a portrait according to the user of described incompleteness, and the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, obtain complete user and draw a portrait matrix.
Described processor 403 can export complete user by output interface 404 and draw a portrait matrix.
Preferably, described processor 403 is also for drawing a portrait matrix, described customer parameter matrix P and described tag parameter matrix Q according to the user of described incompleteness, before calculating the first portrait error of the Part I user in described multiple user, calculate the change difference of described multiple user, and the change difference of described multiple user is sorted, select the user of predetermined ratio as described Part I user, the change difference of the user of described predetermined ratio is all greater than the change difference of non-selected user in described multiple user.
Preferably, described processor 403 is also for after the R time upgrades described customer parameter matrix P and described tag parameter matrix Q, described customer parameter matrix P is carried out S deep decomposition, and obtains customer parameter matrix and the tag parameter matrix of S the level degree of depth; Described is for S time the number of times set;
Wherein, customer parameter matrix P described in the w time deep decomposition, w is more than or equal to 1 and is less than or equal to S, comprising:
Stochastic generation customer parameter matrix P
wwith tag parameter matrix Q
w; Wherein, described customer parameter matrix P
w, described tag parameter matrix Q
widentical multiple users corresponding to described customer parameter matrix P; Customer parameter matrix P
wwith tag parameter matrix Q
wfor the split-matrix of described customer parameter matrix P;
According to described customer parameter matrix P, upgrade described customer parameter matrix P Y time
wwith described tag parameter matrix Q
w, Y is for making the update times needed for the second portrait error convergence; Described second portrait error is the difference in described customer parameter matrix P between nonzero element and the second corresponding predicted value, and described second predicted value is described customer parameter matrix P
wwith described tag parameter matrix Q
wmerging entry of a matrix element;
Wherein, described customer parameter matrix P is upgraded the y time
wwith described tag parameter matrix Q
w, y is more than or equal to 1 and is less than or equal to Y, comprising:
As y=1, upgrade described customer parameter matrix P the y time
wwith described tag parameter matrix Q
w, comprising: according to described customer parameter matrix P, the customer parameter matrix P that stochastic generation is initial
wwith tag parameter matrix Q
w, the user calculating described incompleteness draws a portrait the second portrait error of nonzero element in matrix;
According to described second portrait error, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
When y is more than or equal to 2, upgrade described customer parameter matrix P the y time
wwith described tag parameter matrix Q
w, comprising:
According to described customer parameter matrix P, described customer parameter matrix P
wwith described tag parameter matrix Q
w, calculate the second portrait error of the Part II user in described multiple user; The second change difference of described Part II user is greater than the second change difference of the second remaining users, and described second remaining users is the user except described Part II user in described multiple user;
When y equals 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wthe second predicted value obtained with according to the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q
wdifference between the second predicted value obtained;
When y is greater than 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P and tag parameter matrix Q that obtain
wobtain the second predicted value with upgrade according to the y-2 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wdifference between the second predicted value obtained;
According to the second portrait error of described Part II user nonzero element, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
The described customer parameter matrix P obtained is upgraded by Y time
wwith described tag parameter matrix Q
was the customer parameter matrix P of the w level degree of depth
wwith tag parameter matrix Q
w;
When described w is less than described S, upgrade Y time the described customer parameter matrix P obtained
was the w+1 time want the customer parameter matrix P of deep decomposition;
Described processor 403 is also for carrying out S deep decomposition by described customer parameter matrix P, and after the customer parameter matrix obtaining S the level degree of depth and tag parameter matrix, matrix is drawn a portrait according to the user of described incompleteness, the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, and the customer parameter matrix of S the level degree of depth obtained and tag parameter matrix, obtain complete user and draw a portrait matrix.
One of ordinary skill in the art will appreciate that all or part of step in the various methods of above-described embodiment is that the hardware that can carry out instruction relevant by program has come, this program can be stored in a computer-readable recording medium, storage medium can comprise: ROM (read-only memory) (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), disk or CD etc.
Above to the embodiment of the present invention provide a kind of obtain whole user portrait method and device be described in detail, apply specific case herein to set forth principle of the present invention and embodiment, the explanation of above embodiment just understands method of the present invention and core concept thereof for helping; Meanwhile, for those skilled in the art, according to thought of the present invention, all will change in specific embodiments and applications, in sum, this description should not be construed as limitation of the present invention.
Claims (6)
1. obtain a method for whole user portrait, it is characterized in that, comprising:
Obtain incomplete user and draw a portrait matrix, and the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q; Wherein, the user of described incompleteness draws a portrait matrix, multiple users that described customer parameter matrix P is corresponding identical with described tag parameter matrix Q, and described customer parameter matrix P and described tag parameter matrix Q is that the user of described incompleteness draws a portrait the split-matrix of matrix;
Drawing a portrait matrix according to the user of described incompleteness, upgrading described customer parameter matrix P and described tag parameter matrix Q, R for R time for making the update times needed for the first portrait error convergence; Described first portrait error is the difference that the user of described incompleteness draws a portrait in matrix between nonzero element and the first corresponding predicted value, and described first predicted value is that the merging entry of a matrix of described customer parameter matrix P and described tag parameter matrix Q is plain;
Wherein, upgrade described customer parameter matrix P and described tag parameter matrix Q the r time, r is more than or equal to 1 and is less than or equal to R, comprising:
As r=1, draw a portrait matrix, the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q according to the user of described incompleteness, the user calculating described incompleteness draws a portrait the first portrait error of nonzero element in matrix; According to described first portrait error, upgrade described customer parameter matrix P and described tag parameter matrix Q;
When r is more than or equal to 2, upgrades described customer parameter matrix P and described tag parameter matrix Q, comprising for described the r time:
Draw a portrait matrix, described customer parameter matrix P and described tag parameter matrix Q according to the user of described incompleteness, calculate the first portrait error of the Part I user in described multiple user; The first change difference of described Part I user is greater than the first change difference of the first remaining users, and described first remaining users is the user except described Part I user in described multiple user;
When r equals 2, described first change difference is the difference according to upgrading for the r-1 time between first predicted value of the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition and the first predicted value obtained according to initial described customer parameter matrix P and the described tag parameter matrix Q of stochastic generation;
When r is greater than 2, described first change difference be according to upgrade for the r-1 time the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition the first predicted value and according to the difference upgraded for the r-2 time between the first predicted value that the described customer parameter matrix P that obtains and described tag parameter matrix Q obtains;
According to the first portrait error of described Part I user, upgrade described customer parameter matrix P and described tag parameter matrix Q;
Upgrade after described customer parameter matrix P and described tag parameter matrix Q at the R time, draw a portrait matrix and the described customer parameter matrix P of the R time renewal and described tag parameter matrix Q according to the user of described incompleteness, obtain complete user and draw a portrait matrix.
2. the method for acquisition whole user portrait according to claim 1, is characterized in that, upgrade described customer parameter matrix P and described tag parameter matrix Q described the r time, r is more than or equal to 1 and is less than or equal to R, comprising:
Matrix, described customer parameter matrix P and described tag parameter matrix Q is being drawn a portrait according to the user of described incompleteness, before calculating the first portrait error of the Part I user in described multiple user, calculate the change difference of described multiple user, and the change difference of described multiple user is sorted, select the user of predetermined ratio as described Part I user, the change difference of the user of described predetermined ratio is all greater than the change difference of non-selected user in described multiple user.
3. the method for acquisition whole user portrait according to claim 1 and 2, is characterized in that, after the R time upgrades described customer parameter matrix P and described tag parameter matrix Q, described method also comprises:
Described customer parameter matrix P is carried out S deep decomposition, and obtains customer parameter matrix and the tag parameter matrix of S the level degree of depth; Described is for S time the number of times set;
Wherein, customer parameter matrix P described in the w time deep decomposition, w is more than or equal to 1 and is less than or equal to S, comprising:
Stochastic generation customer parameter matrix P
wwith tag parameter matrix Q
w; Wherein, described customer parameter matrix P
w, described tag parameter matrix Q
widentical multiple users corresponding to described customer parameter matrix P; Customer parameter matrix P
wwith tag parameter matrix Q
wfor the split-matrix of described customer parameter matrix P;
According to described customer parameter matrix P, upgrade described customer parameter matrix P Y time
wwith described tag parameter matrix Q
w, Y is for making the update times needed for the second portrait error convergence; Described second portrait error is the difference in described customer parameter matrix P between nonzero element and the second corresponding predicted value, and described second predicted value is described customer parameter matrix P
wwith described tag parameter matrix Q
wmerging entry of a matrix element;
Wherein, described customer parameter matrix P is upgraded the y time
wwith described tag parameter matrix Q
w, y is more than or equal to 1 and is less than or equal to Y, comprising:
As y=1, according to described customer parameter matrix P, the customer parameter matrix P that stochastic generation is initial
wwith tag parameter matrix Q
w, the user calculating described incompleteness draws a portrait the second portrait error of nonzero element in matrix;
According to described second portrait error, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
When y is more than or equal to 2, according to described customer parameter matrix P, described customer parameter matrix P
wwith described tag parameter matrix Q
w, calculate the second portrait error of the Part II user in described multiple user; The second change difference of described Part II user is greater than the second change difference of the second remaining users, and described second remaining users is the user except described Part II user in described multiple user;
When y equals 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wthe second predicted value obtained with according to the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q
wdifference between the second predicted value obtained;
When y is greater than 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P and tag parameter matrix Q that obtain
wobtain the second predicted value with upgrade according to the y-2 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wdifference between the second predicted value obtained;
According to the second portrait error of described Part II user nonzero element, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
The described customer parameter matrix P obtained is upgraded by Y time
wwith described tag parameter matrix Q
was the customer parameter matrix P of the w level degree of depth
wwith tag parameter matrix Q
w;
When described w is less than described S, upgrade Y time the described customer parameter matrix P obtained
was the w+1 time want the customer parameter matrix P of deep decomposition;
The described user according to described incompleteness draws a portrait matrix, and the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, obtains complete user and draws a portrait matrix, comprising:
Described customer parameter matrix P is being carried out S deep decomposition, and after the customer parameter matrix obtaining S the level degree of depth and tag parameter matrix, matrix is drawn a portrait according to the user of described incompleteness, the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, and the customer parameter matrix of S the level degree of depth obtained and tag parameter matrix, obtain complete user and draw a portrait matrix.
4. a signal conditioning package, is characterized in that, comprising:
Load module, draws a portrait matrix for obtaining incomplete user, and incomplete user described in stochastic generation draws a portrait the customer parameter matrix P of matrix and tag parameter matrix Q; Wherein, the user of described incompleteness draws a portrait multiple users corresponding to matrix, described customer parameter matrix P and described tag parameter matrix Q;
Matrix decomposition module, for: obtain incomplete user at described load module and draw a portrait matrix, and incomplete user described in stochastic generation draws a portrait after the customer parameter matrix P of matrix and tag parameter matrix Q, matrix is drawn a portrait according to the user of described incompleteness, R the described customer parameter matrix P of renewal and described tag parameter matrix Q, R are for making the update times needed for the first portrait error convergence; Described first portrait error is the difference that the user of described incompleteness draws a portrait in matrix between nonzero element and the first corresponding predicted value, and described first predicted value is that the merging entry of a matrix of described customer parameter matrix P and described tag parameter matrix Q is plain;
Wherein, upgrade described customer parameter matrix P and described tag parameter matrix Q the r time, r is more than or equal to 1 and is less than or equal to R, comprising:
As r=1, draw a portrait matrix, the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q according to the user of described incompleteness, the user calculating described incompleteness draws a portrait the first portrait error of nonzero element in matrix; According to described first portrait error, upgrade described customer parameter matrix P and described tag parameter matrix Q;
When r is more than or equal to 2, upgrades described customer parameter matrix P and described tag parameter matrix Q, comprising for described the r time:
Draw a portrait matrix, described customer parameter matrix P and described tag parameter matrix Q according to the user of described incompleteness, calculate the first portrait error of the Part I user in described multiple user; The first change difference of described Part I user is greater than the first change difference of the first remaining users, and described first remaining users is the user except described Part I user in described multiple user;
When r equals 2, described first change difference is the difference according to upgrading for the r-1 time between first predicted value of the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition and the first predicted value obtained according to initial described customer parameter matrix P and the described tag parameter matrix Q of stochastic generation;
When r is greater than 2, described first change difference be according to upgrade for the r-1 time the described customer parameter matrix P that obtains and described tag parameter matrix Q acquisition the first predicted value and according to the difference upgraded for the r-2 time between the first predicted value that the described customer parameter matrix P that obtains and described tag parameter matrix Q obtains;
According to the first portrait error of described Part I user, upgrade described customer parameter matrix P and described tag parameter matrix Q;
Output module, for after the R time upgrades described customer parameter matrix P and described tag parameter matrix Q, draw a portrait matrix according to the user of described incompleteness, and the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, obtain complete user and draw a portrait matrix.
5. device according to claim 4, it is characterized in that, described matrix decomposition module comprises dynamic dispatching module, described dynamic dispatching module is used for drawing a portrait matrix according to the user of described incompleteness, described customer parameter matrix P and described tag parameter matrix Q, before calculating the first portrait error of the Part I user in described multiple user, calculate the change difference of described multiple user, and the change difference of described multiple user is sorted, select the user of predetermined ratio as described Part I user, the change difference of the user of described predetermined ratio is all greater than the change difference of non-selected user in described multiple user.
6. the device according to claim 4 or 5, it is characterized in that, described matrix decomposition module also comprises deep decomposition module, described deep decomposition module is used for: after the R time upgrades described customer parameter matrix P and described tag parameter matrix Q, described customer parameter matrix P is carried out S deep decomposition, and obtains customer parameter matrix and the tag parameter matrix of S the level degree of depth; Described is for S time the number of times set;
Wherein, customer parameter matrix P described in the w time deep decomposition, w is more than or equal to 1 and is less than or equal to S, comprising:
Stochastic generation customer parameter matrix P
wwith tag parameter matrix Q
w; Wherein, described customer parameter matrix P
w, described tag parameter matrix Q
widentical multiple users corresponding to described customer parameter matrix P; Customer parameter matrix P
wwith tag parameter matrix Q
wfor the split-matrix of described customer parameter matrix P;
According to described customer parameter matrix P, upgrade described customer parameter matrix P Y time
wwith described tag parameter matrix Q
w, Y is for making the update times needed for the second portrait error convergence; Described second portrait error is the difference in described customer parameter matrix P between nonzero element and the second corresponding predicted value, and described second predicted value is described customer parameter matrix P
wwith described tag parameter matrix Q
wmerging entry of a matrix element;
Wherein, described customer parameter matrix P is upgraded the y time
wwith described tag parameter matrix Q
w, y is more than or equal to 1 and is less than or equal to Y, comprising:
As y=1, according to described customer parameter matrix P, the customer parameter matrix P that stochastic generation is initial
wwith tag parameter matrix Q
w, the user calculating described incompleteness draws a portrait the second portrait error of nonzero element in matrix;
According to described second portrait error, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
When y is more than or equal to 2, according to described customer parameter matrix P, described customer parameter matrix P
wwith described tag parameter matrix Q
w, calculate the second portrait error of the Part II user in described multiple user; The second change difference of described Part II user is greater than the second change difference of the second remaining users, and described second remaining users is the user except described Part II user in described multiple user;
When y equals 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wthe second predicted value obtained with according to the initial customer parameter matrix P of stochastic generation and tag parameter matrix Q
wdifference between the second predicted value obtained;
When y is greater than 2, described second change difference is upgrade according to the y-1 time the customer parameter matrix P and tag parameter matrix Q that obtain
wobtain the second predicted value with upgrade according to the y-2 time the customer parameter matrix P obtained
wwith tag parameter matrix Q
wdifference between the second predicted value obtained;
According to the second portrait error of described Part II user nonzero element, upgrade described customer parameter matrix P
wwith described tag parameter matrix Q
w;
The described customer parameter matrix P obtained is upgraded by Y time
wwith described tag parameter matrix Q
was the customer parameter matrix P of the w level degree of depth
wwith tag parameter matrix Q
w;
When described w is less than described S, upgrade Y time the described customer parameter matrix P obtained
was the w+1 time want the customer parameter matrix P of deep decomposition;
Described output module is also for carrying out S deep decomposition by described customer parameter matrix P, and after the customer parameter matrix obtaining S the level degree of depth and tag parameter matrix, matrix is drawn a portrait according to the user of described incompleteness, the described customer parameter matrix P upgraded for the R time and described tag parameter matrix Q, and the customer parameter matrix of S the level degree of depth obtained and tag parameter matrix, obtain complete user and draw a portrait matrix.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310745711.7A CN104750731B (en) | 2013-12-30 | 2013-12-30 | A kind of method and device obtaining whole user portrait |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310745711.7A CN104750731B (en) | 2013-12-30 | 2013-12-30 | A kind of method and device obtaining whole user portrait |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104750731A true CN104750731A (en) | 2015-07-01 |
CN104750731B CN104750731B (en) | 2018-09-21 |
Family
ID=53590430
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310745711.7A Active CN104750731B (en) | 2013-12-30 | 2013-12-30 | A kind of method and device obtaining whole user portrait |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104750731B (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105893406A (en) * | 2015-11-12 | 2016-08-24 | 乐视云计算有限公司 | Group user profiling method and system |
CN106489159A (en) * | 2016-06-29 | 2017-03-08 | 深圳狗尾草智能科技有限公司 | A kind of user's portrait based on deep neural network represents learning system and method |
CN106503015A (en) * | 2015-09-07 | 2017-03-15 | 国家计算机网络与信息安全管理中心 | A kind of method for building user's portrait |
WO2017186106A1 (en) * | 2016-04-29 | 2017-11-02 | 腾讯科技(深圳)有限公司 | Method and device for acquiring user portrait |
CN107480271A (en) * | 2017-08-18 | 2017-12-15 | 晶赞广告(上海)有限公司 | Crowd portrayal method and system with index search are searched based on sampling |
CN107644047A (en) * | 2016-07-22 | 2018-01-30 | 华为技术有限公司 | Tag Estimation generation method and device |
CN108846097A (en) * | 2018-06-15 | 2018-11-20 | 北京搜狐新媒体信息技术有限公司 | The interest tags representation method of user, article recommended method and device, equipment |
CN111322716A (en) * | 2020-02-24 | 2020-06-23 | 青岛海尔工业智能研究院有限公司 | Air conditioner temperature automatic setting method, air conditioner, equipment and storage medium |
CN111522828A (en) * | 2020-04-23 | 2020-08-11 | 中国农业银行股份有限公司 | User portrait label value analysis method and device |
CN111523026A (en) * | 2020-04-15 | 2020-08-11 | 咪咕文化科技有限公司 | User portrait updating method, system, network equipment and storage medium |
CN111931107A (en) * | 2020-07-31 | 2020-11-13 | 上海博泰悦臻电子设备制造有限公司 | Digital citizen system construction method, system and storage medium |
CN117440192A (en) * | 2023-12-21 | 2024-01-23 | 辽宁云科智造产业技术研究院有限公司 | User demand analysis method and system based on intelligent cloud service platform |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090112859A1 (en) * | 2007-10-25 | 2009-04-30 | Dehlinger Peter J | Citation-based information retrieval system and method |
CN103345474A (en) * | 2013-07-25 | 2013-10-09 | 苏州大学 | Online tracking method for document theme |
-
2013
- 2013-12-30 CN CN201310745711.7A patent/CN104750731B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090112859A1 (en) * | 2007-10-25 | 2009-04-30 | Dehlinger Peter J | Citation-based information retrieval system and method |
CN103345474A (en) * | 2013-07-25 | 2013-10-09 | 苏州大学 | Online tracking method for document theme |
Non-Patent Citations (1)
Title |
---|
叶芸: "主题模型的在线消息传递算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106503015A (en) * | 2015-09-07 | 2017-03-15 | 国家计算机网络与信息安全管理中心 | A kind of method for building user's portrait |
CN105893406A (en) * | 2015-11-12 | 2016-08-24 | 乐视云计算有限公司 | Group user profiling method and system |
WO2017186106A1 (en) * | 2016-04-29 | 2017-11-02 | 腾讯科技(深圳)有限公司 | Method and device for acquiring user portrait |
CN107341679A (en) * | 2016-04-29 | 2017-11-10 | 腾讯科技(深圳)有限公司 | Obtain the method and device of user's portrait |
US11394798B2 (en) | 2016-04-29 | 2022-07-19 | Tencent Technology (Shenzhen) Company Limited | User portrait obtaining method, apparatus, and storage medium according to user behavior log records on features of articles |
CN106489159A (en) * | 2016-06-29 | 2017-03-08 | 深圳狗尾草智能科技有限公司 | A kind of user's portrait based on deep neural network represents learning system and method |
CN107644047B (en) * | 2016-07-22 | 2021-01-15 | 华为技术有限公司 | Label prediction generation method and device |
CN107644047A (en) * | 2016-07-22 | 2018-01-30 | 华为技术有限公司 | Tag Estimation generation method and device |
CN107480271A (en) * | 2017-08-18 | 2017-12-15 | 晶赞广告(上海)有限公司 | Crowd portrayal method and system with index search are searched based on sampling |
CN107480271B (en) * | 2017-08-18 | 2020-09-18 | 晶赞广告(上海)有限公司 | Crowd image drawing method and system based on sampling search and index search |
CN108846097B (en) * | 2018-06-15 | 2021-01-29 | 北京搜狐新媒体信息技术有限公司 | User interest tag representation method, article recommendation device and equipment |
CN108846097A (en) * | 2018-06-15 | 2018-11-20 | 北京搜狐新媒体信息技术有限公司 | The interest tags representation method of user, article recommended method and device, equipment |
CN111322716A (en) * | 2020-02-24 | 2020-06-23 | 青岛海尔工业智能研究院有限公司 | Air conditioner temperature automatic setting method, air conditioner, equipment and storage medium |
CN111523026A (en) * | 2020-04-15 | 2020-08-11 | 咪咕文化科技有限公司 | User portrait updating method, system, network equipment and storage medium |
CN111523026B (en) * | 2020-04-15 | 2023-10-17 | 咪咕文化科技有限公司 | User portrait updating method, system, network equipment and storage medium |
CN111522828A (en) * | 2020-04-23 | 2020-08-11 | 中国农业银行股份有限公司 | User portrait label value analysis method and device |
CN111522828B (en) * | 2020-04-23 | 2023-08-01 | 中国农业银行股份有限公司 | User portrait tag value analysis method and device |
CN111931107A (en) * | 2020-07-31 | 2020-11-13 | 上海博泰悦臻电子设备制造有限公司 | Digital citizen system construction method, system and storage medium |
CN111931107B (en) * | 2020-07-31 | 2024-03-22 | 博泰车联网科技(上海)股份有限公司 | Digital citizen system construction method, system and storage medium |
CN117440192A (en) * | 2023-12-21 | 2024-01-23 | 辽宁云科智造产业技术研究院有限公司 | User demand analysis method and system based on intelligent cloud service platform |
CN117440192B (en) * | 2023-12-21 | 2024-02-23 | 辽宁云科智造产业技术研究院有限公司 | User demand analysis method and system based on intelligent cloud service platform |
Also Published As
Publication number | Publication date |
---|---|
CN104750731B (en) | 2018-09-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104750731A (en) | Method and device for obtaining complete user portrait | |
KR102170105B1 (en) | Method and apparatus for generating neural network structure, electronic device, storage medium | |
US11468366B2 (en) | Parallel development and deployment for machine learning models | |
CN110287324B (en) | Data dynamic labeling method and device for coarse-grained text classification | |
Katoen et al. | The ins and outs of the probabilistic model checker MRMC | |
KR20180091043A (en) | Method and apparatus for obtaining user portraits | |
CN110366734A (en) | Optimization neural network framework | |
US9804580B2 (en) | Feasible tracking control of machine | |
Lambert et al. | Methods for optimum and near optimum disassembly sequencing | |
US20170178163A1 (en) | Method and system for generating a split questionnaire | |
US20140136460A1 (en) | Multi objective design selection | |
CN106156857B (en) | The method and apparatus of the data initialization of variation reasoning | |
US20200334557A1 (en) | Chained influence scores for improving synthetic data generation | |
CN113673201A (en) | Text representation vector generation method and device, storage medium and electronic equipment | |
CN104809098A (en) | Method and device for determining statistical model parameter based on expectation-maximization algorithm | |
CN116304699A (en) | Critical sample set generation method and system based on new energy multi-station short circuit ratio | |
CN115758222A (en) | Signal category identification method and device, electronic equipment and storage medium | |
CN112433952B (en) | Method, system, device and medium for testing fairness of deep neural network model | |
KR102454317B1 (en) | Augmenting virtual users and items in collaborative filtering for addressing cold start problems | |
CN114462581A (en) | Network structure searching method and device | |
CN109582295B (en) | Data processing method and device, storage medium and processor | |
US7650579B2 (en) | Model correspondence method and device | |
CN113609720B (en) | Master-slave degree of freedom processing method, device and storage medium for finite element analysis | |
JP7026922B1 (en) | Information processing equipment, methods, programs and systems | |
CN111179037B (en) | Financial product recommendation method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |