CN104090892A - Method and device for on-line computing of off-line algorithm - Google Patents

Abstract

The invention discloses a method and device for on-line computing of an off-line algorithm, and belongs to the field of Internet. The method comprises the following steps: obtaining a history record corresponding to each service information operated by a user within a first time period, wherein the first time period is a time period which is the time period closest to the present and is in a preset time length; building a first user interest matrix according to the history record corresponding to each service information; according to the first user interest matrix, correcting a second user interest matrix which is stored to obtain a third user interest matrix; according to a preset service information matrix and the third user interest matrix, computing an on-line service recommending matrix. The device comprises an obtaining module, a building module, a correcting module and a computing module. Through the adoption of the method and device, disclosed by the invention, the accuracy of the recommended service information can be improved.

Description

A kind of method that off-line algorithm is calculated online and device

Technical field

The present invention relates to field of Internet communication, particularly a kind of method that off-line algorithm is calculated online and device.

Background technology

At present, user browses dissimilar business information on the network of being everlasting, as video, advertisement and merchandise news etc., the business information that server can be browsed in the past according to user, the interested business information of the current possibility of predictive user, and the business information of prediction is recommended to user.

Server recommends the process of business information as follows:

Server is the in the situation that of off-line, first obtain user from current recently and historical record corresponding to each business information that operate in the time period for default duration of duration, the corresponding historical record of this business information at least comprises the sign of this business information, operates the running time of operation behavior type and this business information of operation of this business information.According to historical record calculated off-line corresponding to each business information, go out this user's user interest matrix B _k*N, user interest matrix B _k*Na kind of attribute factor of the corresponding business information of every a line, each is listed as corresponding a kind of operation behavior, user interest matrix B _k*Nin element B _ijrepresent that the interest value of operation behavior j occurs attribute factor i this user, and be kept at the user's who calculates under off-line case user interest matrix B _k*N.Then server is every the time of default duration, this user's of calculated off-line user interest matrix as stated above, and by the user interest matrix update of preserving, be the user interest matrix of new calculating.

When needs are recommended business information to this user, obtain default business information matrix A _m*K, business information matrix A _m*Kthe corresponding business information of every a line, each is listed as a kind of attribute factor of corresponding business information, business information matrix A _m*Kin elements A _ijrepresent the weight of attribute factor j in business information i.According to the business information matrix A of obtaining _m*Kthis user's who obtains with the calculated off-line of preserving user interest matrix B _k*N, calculate offline business and recommend matrix W _m*N, offline business is recommended matrix W _m*Nthe corresponding business information of every a line, each be listed as corresponding a kind of operation behavior, offline business recommendation matrix W _m*Nin element W _ijrepresent that the level of interest of operation behavior j occurs business information i this user.According to offline business, recommend matrix W _m*Nobtain several business information that the level of interest maximum of operation behavior occurs user, using the business information of obtaining as user is current may interested business information and recommend user.

In realizing process of the present invention, inventor finds prior art, and at least there are the following problems:

Server upgrades once this user's user interest matrix every the time of default duration, and default duration is often two weeks or one month etc.And user is constantly to change to the interest level of business information, the larger business information of level of interest that the offline business so obtaining according to user interest matrix recommends matrix to comprise may not be the interested business information of active user, thereby has reduced the accuracy of recommending business information.

Summary of the invention

In order to improve the accuracy of recommending business information, the invention provides a kind of method that off-line algorithm is calculated online and device.Described technical scheme is as follows:

The method that off-line algorithm is calculated online, described method comprises:

Obtain historical record corresponding to each business information that user operates in very first time section, described very first time section be from current recently and duration be default duration time period;

According to historical record corresponding to described each business information, build first user interests matrix;

According to described first user interests matrix, the second user interest matrix that preservation calculated off-line is obtained is revised, and obtains the 3rd user interest matrix;

According to default business information matrix and described the 3rd user interest matrix, calculate online business recommended matrix.

The device that off-line algorithm is calculated online, described device comprises:

Acquisition module, historical record corresponding to each business information operating in very first time section for obtaining user, described very first time section be from current recently and duration be default duration time period;

Build module, for according to historical record corresponding to described each business information, build first user interests matrix;

Correcting module, for according to described first user interests matrix, revises preserving the second user interest matrix that calculated off-line obtains, and obtains the 3rd user interest matrix;

Computing module, for according to default business information matrix and described the 3rd user interest matrix, calculates online business recommended matrix.

In embodiments of the present invention, obtain historical record corresponding to each business information that user operates in very first time section, very first time section be from current recently and duration be default duration time period; The historical record corresponding according to each business information, builds first user interests matrix; According to first user interests matrix, the second user interest matrix that preservation calculated off-line is obtained is revised, and obtains the 3rd user interest matrix; According to default business information matrix and the 3rd user interest matrix, calculate online business recommended matrix.Due to historical record corresponding to each business information operating in very first time section according to user, built first user interests matrix, and utilize first user interests matrix, the second user interest matrix that preservation calculated off-line is obtained is revised, obtain the 3rd user interest matrix, the larger business information of level of interest that the online business recommended matrix so obtaining according to the 3rd user interest matrix the comprises current interested business information of being more close to the users, thus the accuracy of recommending business information improved.

Accompanying drawing explanation

Fig. 1 is a kind of process flow diagram that off-line algorithm is calculated online that the embodiment of the present invention 1 provides;

Fig. 2 is a kind of process flow diagram that off-line algorithm is calculated online that the embodiment of the present invention 2 provides;

Fig. 3 is a kind of apparatus structure schematic diagram that off-line algorithm is calculated online that the embodiment of the present invention 3 provides.

Embodiment

For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with accompanying drawing, embodiment of the present invention is described further in detail.

Embodiment 1

Referring to Fig. 1, the embodiment of the present invention provides a kind of method that off-line algorithm is calculated online, comprising:

Step 101: obtain historical record corresponding to each business information that user operates in very first time section, very first time section be from current recently and duration be default duration time period;

Step 102: the historical record corresponding according to each business information, builds first user interests matrix;

Step 103: according to first user interests matrix, the second user interest matrix that preservation calculated off-line is obtained is revised, and obtains the 3rd user interest matrix;

Step 104: according to default business information matrix and the 3rd user interest matrix, calculate online business recommended matrix.

Preferably, obtain historical record corresponding to each business information that this user operates in very first time section, comprising:

From this user's history file, obtain historical record corresponding to the business information of running time in very first time section;

If the number of the historical record that the business information of obtaining is corresponding is greater than default number, from historical record corresponding to the business information obtained, obtain the running time from a current nearest default historical record corresponding to several business information.

Further, obtain historical record corresponding to each business information that this user operates in very first time section, also comprise:

If the number of the historical record that the business information of obtaining is corresponding is less than or equal to default number, select historical record corresponding to each business information obtaining.

Preferably, the historical record corresponding according to each business information, builds first user interests matrix, comprising:

The historical record corresponding according to each business information builds business conduct matrix, the corresponding business information of every a line of business conduct matrix, each is listed as corresponding a kind of operation behavior, the number of times of this user of the element representation in business conduct matrix to business information generation operation behavior;

From default business information matrix, obtain a line item corresponding to each business information and form the first business information matrix;

According to business conduct matrix and the first business information matrix, obtain first user interests matrix.

Preferably, according to business conduct matrix and the first business information matrix, obtain first user interests matrix, comprising:

The historical record corresponding according to each business information, calculates respectively time correlation coefficient corresponding to each business information;

In the first business information matrix, by a line item corresponding to each business information time correlation multiplication corresponding with each business information respectively, obtain the second business information matrix;

According to the second business information transpose of a matrix matrix and business conduct matrix, calculate first user interests matrix.

Preferably, according to the second business information transpose of a matrix matrix and business conduct matrix, calculate first user interests matrix, comprising:

According to the second business information transpose of a matrix matrix and business conduct matrix, by formula as follows (1), calculate first user interests matrix;

B ¹ _K*N=（R _X*K） ^T*F _X*N……（1）

In formula (1), B ¹ _k*Nfor first user interests matrix, first user interests matrix B ¹ _k*Nline number be that K and columns are N; (R _x*K) ^tbe the second business interests matrix R _x*Ktransposed matrix, the second business interests matrix R _x*Kline number be that X and columns are K; F _x*Nfor business conduct matrix, business conduct matrix F _x*Nline number be that X and columns are N.

Preferably, according to first user interests matrix, the second user interest matrix that the calculated off-line of preserving is obtained is revised, and obtains the 3rd user interest matrix, comprising:

According to first user interests matrix and second user's matrix, by formula as follows (2), calculate the 3rd user interest matrix;

B ² _K*N=B _K*N+B ¹ _K*N……（2）

In formula (2), B ² _k*Nbe the 3rd user interest matrix, the 3rd user interest matrix B ² _k*Nline number be that K and columns are N; B _k*Nbe the second user interest matrix, the second user interest matrix B _k*Nline number be that K and columns are N.

Preferably, according to default business information matrix and the 3rd user interest matrix, calculate online business recommended matrix, comprising:

Determine the columns that default business information matrix comprises, build a diagonal matrix, the line number of diagonal matrix and columns equate with definite columns, and the element on the diagonal line in diagonal matrix is all default value;

According to default business information matrix, diagonal matrix and the 3rd user interest matrix, calculate online business recommended matrix.

Preferably, according to default business information matrix, diagonal matrix and the 3rd user interest matrix, calculate online business recommended matrix, comprising:

According to default business information matrix, diagonal matrix and the 3rd user interest matrix, according to formula as follows (3), calculate online business recommended matrix;

W _M*N=A _M*K*C _K*K*B ² _K*N……（3）

In formula (3), W _m*Nfor online business recommended matrix, online business recommended matrix W _m*Nline number be that M and columns are N; A _m*Kfor business information matrix, business information matrix A _m*Kline number be that M and columns are K; C _k*Kfor diagonal matrix, diagonal matrix C _k*Kline number and columns be all K.

Further, the method also comprises:

A line item corresponding to each business information comprising for online business recommended matrix, there is level of interest and every kind of weight coefficient that operation behavior is corresponding of every kind of operation behavior in this user who comprises according to record, calculates the total level of interest of this user to business information generation operation behavior to business information.

In embodiments of the present invention, obtain historical record corresponding to each business information that user operates in very first time section, very first time section be from current recently and duration be default duration time period; The historical record corresponding according to each business information, builds first user interests matrix; According to first user interests matrix, the second user interest matrix that preservation calculated off-line is obtained is revised, and obtains the 3rd user interest matrix; According to default business information matrix and the 3rd user interest matrix, calculate online business recommended matrix.Due to historical record corresponding to each business information operating in very first time section according to user, built first user interests matrix, and utilize first user interests matrix, the second user interest matrix that preservation calculated off-line is obtained is revised, obtain the 3rd user interest matrix, the larger business information of level of interest that the online business recommended matrix so obtaining according to the 3rd user interest matrix the comprises current interested business information of being more close to the users, thus the accuracy of recommending business information improved.

Embodiment 2

The embodiment of the present invention provides a kind of method that off-line algorithm is calculated online.

When server need to be recommended business information to user, server obtains online business recommended matrix according to the method providing by the embodiment of the present invention, and then recommends business information to user according to online business recommended matrix.

Referring to Fig. 2, the method specifically comprises:

Step 201: obtain historical record corresponding to each business information that user operates in very first time section, very first time section be from current recently and duration be default duration time period;

Wherein, business information can be the information of the different information types such as video, advertisement or merchandise news, at this, just differs one for example.When user operates a business information, for example, when user watches, clicks or collects a business information, server can operate the sign of this business information, this user the operation behavior type of this business information and operate a historical record of running time composition of this business information, and this historical record is kept in this user's history file.Operation behavior type comprises to be watched, clicks or collection etc.Default duration can be half an hour, ten minutes or two minutes etc.

This step is specially, and obtains historical record corresponding to the business information of running time in very first time section from this user's history file; If the number of the historical record that the business information of obtaining is corresponding is greater than default number, from historical record corresponding to the business information obtained, select the running time from a current nearest default historical record corresponding to several business information; For example, default number can be the numerical value such as 15,20 or 30, suppose, the number of the historical record that the business information obtained is corresponding is 30, and default number is 20, from historical record corresponding to 30 business information obtaining, select the running time from current 20 nearest historical records that business information is corresponding.If the number of the historical record that the business information of obtaining is corresponding is less than or equal to default number, select historical record corresponding to each business information obtaining, suppose, the number of the historical record that the business information obtained is corresponding is 15, and default number is 20, select 15 historical records that business information is corresponding that obtain.

Step 202: the historical record corresponding according to each business information, builds business conduct matrix;

Wherein, for each business information, according to the historical record of this business information, calculate user this business information is occurred to for the number of times of every kind of operation behavior, the number of times that user, to this business information, every kind of operation behavior is occurred to forms a line item corresponding to this business information, obtain in a manner described a line item corresponding to each business information, and a line item corresponding to each business information formed to business conduct matrix F _x*N, and the business conduct matrix F forming _x*Nas follows.

Business conduct matrix F _x*Nthe corresponding business information of every a line, each is listed as corresponding a kind of operation behavior, X is business conduct matrix F _x*Nthe line number comprising and equating with the number of the business information of obtaining, N is business conduct matrix F _x*Nthe columns comprising and equate F in business conduct matrix with the number of operation behavior _x*Nelement F _ijrepresent that the number of times of operation behavior j occurs business information i user.For example, business conduct matrix F as follows _x*N, business conduct matrix F _x*Nthe corresponding business information 2 of corresponding business information 1, the second row of the first row ..., the capable corresponding business information X of X, business conduct matrix F _x*Nfirst row respective operations behavior 1, secondary series respective operations behavior 2 ..., N row respective operations behavior N, business conduct matrix F _x*Nelement F ₁₁represent that the number of times of operation behavior 1, element F occur business information 1 user ₁₂represent that the number of times of operation behavior 2 occurs business information 1 user ..., element F _xNrepresent that the number of times of operation behavior N occurs business information X user.

$F_{X*N}=\left[\begin{array}{cccc}{F}_{11}& F_{12}& ...& F_{1N}\\ F_{21}& F_{22}& ...& F_{2N}\\ ...& ...& ...& ...\\ F_{X1}& F_{X2}& ...& F_{\mathrm{XN}}\end{array}\right]$

Step 203: obtain a line item corresponding to each business information and form the first business information matrix from default business information matrix;

In advance default business information matrix A as follows _m*K, business information matrix A _m*Kin every a line corresponding server in the business information of storing, business information matrix A _m*Keach be listed as a kind of attribute factor of corresponding business information, business information matrix A _m*Kin elements A _ijrepresent the weight of attribute factor j in business information i.

For example, in server, store business information 1,2 ..., M, the attribute factor of business information comprise attribute factor 1,2 ..., K.In business information matrix A _m*Kin, the corresponding business information 2 of corresponding business information 1, the second row of the first row ..., the capable corresponding business information M of M, and, the corresponding attribute factor 1 of first row, the corresponding attribute factor 2 of secondary series ..., K is listed as corresponding attribute factor K, business information matrix A _m*Kelements A ₁₁represent the weight of attribute factor 1 in business information 1, elements A ₁₂represent the weight of attribute factor 2 in business information 1 ..., elements A _kNrepresent the weight of attribute factor N in business information K.

$A_{M*K}=\left[\begin{array}{cccc}{A}_{11}& A_{12}& ...& A_{1K}\\ A_{21}& A_{22}& ...& A_{2K}\\ ...& ...& ...& ...\\ A_{X1}& A_{X2}& ...& A_{\mathrm{XK}}\\ ...& ...& ...& ...\\ A_{M1}& A_{M2}& ...& A_{\mathrm{MK}}\end{array}\right]$

Wherein, each business information is corresponding multiple attribute factor all, and different for the attribute factor that the business information of different information types is corresponding, such as attribute factor corresponding to video, can be terror, suspense and describing love affairs etc., attribute factor corresponding to advertisement can be automobile, house property and books etc., attribute factor corresponding to merchandise news can be clothes, food and furniture etc., at this, just differs one for example.

Suppose that step 201 obtained x the business information that this user operates in very first time section, from default business information matrix A _m*Kin obtain the first business information matrix S of a line item corresponding to each business information in this x business information shown in composed as follows _x*K.

$S_{X*K}=\left[\begin{array}{cccc}{A}_{11}& A_{12}& ...& A_{1K}\\ A_{21}& A_{22}& ...& A_{2K}\\ ...& ...& ...& ...\\ A_{X1}& A_{X2}& ...& A_{\mathrm{XK}}\end{array}\right]$

Step 204: according to business conduct matrix and the first business information matrix, obtain first user interests matrix;

Particularly, the historical record corresponding according to each business information, calculates respectively time correlation coefficient corresponding to each business information; In the first business information matrix, by a line item corresponding to each business information time correlation multiplication corresponding with each business information respectively, obtain the second business information matrix; According to the second business information transpose of a matrix matrix and business conduct matrix, calculate first user interests matrix.

For example, the operation behavior type of this user's operation service information 1 comprising according to the historical record of business information 1 correspondence and the running time of operation service information 1, by time decay algorithm, calculate the time correlation factor beta of business information 1 correspondence ₁, in the first business information matrix S _x*Kin, the first row of business information 1 correspondence is recorded to the time correlation factor beta that is multiplied by business information 1 correspondence ₁.For other each business information in this X business information with business information 1 according to aforesaid operations, calculate the time correlation factor beta of business information 2 correspondences ₂, the time correlation factor beta of business information 3 correspondences ₃..., the time correlation factor beta that business information X is corresponding _x, and, in the first business information matrix S _x*Kin, the second line item of business information 2 correspondences is multiplied by the time correlation factor beta of business information 2 correspondences ₂, the third line of business information 3 correspondences is recorded to the time correlation factor beta that is multiplied by business information 3 correspondences ₃..., X line item corresponding to business information X is multiplied by the time correlation factor beta that business information X is corresponding _x, obtain the second business information matrix R as follows _x*K.

$R_{X*K}=\left[\begin{array}{cccc}{\mathrm{\β}}_1{A}_{11}& {\mathrm{\β}}_1{A}_{12}& ...& {\mathrm{\β}}_1{A}_{1K}\\ {\mathrm{\β}}_2{A}_{21}& {\mathrm{\β}}_2{A}_{22}& ...& {\mathrm{\β}}_2{A}_{2K}\\ ...& ...& ...& ...\\ {\mathrm{\β}}_X{A}_{X1}& {\mathrm{\β}}_X{A}_{X2}& ...& {\mathrm{\β}}_X{A}_{\mathrm{XK}}\end{array}\right]$

Obtain the second business information matrix R _x*Ktransposed matrix, as follows:

${\left(R_{X*K}\right)}^T=\left[\begin{array}{cccc}{\mathrm{\β}}_1{A}_{11}& {\mathrm{\β}}_2{A}_{21}& ...& {\mathrm{\β}}_X{A}_{X1}\\ {\mathrm{\β}}_1{A}_{12}& {\mathrm{\β}}_2{A}_{22}& ...& {\mathrm{\β}}_X{A}_{X2}\\ ...& ...& ...& ...\\ {\mathrm{\β}}_1{A}_{1K}& {\mathrm{\β}}_2{A}_{2K}& ...& {\mathrm{\β}}_X{A}_{\mathrm{XK}}\end{array}\right]$

According to the second business information matrix R _x*Ktransposed matrix (R _x*K) ^twith business conduct matrix F _x*N, calculate first user interests matrix B as follows ¹ _k*N, first user interests matrix B ¹ _k*Na kind of attribute factor of the corresponding business information of every a line, each is listed as corresponding a kind of operation behavior, first user interests matrix B ¹ _k*Nin element B ¹ _ijrepresent that the interest value of operation behavior j occurs attribute factor i this user.For example,, at first user interests matrix B ¹ _k*Nin, first user interests matrix B ¹ _k*Nthe corresponding attribute factor 2 of corresponding attribute factor 1, the second row of the first row ..., the capable corresponding attribute factor K of K, first user interests matrix B ¹ _k*Nfirst row respective operations behavior 1, secondary series respective operations behavior 2 ..., N row respective operations behavior N, and, first user interests matrix B ¹ _k*Nelement B ¹ ₁₁represent that the interest value of operation behavior 1, element B occur attribute factor 1 this user ¹ ₁₂represent that the interest value of operation behavior 2, element B occur attribute factor 1 this user ¹ _1Nrepresent that the interest value of operation behavior N occurs attribute factor 1 this user ..., element B ¹ _kNrepresent that the interest value of operation behavior N occurs attribute factor K this user.

${B^1}_{K*N}={\left(R_{X*K}\right)}^T*F_{X*N}=\left[\begin{array}{cccc}{B}_{11}^1& B_{12}^1& ...& B_{1N}^1\\ B_{21}^1& B_{22}^1& ...& B_{2N}^1\\ ...& ...& ...& ...\\ B_{K1}^1& B_{K2}^1& ...& B_{\mathrm{KN}}^1\end{array}\right]$

Step 204: according to first user interests matrix, the second user interest matrix that the calculated off-line of preserving is obtained is revised, and obtains the 3rd user interest matrix;

Wherein, in server, preserved the second user interest matrix B as follows _k*N, the second user interest matrix B _k*Nhistorical record calculated off-line corresponding to each business information periodically operating in one-period according to user for server obtains, and the initial time in corresponding cycle of the second user interest matrix is early than the initial time of very first time section.The second user interest matrix B _k*Na kind of attribute factor of the corresponding business information of every a line, each is listed as corresponding a kind of operation behavior, the second user interest matrix B _k*Nin element B _ijrepresent that the interest value of operation behavior j occurs attribute factor i this user, for example, in the second user interest matrix B _k*Nin, the second user interest matrix B _k*Nthe corresponding attribute factor 2 of corresponding attribute factor 1, the second row of the first row ..., the capable corresponding attribute factor K of K, the second user interest matrix B _k*Nfirst row respective operations behavior 1, secondary series respective operations behavior 2 ..., N row respective operations behavior N, and, the second user interest matrix B _k*Nelement B ₁₁represent that the interest value of operation behavior 1, element B occur attribute factor 1 this user ₁₂represent that the interest value of operation behavior 2, element B occur attribute factor 1 this user _1Nrepresent that the interest value of operation behavior N occurs attribute factor 1 this user ..., element B _kNrepresent that the interest value of operation behavior N occurs attribute factor K this user.

$B_{K*N}=\left[\begin{array}{cccc}{B}_{11}& B_{12}& ...& B_{1N}\\ B_{21}& B_{22}& ...& B_{2N}\\ ...& ...& ...& ...\\ B_{K1}& B_{K2}& ...& B_{\mathrm{KN}}\end{array}\right]$

According to first user interests matrix B ¹ _k*N, the second user interest matrix B that the calculated off-line of preserving is obtained _k*Nrevise, obtain the 3rd user interest matrix B as follows ² _k*N, the 3rd user interest matrix B ² _k*Na kind of attribute factor of the corresponding business information of every a line, each is listed as corresponding a kind of operation behavior, the 3rd user interest matrix B ² _k*Nin element B ² _ijrepresent that the interest value of operation behavior j occurs attribute factor i this user, for example, in the 3rd user interest matrix B ² _k*Nin, the 3rd user interest matrix B ² _k*Nthe corresponding attribute factor 2 of corresponding attribute factor 1, the second row of the first row ..., the capable corresponding attribute factor K of K, the 3rd user interest matrix B ² _k*Nfirst row respective operations behavior 1, secondary series respective operations behavior 2 ..., N row respective operations behavior N, and, the 3rd user interest matrix B ² _k*Nelement B ² ₁₁represent that the interest value of operation behavior 1, element B occur attribute factor 1 this user ² ₁₂represent that the interest value of operation behavior 2, element B occur attribute factor 1 this user ² _1Nrepresent that the interest value of operation behavior N occurs attribute factor 1 this user ..., element B ² _kNrepresent that the interest value of operation behavior N occurs attribute factor K this user.

${B^2}_{K*N}=B_{K*N}+{B^1}_{K*N}=\left[\begin{array}{cccc}{B}_{11}^2& B_{12}^2& ...& B_{1N}^2\\ B_{21}^2& B_{22}^2& ...& B_{2N}^2\\ ...& ...& ...& ...\\ B_{K1}^2& B_{K2}^2& ...& B_{\mathrm{KN}}^2\end{array}\right]$

Further, calculating the 3rd user interest matrix B ² _k*Nbefore, can also be to a default value on the second user interest Matrix Multiplication, to dwindle the second user interest matrix, this default value can be for being less than 1 positive number, as 0.1,0.5 or 0.7 etc.

Step 205: according to default business information matrix and the 3rd user interest matrix, calculate online business recommended matrix.

Particularly, obtain the default business information matrix A of having stored _m*K, determine default business information matrix A _m*Kthe columns K comprising, builds diagonal matrix C as follows _k*K, this diagonal matrix C _k*Kline number and columns be all K, and diagonal matrix C _k*Kin diagonal line on element be all the default value (Elements C on diagonal line ₁₁, C ₂₂..., C _kKbe all the numerical value pre-setting).

$C_{K*K}=\left[\begin{array}{cccc}{C}_{11}& 0& ...& 0\\ 0& C_{22}& ..& 0\\ ...& ...& ...& ...\\ 0& 0& ...& C_{\mathrm{KK}}\end{array}\right]$

According to this diagonal matrix C _k*K, the 3rd user interest matrix B ² _k*Nwith the business information matrix A of obtaining _m*K, calculate online business recommended matrix W as follows _m*N, online business recommended matrix W _m*Nthe corresponding business information of every a line, each is listed as corresponding a kind of operation behavior, online business recommended matrix W _m*Nin element W _ijrepresent that the level of interest of operation behavior j occurs business information i this user.For example, online business recommended matrix W _m*Nthe corresponding business information 2 of corresponding business information 1, the second row of the first row ..., the capable corresponding business information M of M, online business recommended matrix W _m*Nfirst row respective operations behavior 1, secondary series respective operations behavior 2 ..., N row respective operations behavior N, and, online business recommended matrix W _m*Nelement W ₁₁represent that the level of interest of operation behavior 1, the 5 plain W of unit occur business information 1 this user ₁₂represent that the level of interest of operation behavior 2, element W occur business information 1 this user _1Nrepresent that the level of interest of operation behavior N occurs business information 1 this user ..., element W _mNrepresent that the level of interest of operation behavior N occurs business information M this user.

$W_{M*N}=A_{M*K}*C_{K*K}*{B^2}_{K*N}=\left[\begin{array}{cccc}{W}_{11}& W_{12}& ...& W_{1N}\\ W_{21}& W_{22}& ...& W_{2N}\\ ...& ...& ...& ...\\ W_{M1}& W_{M2}& ...& W_{\mathrm{MN}}\end{array}\right]$

Further, according to online business recommended matrix W _m*N, this user may interested business information be recommended to this user, be specifically as follows:

From user's history file, obtain the business information that user has operated, from online business recommended matrix W _m*Nin filter out a line item corresponding to each business information that user has operated.For the online business recommended matrix W after filtering _m*Neach business information comprising, the user who comprises according to a line item corresponding to this business information, to level of interest corresponding to every kind of operation behavior of this business information generation and the weight coefficient of every kind of operation behavior, calculates the total interest-degree of this user to this business information transmit operation behavior by default algorithm.

For example, suppose that this business information is business information i, and a line corresponding to business information i is recorded as [W _i1, W _i2..., W _iN].The level of interest of every kind of operation behavior is occurred business information i the user who comprises according to a line item corresponding to business information i to and the weight coefficient of every kind of operation behavior can calculate the total interest-degree W of this user to business information i generation operation behavior by formula as follows (1) _i.

W _i=ω ₁*W _i1+ω ₂*W _i2+……+ω _N*W _iN……（1）

Wherein, ω ₁for the weight coefficient of operation behavior 1, ω ₂for the weight coefficient of operation behavior 2 ..., ω _nthe weight coefficient of operation behavior N.

Wherein, for the online business recommended matrix W after filtering _m*Neach business information comprising, calculates the total interest-degree of this user to each business information generation operation behavior with business information i is the same according to aforesaid operations, then each business information is sorted according to total interest-degree order from big to small.Online business recommended matrix W from sequence _m*Nfirst business information comprising starts, and obtains a preset number business information, and the preset number of an obtaining business information is defined as to the interested business information of user's possibility, then these business information is sent to terminal corresponding to this user.

Further, in step 205, can also first obtain the business information that user has operated, from default business information matrix, filter out a line item corresponding to each business information that user has operated, according to business information matrix and the 3rd user interest matrix after filtering, calculate online business recommended matrix again.

In embodiments of the present invention, obtain historical record corresponding to each business information that user operates in very first time section, very first time section be from current recently and duration be default duration time period; The historical record corresponding according to each business information, builds first user interests matrix; According to first user interests matrix, the second user interest matrix that preservation calculated off-line is obtained is revised, and obtains the 3rd user interest matrix; According to default business information matrix and the 3rd user interest matrix, calculate online business recommended matrix.Due to historical record corresponding to each business information operating in very first time section according to user, built first user interests matrix, and utilize first user interests matrix, the second user interest matrix that preservation calculated off-line is obtained is revised, obtain the 3rd user interest matrix, the larger business information of level of interest that the online business recommended matrix so obtaining according to the 3rd user interest matrix the comprises current interested business information of being more close to the users, thus the accuracy of recommending business information improved.

Embodiment 3

Referring to Fig. 3, the embodiment of the present invention provides a kind of device that off-line algorithm is calculated online, comprising:

Acquisition module 301, historical record corresponding to each business information operating in very first time section for obtaining this user, very first time section be from current recently and duration be default duration time period;

Build module 302, for the historical record corresponding according to each business information, build first user interests matrix;

Correcting module 303, for according to first user interests matrix, revises preserving the second user interest matrix that calculated off-line obtains, and obtains the 3rd user interest matrix;

Computing module 304, for according to default business information matrix and the 3rd user interest matrix, calculates online business recommended matrix.

Wherein, acquisition module 301 comprises:

The first acquiring unit, obtains historical record corresponding to the business information of running time in very first time section for the history file from this user;

Second acquisition unit, if the number for historical record corresponding to the business information obtained is greater than default number, from historical record corresponding to the business information obtained, obtain the running time from a current nearest default historical record corresponding to several business information.

Wherein, acquisition module 301, if be also less than or equal to default number for the number of historical record corresponding to the business information obtained, selects historical record corresponding to each business information obtaining.

Wherein, building module 302 comprises:

Construction unit, for building business conduct matrix according to historical record corresponding to each business information, the corresponding business information of every a line of business conduct matrix, each is listed as corresponding a kind of operation behavior, the number of times of this user of the element representation in business conduct matrix to business information generation operation behavior;

Obtain component units, for the business information matrix from default, obtain a line item corresponding to each business information and form the first business information matrix;

The 3rd acquiring unit, for according to business conduct matrix and the first business information matrix, obtains first user interests matrix.

Wherein, the 3rd acquiring unit comprises:

The first computation subunit, for the historical record corresponding according to each business information, calculates respectively time correlation coefficient corresponding to each business information;

The second computation subunit, at the first business information matrix by a line item corresponding to each business information time correlation multiplication corresponding with each business information respectively, obtain the second business information matrix;

The 3rd computation subunit, for according to the second business information transpose of a matrix matrix and business conduct matrix, calculates first user interests matrix.

Wherein, the 3rd computation subunit, specifically for according to the second business information transpose of a matrix matrix and business conduct matrix, by formula as follows (1), is calculated first user interests matrix;

B ¹ _K*N=（R _X*K） ^T*F _X*N……（1）

In formula (1), B ¹ _k*Nfor first user interests matrix, first user interests matrix B ¹ _k*Nline number be that K and columns are N; (R _x*K) ^tbe the second business interests matrix R _x*Ktransposed matrix, described the second business interests matrix R _x*Kline number be that X and columns are K; F _x*Nfor business conduct matrix, business conduct matrix F _x*Nline number be that X and columns are N.

Wherein, correcting module 303, comprising:

The first computing unit, for according to first user interests matrix and second user's matrix, by formula as follows (2), calculates the 3rd user interest matrix;

B ² _K*N=B _K*N+B ¹ _K*N……（2）

In formula (2), B ² _k*Nbe the 3rd user interest matrix, the 3rd user interest matrix B ² _k*Nline number be that K and columns are N; B _k*Nbe the second user interest matrix, the second user interest matrix B _k*Nline number be that K and columns are N.

Wherein, computing module 304, comprising:

The second construction unit, the columns for determining that default business information matrix comprises, builds a diagonal matrix, and the line number of diagonal matrix and columns equate with definite columns, and the element on the diagonal line in diagonal matrix is all default value;

The second computing unit, for according to default business information matrix, diagonal matrix and the 3rd user interest matrix, calculates online business recommended matrix.

Wherein, the second computing unit, comprising:

The 4th computation subunit, for according to according to default business information matrix, diagonal matrix and the 3rd user interest matrix, according to formula as follows (3), calculates online business recommended matrix;

W _M*N=A _M*K*C _K*K*B ² _K*N……（3）

In formula (3), W _m*Nfor online business recommended matrix, online business recommended matrix W _m*Nline number be that M and columns are N; A _m*Kfor business information matrix, business information matrix A _m*Kline number be that M and columns are K; C _k*Kfor diagonal matrix, diagonal matrix C _k*Kline number and columns be all N.

Further, this device, a line item corresponding to each business information for comprising for online business recommended matrix also, there is level of interest and every kind of weight coefficient that operation behavior is corresponding of every kind of operation behavior in this user who comprises according to record, calculates the total level of interest of user to business information generation operation behavior to business information.

In embodiments of the present invention, obtain historical record corresponding to each business information that user operates in very first time section, very first time section be from current recently and duration be default duration time period; The historical record corresponding according to each business information, builds first user interests matrix; According to first user interests matrix, the second user interest matrix that preservation calculated off-line is obtained is revised, and obtains the 3rd user interest matrix; According to default business information matrix and the 3rd user interest matrix, calculate online business recommended matrix.Due to historical record corresponding to each business information operating in very first time section according to user, built first user interests matrix, and utilize first user interests matrix, the second user interest matrix that preservation calculated off-line is obtained is revised, obtain the 3rd user interest matrix, the larger business information of level of interest that the online business recommended matrix so obtaining according to the 3rd user interest matrix the comprises current interested business information of being more close to the users, thus the accuracy of recommending business information improved.

One of ordinary skill in the art will appreciate that all or part of step that realizes above-described embodiment can complete by hardware, also can come the hardware that instruction is relevant to complete by program, described program can be stored in a kind of computer-readable recording medium, the above-mentioned storage medium of mentioning can be ROM (read-only memory), disk or CD etc.

The foregoing is only preferred embodiment of the present invention, in order to limit the present invention, within the spirit and principles in the present invention not all, any modification of doing, be equal to replacement, improvement etc., within all should being included in protection scope of the present invention.

Claims (20)

1. a method of off-line algorithm being calculated online, is characterized in that, described method comprises:

Obtain historical record corresponding to each business information that user operates in very first time section, described very first time section be from current recently and duration be default duration time period;

According to historical record corresponding to described each business information, build first user interests matrix;

According to described first user interests matrix, the second user interest matrix that the calculated off-line of preserving is obtained is revised, and obtains the 3rd user interest matrix;

According to default business information matrix and described the 3rd user interest matrix, calculate online business recommended matrix.

2. the method for claim 1, is characterized in that, described in obtain historical record corresponding to each business information that user operates in very first time section, comprising:

From described user's history file, obtain historical record corresponding to the business information of running time in very first time section;

If described in the number of historical record corresponding to the business information obtained be greater than default number, from historical record corresponding to the described business information of obtaining, obtain the running time from a current nearest default historical record corresponding to several business information.

3. method as claimed in claim 2, is characterized in that, described method also comprises:

If described in the number of historical record corresponding to the business information obtained be less than or equal to default number, select historical record corresponding to each business information obtaining.

4. the method for claim 1, is characterized in that, historical record corresponding to each business information described in described basis builds first user interests matrix, comprising:

The historical record corresponding according to each business information builds business conduct matrix, the corresponding business information of every a line of described business conduct matrix, each is listed as corresponding a kind of operation behavior, the number of times of user to business information generation operation behavior described in the element representation in described business conduct matrix;

From default business information matrix, obtain a line item corresponding to described each business information and form the first business information matrix;

According to described business conduct matrix and described the first business information matrix, obtain first user interests matrix.

5. method as claimed in claim 4, is characterized in that, describedly according to described business conduct matrix and described the first business information matrix, obtains first user interests matrix, comprising:

According to historical record corresponding to described each business information, calculate respectively described time correlation coefficient corresponding to each business information;

In described the first business information matrix, by a line item corresponding to described each business information time correlation multiplication corresponding with described each business information respectively, obtain the second business information matrix;

According to described the second business information transpose of a matrix matrix and described business conduct matrix, calculate first user interests matrix.

6. method as claimed in claim 5, is characterized in that, described according to described the second business information transpose of a matrix matrix and described business conduct matrix, calculates first user interests matrix, comprising:

According to described the second business information transpose of a matrix matrix and described business conduct matrix, by formula as follows (1), calculate described first user interests matrix;

B ¹ _K*N=（R _X*K） ^T*F _X*N……（1）

In formula (1), B ¹ _k*Nfor described first user interests matrix, described first user interests matrix B ¹ _k*Nline number be that K and columns are N; (R _x*K) ^tfor described the second business interests matrix R _x*Ktransposed matrix, described the second business interests matrix R _x*Kline number be that X and columns are K; F _x*Nfor described business conduct matrix, described business conduct matrix F _x*Nline number be that X and columns are N.

7. the method for claim 1, is characterized in that, described according to described first user interests matrix, and the second user interest matrix that the calculated off-line of preserving is obtained is revised, and obtains the 3rd user interest matrix, comprising:

According to described first user interests matrix and described second user's matrix, by formula as follows (2), calculate described the 3rd user interest matrix;

B ² _K*N=B _K*N+B ¹ _K*N……（2）

In formula (2), B ² _k*Nfor described the 3rd user interest matrix, described the 3rd user interest matrix B ² _k*Nline number be that K and columns are N; B _k*Nfor described the second user interest matrix, described the second user interest matrix B _k*Nline number be that K and columns are N.

8. the method for claim 1, is characterized in that, the business information matrix that described basis is default and described the 3rd user interest matrix, calculate online business recommended matrix, comprising:

Determine the columns that described default business information matrix comprises, build a diagonal matrix, the line number of described diagonal matrix and columns equate with described definite columns, and the element on the diagonal line in described diagonal matrix is all default value;

According to described default business information matrix, described diagonal matrix and described the 3rd user interest matrix, calculate described online business recommended matrix.

9. method as claimed in claim 8, is characterized in that, describedly according to described default business information matrix, described diagonal matrix and described the 3rd user interest matrix, calculates described online business recommended matrix, comprising:

According to described default business information matrix, described diagonal matrix and described the 3rd user interest matrix, according to formula as follows (3), calculate described online business recommended matrix;

W _M*N=A _M*K*C _K*K*B ² _K*N……（3）

In formula (3), W _m*Nfor described online business recommended matrix, described online business recommended matrix W _m*Nline number be that M and columns are N; A _m*Kfor described business information matrix, described business information matrix A _m*Kline number be that M and columns are K; C _k*Kfor described diagonal matrix, described diagonal matrix C _k*Kline number and columns be all K.

10. the method as described in claim 1 to 9 any one claim, is characterized in that, the business information matrix that described basis is default and described the 3rd user interest matrix, after calculating online business recommended matrix, also comprise:

A line item corresponding to each business information comprising for described online business recommended matrix, there is level of interest and the weight coefficient corresponding to described every kind of operation behavior of every kind of operation behavior in the described user who comprises according to described record, calculates the total level of interest of described user to described business information generation operation behavior to described business information.

11. 1 kinds of devices that off-line algorithm is calculated online, is characterized in that, described device comprises:

Acquisition module, historical record corresponding to each business information operating in very first time section for obtaining user, described very first time section be from current recently and duration be default duration time period;

Build module, for according to historical record corresponding to described each business information, build first user interests matrix;

Correcting module, for according to described first user interests matrix, the second user interest matrix that the calculated off-line of preserving is obtained is revised, and obtains the 3rd user interest matrix;

Computing module, for according to default business information matrix and described the 3rd user interest matrix, calculates online business recommended matrix.

12. devices as claimed in claim 11, is characterized in that, described acquisition module comprises:

The first acquiring unit, obtains historical record corresponding to the business information of running time in very first time section for the history file from described user;

Second acquisition unit, if the number of the historical record that the business information of obtaining described in being used for is corresponding is greater than default number, from historical record corresponding to the described business information of obtaining, obtain the running time from a current nearest default historical record corresponding to several business information.

13. devices as claimed in claim 12, is characterized in that, described acquisition module, if also for described in the number of historical record corresponding to the business information obtained be less than or equal to default number, select historical record corresponding to each business information obtaining.

14. devices as claimed in claim 11, is characterized in that, described structure module comprises:

The first construction unit, for building business conduct matrix according to historical record corresponding to each business information, the corresponding business information of every a line of described business conduct matrix, each is listed as corresponding a kind of operation behavior, the number of times of user to business information generation operation behavior described in the element representation in described business conduct matrix;

Obtain component units, for the business information matrix from default, obtain a line item corresponding to described each business information and form the first business information matrix;

The 3rd acquiring unit, for according to described business conduct matrix and described the first business information matrix, obtains first user interests matrix.

15. devices as claimed in claim 14, is characterized in that, described the 3rd acquiring unit comprises:

The first computation subunit, for according to historical record corresponding to described each business information, calculates respectively described time correlation coefficient corresponding to each business information;

The second computation subunit, at described the first business information matrix by a line item corresponding to described each business information time correlation multiplication corresponding with described each business information respectively, obtain the second business information matrix;

The 3rd computation subunit, for according to described the second business information transpose of a matrix matrix and described business conduct matrix, calculates first user interests matrix.

16. devices as claimed in claim 15, is characterized in that,

Described the 3rd computation subunit, specifically for according to described the second business information transpose of a matrix matrix and described business conduct matrix, by formula as follows (1), is calculated described first user interests matrix;

B ¹ _K*N=（R _X*K） ^T*F _X*N……（1）

In formula (1), B ¹ _k*Nfor described first user interests matrix, described first user interests matrix B ¹ _k*Nline number be that K and columns are N; (R _x*K) ^tfor described the second business interests matrix R _x*Ktransposed matrix, described the second business interests matrix R _x*Kline number be that X and columns are K; F _x*Nfor described business conduct matrix, described business conduct matrix F _x*Nline number be that X and columns are N.

17. devices as claimed in claim 11, is characterized in that, described correcting module, comprising:

The first computing unit, for according to described first user interests matrix and described second user's matrix, by formula as follows (2), calculates described the 3rd user interest matrix;

B ² _K*N=B _K*N+B ¹ _K*N……（2）

In formula (2), B ² _k*Nfor described the 3rd user interest matrix, described the 3rd user interest matrix B ² _k*Nline number be that K and columns are N; B _k*Nfor described the second user interest matrix, described the second user interest matrix B _k*Nline number be that K and columns are N.

18. devices as claimed in claim 11, is characterized in that, described computing module, comprising:

The second construction unit, for the columns of determining that described default business information matrix comprises, build a diagonal matrix, the line number of described diagonal matrix and columns equate with described definite columns, and the element on the diagonal line in described diagonal matrix is all default value;

The second computing unit, for according to described default business information matrix, described diagonal matrix and described the 3rd user interest matrix, calculates described online business recommended matrix.

19. devices as claimed in claim 18, is characterized in that, described the second computing unit, comprising:

The 4th computation subunit, for according to according to described default business information matrix, described diagonal matrix and described the 3rd user interest matrix, according to formula as follows (3), calculates described online business recommended matrix;

W _M*N=A _M*K*C _K*K*B ² _K*N……（3）

In formula (3), W _m*Nfor described online business recommended matrix, described online business recommended matrix W _m*Nline number be that M and columns are N; A _m*Kfor described business information matrix, described business information matrix A _m*Kline number be that M and columns are K; C _k*Kfor described diagonal matrix, described diagonal matrix C _k*Kline number and columns be all K.

20. devices as described in claim 11 to 19 any one claim, it is characterized in that, described device, a line item corresponding to each business information for comprising for described online business recommended matrix also, there is level of interest and the weight coefficient corresponding to described every kind of operation behavior of every kind of operation behavior in the described user who comprises according to described record, calculates the total level of interest of described user to described business information generation operation behavior to described business information.