CN105045827A - Familiarity based information recommendation method and apparatus - Google Patents

Abstract

The invention provides a familiarity based information recommendation method and apparatus. The familiarity based information recommendation method provided by the invention comprises: according to operation behaviors of M users for N target objects, establishing a first interaction matrix, wherein elements r,i and j in the first interaction matrix represent the operation behavior of an ith user for a jth target object, i is greater than or equal to 1, is less than or equal to M i and is an integer, j is greater than or equal to 1, is less than or equal to N and is an integer, M is an integer greater than or equal to 2, and N is an integer greater than or equal to 2; determining familiarity of each user with different target objects according to feature model vectors of each user and feature model vectors of each target object, wherein the feature model vectors of each user represent relationships between each user and different labels and the feature model vectors of each target object represent relationships between each target object and the different labels; and determining the target object recommended to each user according to the first interaction matrix and the familiarity of each user with different target objects. According to the method provides by the invention, accuracy of recommending the target object to the user is improved.

Description

Based on information recommendation method and the device of familiarity

Technical field

The present invention relates to computer technology, particularly relate to a kind of information recommendation method based on familiarity and device.

Background technology

Along with the development of infotech, the establishment of information becomes more and more easier with sharing, and cause various information to be explosive growth, user is faced with heavy screening burden.How to help user accurately to obtain within the shortest time and a major issue has been become to oneself valuable information.Such as, in job hunting process, increasing people is by network job search, and the functions such as job hunting website has that recruitment information is issued, resume is downloaded, customization recruitment prefecture, job seeker resume generate, position search and salary query, meet the demand of user.In order to help user to get at short notice oneself valuable job hunting information, job hunting website can recommend talent market to user.

Existing talent market recommends to carry out by the following method: the job information during the resume of user and enterprises recruitment require by job hunting website mates, such as, mate by extracting key word respectively from the resume of user and the job information of enterprise, if matching degree is higher, then this position is recommended user.

But the Limited information that the resume due to user comprises, the key word extracted from resume can not react the preference of user to position exactly, and therefore, the recommendation precision of above-mentioned recommend method is not high.

Summary of the invention

The invention provides a kind of information recommendation method based on familiarity and device, to solve the problem of recommending precision not high.

The invention provides a kind of information recommendation method based on familiarity, comprising:

According to M user to the operation behavior of N number of destination object, set up the first Interactive matrix, the element r in described first Interactive matrix _ijrepresent that i-th user is to the operation behavior of a jth destination object, wherein, 1≤i≤M and be integer, 1≤j≤N and be integer, described M be more than or equal to 2 integer, described N be more than or equal to 2 integer; According to the characteristic model vector of each destination object of characteristic model vector sum of each user, determine the familiarity of each user to different destination objects, wherein, the each user of characteristic model vector representation of each user and the relation of different label, each destination object of characteristic model vector representation of each destination object and the relation of different label; According to described first Interactive matrix and each user, the familiarity to different destination objects determines the destination object recommended to each user.

The present invention also provides a kind of information recommending apparatus based on familiarity, comprising:

Set up module, for according to M user to the operation behavior of N number of destination object, set up the first Interactive matrix, the element r in described first Interactive matrix _ijrepresent that i-th user is to the operation behavior of a jth destination object, wherein, 1≤i≤M and be integer, 1≤j≤N and be integer, described M be more than or equal to 2 integer, described N be more than or equal to 2 integer; First determination module, for the characteristic model vector of each destination object of characteristic model vector sum according to each user, determine the familiarity of each user to different destination objects, wherein, the each user of characteristic model vector representation of each user and the relation of different label, each destination object of characteristic model vector representation of each destination object and the relation of different label; Second determination module, for according to described first Interactive matrix and each user the familiarity to different destination objects determine the destination object recommended to each user.

Information recommendation method based on familiarity provided by the invention and device, by according to M user to the operation behavior of N number of destination object, set up the first Interactive matrix, according to the characteristic model vector of each destination object of characteristic model vector sum of each user, determine the familiarity of each user to different destination objects, wherein, the each user of characteristic model vector representation of each user and the relation of different label, the each destination object of characteristic model vector representation of each destination object and the relation of different label, according to the first Interactive matrix and each user, the familiarity to different destination objects determines the destination object recommended to each user, when recommending destination object to user, take into account the operation behavior of user to destination object, and, based on user, tagged behavior is added to destination object, take into account by representing the user characteristics model vector of user and the relation of different label and representing that the vectorial user determined of the destination object characteristic model of destination object and different label relation is to the familiarity of different target object, thus, improve the precision of recommending destination object to user.

Accompanying drawing explanation

In order to be illustrated more clearly in the present invention or technical scheme of the prior art, be briefly described to the accompanying drawing used required in embodiment or description of the prior art below, apparently, accompanying drawing in the following describes is 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 the process flow diagram of the information recommendation method embodiment one that the present invention is based on familiarity;

Fig. 2 is the process flow diagram of the information recommendation method embodiment two that the present invention is based on familiarity;

Fig. 3 is the schematic flow sheet of a kind of implementation of Fig. 2;

Fig. 4 is the structural representation of the information recommending apparatus embodiment one that the present invention is based on familiarity;

Fig. 5 is the structural representation of the information recommending apparatus embodiment two that the present invention is based on familiarity.

Embodiment

For making the object of the embodiment of the present invention, technical scheme and advantage clearly, below in conjunction with the accompanying drawing in the embodiment of the present invention, technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is 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 paying the every other embodiment obtained under creative work prerequisite, belong to the scope of protection of the invention.

Fig. 1 is the process flow diagram of the information recommendation method embodiment one that the present invention is based on familiarity.As shown in Figure 1, the information recommendation method based on familiarity that the present embodiment provides comprises:

S10: according to M user to the operation behavior of N number of destination object, set up the first Interactive matrix, the element r in the first Interactive matrix _ijrepresent that i-th user is to the operation behavior of a jth destination object, wherein, 1≤i≤M and be integer, 1≤j≤N and be integer, M be more than or equal to 2 integer, N be more than or equal to 2 integer.

Particularly, the information recommendation method based on familiarity that the present embodiment provides can be used for talent market recommendation, shopping information is recommended or watch in the scenes such as video recommendations.In different application scenarioss, destination object represents different things.Such as, recommend in scene in talent market, destination object is exactly job information; Recommend in scene at shopping information, destination object is exactly Item Information.The executive agent of the present embodiment can be provide talent market accordingly, provide shopping information or provide the server of video service website.

In server, the set of user is { u ₁, u ₂..., u _i..., u _m, the set of destination object is { c ₁, c ₂..., c _j..., c _n.M represents the total quantity of all users, M be more than or equal to 2 integer.N represents the total quantity of all destination objects, N be more than or equal to 2 integer.At server run duration, user has operation behavior to destination object, and operation behavior here refers to the interbehavior between user and destination object, and such as, user marks to destination object, or user clicks destination object.

According to the user of M in server to the operation behavior of N number of destination object, set up the first Interactive matrix.In statistical server, user is to the operation behavior of destination object, and operation behavior is expressed as numeral according to certain rules abstraction, represents by the numerical values recited of numeral the operation behavior that user is different to destination object.High to high or this destination object and user the correlativity of this destination object interest level with large numeric representation user in the present embodiment.Owing to having M user and N number of destination object, be the matrix of M × N dimension to the first Interactive matrix that the operation behavior of each destination object is set up according to each user.Element r in first Interactive matrix _ijrepresent that i-th user is to the operation behavior of a jth destination object, 1≤i≤M and be integer, 1≤j≤N and be integer.It should be noted that, because in server, the quantity of user is a lot, the quantity of destination object is also a lot, each user can not have operation behavior to all destination objects, the operation behavior of user to a lot of destination object is unknown, represent the unknown element in the first Interactive matrix in the present embodiment with 0, therefore, the first Interactive matrix is a sparse matrix.

S20: according to the characteristic model vector of each destination object of characteristic model vector sum of each user, determine the familiarity of each user to different destination objects, wherein, the each user of characteristic model vector representation of each user and the relation of different label, each destination object of characteristic model vector representation of each destination object and the relation of different label.

Particularly, user can add label to show some speciality of destination object to destination object.Same user can add identical or different labels to different destination objects; Destination object can be added identical or different label by different users.For example, in the scene that talent market is recommended, some positions can be added to the label of " high salary " and/or " software development " with the technology kind of the wages situation and/or this position that show this position by user; In video recommendations scene, some video councils are added to " comedy " to show the type of this video by user.Each user reflects the feature of user from the relation of different labels: user use certain label to show frequently the content that user represents this label is interested or more familiar; Each destination object and the relation table of different label understand some feature of this destination object: destination object is added certain label frequently and indicates that the content relevance that this destination object and this label represent is higher.Label can be the label that provides of server or user-defined label.

According to the characteristic model vector of each destination object of characteristic model vector sum of each user, determine the familiarity of each user to different destination objects.Add up each user and determine that the characteristic model of each user is vectorial from the relation of different label; Add up each destination object and determine that the characteristic model of each destination object is vectorial from the relation of different label.User can be the number of times that user uses certain label from the relation of different label, and destination object can be the number of times that destination object is added certain label from the relation of different label.According to the characteristic model vector of each destination object of characteristic model vector sum of each user, determine the familiarity of each user to different destination objects, the familiarity of user to different destination objects indicates the interest level of user to different destination objects.

S30: the familiarity to different destination objects determines the destination object recommended to each user according to the first Interactive matrix and each user.

Particularly, according to user existing in the first Interactive matrix to the operation behavior of destination object, and each user is to the familiarity of different destination objects, finally determines the destination object recommended to each user.Can be carry out computing according to some element in the first Interactive matrix and user to the familiarity of destination object to obtain a destination object to the recommendation of user, this recommendation and the threshold value preset are compared, if be greater than predetermined threshold value, then this destination object is recommended user.Certainly, also have other implementation, the present embodiment is not as limit.

It should be noted that, there is not the relation in sequential in S10 and S20.Can first set up the first Interactive matrix, also first can determine the familiarity of each user to different destination objects, or, set up the first Interactive matrix and determine that each user carries out the familiarity of different destination objects simultaneously.

The information recommendation method based on familiarity that the present embodiment provides, by according to M user to the operation behavior of N number of destination object, set up the first Interactive matrix, according to the characteristic model vector of each destination object of characteristic model vector sum of each user, determine the familiarity of each user to different destination objects, wherein, the each user of characteristic model vector representation of each user and the relation of different label, the each destination object of characteristic model vector representation of each destination object and the relation of different label, according to the first Interactive matrix and each user, the familiarity to different destination objects determines the destination object recommended to each user, when recommending destination object to user, take into account the operation behavior of user to destination object, and, based on user, tagged behavior is added to destination object, take into account by representing the user characteristics model vector of user and the relation of different label and representing that the vectorial user determined of the destination object characteristic model of destination object and different label relation is to the familiarity of different target object, thus, improve the precision of recommending destination object to user.

Fig. 2 is the process flow diagram of the information recommendation method embodiment two that the present invention is based on familiarity.As shown in Figure 2, the information recommendation method based on familiarity that the present embodiment provides, on the basis of embodiment one, S20 specifically comprises:

S201: the number of times adding often kind of label according to i-th user, adds the total degree of all labels, the size of M, and M user adds the total degree of often kind of label, determines the characteristic model vector of i-th user.

Particularly, in server, tag set is T={t ₁, t ₂..., t _k..., t _l, L is the total quantity of label, L be more than or equal to 2 integer, t _krepresent kth kind label, 1≤k≤L and be integer.The number of times of often kind of label in tag set is added according to i-th user, i-th user adds the total degree of all labels in tag set, the size of the quantity M of user in server, in server, M user adds the characteristic model vector that the total degree of often kind of label in tag set determines i-th user.

Optionally, according to formula ${\stackrel{\‾}{u}}_i=\left(\frac{s_i\left(t_1\right)}{n_{u_i}}\mathrm{\ω}\right(t_1),\frac{s_i\left(t_2\right)}{n_{u_i}}\mathrm{\ω}(t_2),\mathrm{...},\frac{s_i\left(t_L\right)}{n_{u_i}}\mathrm{\ω}(t_L\left)\right)$ Determine the characteristic model vector of i-th user, wherein, u _irepresent i-th user, S _i(t _k) represent that i-th user adds the number of times of kth kind label, represent that i-th user adds the total degree of all labels, represent that M user adds the total degree of kth kind label.The each user of characteristic model vector representation and the relation between different label of user.

S202: the number of times being added often kind of label according to a jth destination object, is added the total degree of all labels, the size of M, M user adds the total degree of often kind of label, determines the characteristic model vector of a jth destination object.

Particularly, the number of times of often kind of label is added according to a jth destination object, a jth destination object is added the total degree of all labels in tag set, the size of the quantity M of user in server, in server, M user adds the characteristic model vector of the total degree determination destination object of often kind of label in tag set.

Optionally, according to formula ${\stackrel{\‾}{c}}_j=\left(\frac{s_j\left(t_1\right)}{m_{c_j}}\mathrm{\ω}\right(t_1),\frac{s_j\left(t_2\right)}{m_{c_j}}\mathrm{\ω}(t_2),\mathrm{...},\frac{s_j\left(t_L\right)}{m_{c_j}}\mathrm{\ω}(t_L\left)\right)$ Determine the characteristic model vector of a jth destination object, wherein, s _j(t _k) represent that a jth destination object is added the number of times of kth kind label, represent that a jth destination object is added the total degree of all labels.The each destination object of characteristic model vector representation and the relation between different label of destination object.

S203: according to the characteristic model vector of a characteristic model vector sum jth destination object of i-th user, determine that i-th user is to the familiarity of a jth destination object.

Particularly, according to the relation of user from different label, with the relation of destination object from different label, the relation between user and destination object can be determined, namely can determine that i-th user is to the familiarity of a jth destination object according to the characteristic model vector of a characteristic model vector sum jth destination object of i-th user.Familiarity in the present embodiment can represent that i-th user is to the interest level of a jth destination object.

Optionally, according to formula determine that i-th user is to the familiarity of a jth destination object, wherein, X represents the characteristic model vector of M user and the maximal value of the characteristic model dot product of N number of destination object, and namely X is M × N number of in maximal value, α is default constant.

The information recommendation method based on familiarity that the present embodiment provides, on the basis of embodiment one, S30 specifically comprises:

S301: the hidden eigenvectors matrix U of initialising subscriber and the hidden eigenvectors matrix V of destination object.

Particularly, the hidden eigenvectors matrix U of user represents that the hidden feature of user, the hidden eigenvectors matrix V of destination object represent the hidden feature of destination object, the feature about user and destination object namely obtained based on the operation behavior between user and destination object.U _ithe hidden proper vector of i-th user, V _jrepresent the hidden proper vector of a jth destination object, U _iand V _jdimension can determine according to the actual requirements, dimension is higher, and recommend precision higher, but correspondingly, computational complexity also can be higher.The hidden eigenvectors matrix of the hidden proper vector composition user of M user, the hidden eigenvectors matrix of the hidden proper vector composition destination object of N number of destination object.During the hidden eigenvectors matrix V of the hidden eigenvectors matrix U of initialising subscriber and destination object, can be for the element in matrix U and matrix V gets initial value.

S302: according to each user to the familiarity of different destination objects and the first Interactive matrix, upgrade the hidden eigenvectors matrix U of user and the hidden eigenvectors matrix V of destination object, until the hidden eigenvectors matrix U of user and the hidden eigenvectors matrix V of destination object restrains.

Particularly, after the hidden eigenvectors matrix U of user and the hidden eigenvectors matrix V of destination object gets initial value, upgrade the hidden eigenvectors matrix U of user and the hidden eigenvectors matrix V of destination object.According to each user to the familiarity of different destination objects and the first Interactive matrix, upgrade the hidden eigenvectors matrix U of user and the hidden eigenvectors matrix V of destination object.Optionally, according to formula: upgrade the hidden eigenvectors matrix U of user, wherein, $\frac{\∂E}{\∂U_i}=\underset{j=1}{\overset{N}{\Σ}}{I}_{ij}\left(S\right({\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j)U_i^T{V}_j-r_{ij})\·S({\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j)V_j+{\mathrm{\λ}}_U{U}_i,$ I _ijrepresentation unit matrix, ${\mathrm{\λ}}_U=\frac{{{\mathrm{\σ}}_R}^2}{{\mathrm{\σ}}_U^2},$ R represents the first Interactive matrix, represent the variance of the first Interactive matrix, represent the variance of the hidden eigenvectors matrix of user, U _irepresent the hidden proper vector of i-th user, η represents learning efficiency.According to formula upgrade the hidden eigenvectors matrix V of described destination object, wherein, $\frac{\∂E}{\∂V_j}=\underset{i=1}{\overset{M}{\Σ}}{I}_{ij}\left(S\right({\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j)U_i^T{V}_j-r_{ij})\·S({\stackrel{\‾}{u}}_j,{\stackrel{\‾}{c}}_j)U_i+{\mathrm{\λ}}_V{V}_j,{\mathrm{\λ}}_V=\frac{{{\mathrm{\σ}}_R}^2}{{\mathrm{\σ}}_V^2},$ represent the variance of the hidden eigenvectors matrix of destination object, V _jrepresent the hidden proper vector of a jth destination object.

The formulation process of the hidden proper vector of the hidden proper vector of i-th user and a jth target is as follows: suppose that the condition of the first Interactive matrix is distributed as $P\left(R\right|U,V,{\mathrm{\σ}}_R^2)=\underset{i=1}{\overset{M}{\Π}}\underset{j=1}{\overset{N}{\Π}}{\[N\left(r_{ij}\right|g\left(U_i^T{V}_j\right),{\mathrm{\σ}}_R^2)\]}^{I_{ij}^R},$ Wherein, represent the variance of the first Interactive matrix R, for indicator function: represent the element r in the first Interactive matrix R _ijtime non-vanishing, then if the element r in the first Interactive matrix R _ijwhen being zero, then represent that average is variance is gaussian probability-density function.

Suppose that the condition of user characteristics vector matrix U and position eigenvectors matrix V is respectively $P\left(U\right|{\mathrm{\σ}}_U^2)=\underset{i=1}{\overset{M}{\Π}}N\left(U_i\right|0,{\mathrm{\σ}}_U^{2}I),P\left(V\right|{\mathrm{\σ}}_V^2)=\underset{j=1}{\overset{N}{\Π}}N\left(V_j\right|0,{\mathrm{\σ}}_V^{2}I),$ Wherein, I is unit matrix, represent the variance of user characteristics vector matrix U, represent the variance of position eigenvectors matrix V.

The familiarity of i-th user to a jth destination object is fused in recommended models, obtains:

$p\left(R\right|U,V,{\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j,{{\mathrm{\σ}}_R}^2)=\underset{i=1}{\overset{M}{\Π}}\underset{j=1}{\overset{N}{\Π}}{\[N\left(r_{ij}\right|f\left(U_i,V_j,{\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j\right),{{\mathrm{\σ}}_R}^2)\]}^{I_{ij}^R}$ Obtain the recommended models based on familiarity, wherein, $f(U_i,V_j,{\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j)=S({\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j)\·U_i^T{V}_j.$

Obtaining probability matrix decomposition model by Bayes' theorem is:

$\begin{array}{c}p(U,V|R,{\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j,{{\mathrm{\σ}}_R}^2,{\mathrm{\σ}}_U^2,{\mathrm{\σ}}_V^2)\∝p\left(R\right|U,V,{\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j,{{\mathrm{\σ}}_R}^2)p\left(U\right|{\mathrm{\σ}}_U^2)p\left(U\right|{\mathrm{\σ}}_V^2)\\ =\underset{i=1}{\overset{M}{\Π}}\underset{j=1}{\overset{N}{\Π}}{\[N\left(r_{ij}\right|f\left(U_i,V_j,{\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j\right),{{\mathrm{\σ}}_R}^2)\]}^{I_{ij}^R}\×\underset{i=1}{\overset{M}{\Σ}}\[N\left(U_i\right|0,{\mathrm{\σ}}_U^{2}I)\]\×\underset{j=1}{\overset{N}{\Π}}\[N\left(V_j\right|0,{\mathrm{\σ}}_V^{2}I)\]\end{array}$

Above formula both sides function of simultaneously taking the logarithm is obtained:

$\begin{array}{c}\ln p(U,V|R,{\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j,{{\mathrm{\σ}}_R}^2,{\mathrm{\σ}}_U^2,{\mathrm{\σ}}_V^2)=-\frac{1}{2{{\mathrm{\σ}}_R}^2}\underset{i=1}{\overset{M}{\Σ}}\underset{j=1}{\overset{M}{\Σ}}{I}_{ij}{(r_{ij}-f(U_i,V_j,{\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j\left)\right)}^2\\ -\frac{1}{2{\mathrm{\σ}}_U^2}\underset{i=1}{\overset{M}{\Σ}}{U}_i^T{U}_i-\frac{1}{2{\mathrm{\σ}}_V^2}\underset{j=1}{\overset{N}{\Σ}}{V}_j^T{V}_j-\frac{1}{2}\[\left(\underset{i=1}{\overset{M}{\Σ}}\underset{j=1}{\overset{N}{\Σ}}{I}_{ij}\right){{\mathrm{ln\σ}}_R}^2+{\mathrm{MDln\σ}}_U^2+{\mathrm{NDln\σ}}_V^2\]+C\end{array}$

Wherein, C is default constant, and D is the dimension of the hidden proper vector of user and the hidden proper vector of destination object.

Maximize above formula and be equivalent to the objective function minimized below:

$E=\frac{1}{2}\underset{i=1}{\overset{M}{\Σ}}\underset{j=1}{\overset{N}{\Σ}}{I}_{ij}{(r_{ij}-f(U_i,V_j,{\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j\left)\right)}^2+\frac{{\mathrm{\λ}}_U}{2}\underset{i=1}{\overset{M}{\Σ}}{||U_i||}_F^2+\frac{{\mathrm{\λ}}_V}{2}\underset{j=1}{\overset{N}{\Σ}}{||V_j||}_F^2,$ || || this Frobenius norm of not Luo Beini crow of representing matrix.Stochastic gradient descent algorithm is used to obtain to above formula:

$\frac{\∂E}{\∂U_i}=\underset{j=1}{\overset{N}{\Σ}}{I}_{ij}\left(S\right(\stackrel{\‾}{u_i},{\stackrel{\‾}{c}}_j)U_i^T{V}_j-r_{ij})\·S(\stackrel{\‾}{u_i},{\stackrel{\‾}{c}}_j)V_j+{\mathrm{\λ}}_U{U}_i,$

$\frac{\∂E}{\∂V_j}=\underset{i=1}{\overset{M}{\Σ}}{I}_{ij}\left(S\right({\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j)U_i^T{V}_j-r_{ij})\·S({\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j)U_i+{\mathrm{\λ}}_V{V}_j$

The iterative formula finally obtaining the hidden eigenvectors matrix of i-th user is: the iterative formula of the hidden eigenvectors matrix of a jth destination object is:

According to iterative formula, iteration is carried out to the hidden eigenvectors matrix of user and the hidden eigenvectors matrix of destination object.Until the result of current iteration is compared with the result of last iteration, element in matrix U and matrix V is tending towards constant, now, think that the hidden eigenvectors matrix U of user and the hidden eigenvectors matrix V of destination object restrains, then finishing iteration, gets the hidden eigenvectors matrix V of the hidden eigenvectors matrix U of the user after renewal and the destination object after upgrading.

S303: the second Interactive matrix obtained according to the hidden eigenvectors matrix U of user after renewal and the hidden eigenvectors matrix V dot product of destination object, determines the destination object recommended to each user.

Particularly, obtain the second Interactive matrix according to the hidden proper vector U of user after renewal and the hidden eigenvectors matrix V dot product of destination object, then the second Interactive matrix is the matrix of M × N dimension, the element r in the second square _ij' represent that i-th user of prediction is to the operation behavior of a jth destination object.According in the second Interactive matrix, each user, to the size order of the numerical value of the operation behavior of different target object, recommends at least one destination object to each user.Such as, can according in the second Interactive matrix, the numerical value of each user to the operation behavior of different target object sorts, and recommends the destination object coming first 5.

The information recommendation method based on familiarity that the present embodiment provides, by the second Interactive matrix obtained according to the hidden eigenvectors matrix dot product of the hidden eigenvectors matrix of the user after renewal and the destination object after upgrading, the destination object that true directional user recommends, make when recommending destination object to user, take into account the operation behavior of user to destination object, and, based on user, tagged behavior is added to destination object, take into account by representing the user characteristics model vector of user and the relation of different label and representing that the vectorial user determined of the destination object characteristic model of destination object and different label relation is to the familiarity of different target object, thus, improve the precision of recommending destination object to user.

Fig. 3 is the schematic flow sheet of a kind of implementation of Fig. 2.The schematic flow sheet that Fig. 3 provides is applicable in the scene that talent market recommends, and as shown in Figure 3, obtains the label information that is added in label information that user adds and talent market, then obtain user characteristics and talent market feature from talent market data; User history information data is obtained from talent market data.Obtain user's familiarity sensor model according to user characteristics, talent market feature and user history information data, pass through the test to model and training, finally produce recommended structure, recommend user.In the process of recommending, consider relation and user's historical data of user characteristics and talent market feature simultaneously, make recommendation results more accurate.

Fig. 4 is the structural representation of the information recommending apparatus embodiment one that the present invention is based on familiarity.As shown in Figure 4, the information recommending apparatus based on familiarity that the present embodiment provides comprises:

Set up module 41, for according to M user to the operation behavior of N number of destination object, set up the first Interactive matrix, the element r in the first Interactive matrix _ijrepresent that i-th user is to the operation behavior of a jth destination object, wherein, 1≤i≤M and be integer, 1≤j≤N and be integer, M be more than or equal to 2 integer, N be more than or equal to 2 integer.

First determination module 42, for the characteristic model vector of each destination object of characteristic model vector sum according to each user, determine the familiarity of each user to different destination objects, wherein, the each user of characteristic model vector representation of each user and the relation of different label, each destination object of characteristic model vector representation of each destination object and the relation of different label.

Second determination module 43, for according to the first Interactive matrix and each user the familiarity to different destination objects determine the destination object recommended to each user.

The device of the present embodiment, may be used for the technical scheme performing embodiment of the method shown in Fig. 1, it is similar that it realizes principle, repeats no more herein.

The information recommending apparatus based on familiarity that this enforcement provides, by setting up module, for according to M user to the operation behavior of N number of destination object, set up the first Interactive matrix, first determination module, for the characteristic model vector of each destination object of characteristic model vector sum according to each user, determine the familiarity of each user to different destination objects, wherein, the each user of characteristic model vector representation of each user and the relation of different label, the each destination object of characteristic model vector representation of each destination object and the relation of different label, second determination module, for according to the first Interactive matrix and each user the familiarity to different destination objects determine the destination object recommended to each user, when recommending destination object to user, take into account the operation behavior of user to destination object, and, based on user, tagged behavior is added to destination object, take into account by representing the user characteristics model vector of user and the relation of different label and representing that the vectorial user determined of the destination object characteristic model of destination object and different label relation is to the familiarity of different target object, thus, improve the precision of recommending destination object to user.

Fig. 5 is the structural representation of the information recommending apparatus embodiment two that the present invention is based on familiarity.As shown in Figure 5, the information recommending apparatus based on familiarity that the present embodiment provides, on the basis of above-described embodiment one, the first determination module 42 specifically comprises:

First determining unit 421, for the number of times adding often kind of label according to i-th user, adds the total degree of all labels, the size of M, and M user adds the total degree of often kind of label, determines the characteristic model vector of i-th user.

Second determining unit 422, for being added the number of times of often kind of label according to a jth destination object, is added the total degree of all labels, the size of M, and M user adds the total degree of often kind of label, determines the characteristic model vector of a jth destination object.

3rd determining unit 423, for the characteristic model vector of the characteristic model vector sum jth destination object according to i-th user, determines that i-th user is to the familiarity of a jth destination object.

Second determination module 43 specifically comprises:

Initialization unit 431, for the hidden eigenvectors matrix U of initialising subscriber and the hidden eigenvectors matrix V of destination object.

Updating block 432, for according to each user to the familiarity of different destination objects and the first Interactive matrix, upgrade the hidden eigenvectors matrix U of user and the hidden eigenvectors matrix V of destination object, until the hidden eigenvectors matrix U of user and the hidden eigenvectors matrix V of destination object restrains.

4th determining unit 433, for the second Interactive matrix obtained according to the hidden eigenvectors matrix U of user after renewal and the hidden eigenvectors matrix V dot product of destination object, determines the destination object recommended to each user.

The device of the present embodiment, may be used for the technical scheme performing embodiment of the method shown in Fig. 2, it is similar that it realizes principle, repeats no more herein.

The information recommending apparatus based on familiarity that the present embodiment provides, by arranging the first determining unit, second determining unit and the 3rd determining unit, by arranging initialization unit, updating block and the 4th determining unit, according to the second Interactive matrix that the hidden eigenvectors matrix of user after upgrading and the hidden eigenvectors matrix dot product of destination object after upgrading obtain, the destination object that true directional user recommends, make when recommending destination object to user, take into account the operation behavior of user to destination object, and, based on user, tagged behavior is added to destination object, take into account by representing the user characteristics model vector of user and the relation of different label and representing that the vectorial user determined of the destination object characteristic model of destination object and different label relation is to the familiarity of different target object, thus, improve the precision of recommending destination object to user.

One of ordinary skill in the art will appreciate that: all or part of step realizing above-mentioned each embodiment of the method can have been come by the hardware that programmed instruction is relevant.Aforesaid program can be stored in a computer read/write memory medium.This program, when performing, performs the step comprising above-mentioned each embodiment of the method; And aforesaid storage medium comprises: ROM, RAM, magnetic disc or CD etc. various can be program code stored medium.

Last it is noted that above each embodiment is only in order to illustrate technical scheme of the present invention, be not intended to limit; Although with reference to foregoing embodiments to invention has been detailed description, those of ordinary skill in the art is to be understood that: it still can be modified to the technical scheme described in foregoing embodiments, or carries out equivalent replacement to wherein some or all of technical characteristic; And these amendments or replacement, do not make the essence of appropriate technical solution depart from the scope of various embodiments of the present invention technical scheme.

Claims (10)

1. based on an information recommendation method for familiarity, it is characterized in that, comprising:

According to M user to the operation behavior of N number of destination object, set up the first Interactive matrix, the element r in described first Interactive matrix _ijrepresent that i-th user is to the operation behavior of a jth destination object, wherein, 1≤i≤M and be integer, 1≤j≤N and be integer, described M be more than or equal to 2 integer, described N be more than or equal to 2 integer;

According to the characteristic model vector of each destination object of characteristic model vector sum of each user, determine the familiarity of each user to different destination objects, wherein, the each user of characteristic model vector representation of each user and the relation of different label, each destination object of characteristic model vector representation of each destination object and the relation of different label;

According to described first Interactive matrix and each user, the familiarity to different destination objects determines the destination object recommended to each user.

2. method according to claim 1, the characteristic model vector of the described each destination object of characteristic model vector sum according to each user, determine the familiarity of each user to different destination objects, comprising:

Add the number of times of often kind of label according to i-th user, add the total degree of all labels, the size of described M, described M user adds the total degree of often kind of label, determines the characteristic model vector of i-th user;

Be added the number of times of often kind of label according to a jth destination object, be added the total degree of all labels, the size of described M, described M user adds the total degree of often kind of label, determines the characteristic model vector of a jth destination object;

According to the characteristic model vector sum of described i-th user, the characteristic model vector of a jth destination object, determines that described i-th user is to the familiarity of a described jth destination object.

3. method according to claim 2, is characterized in that, the described number of times adding often kind of label according to i-th user, add the total degree of all labels, the size of described M, described M user adds the total degree of often kind of label, determine the characteristic model vector of i-th user, comprising:

According to formula $\stackrel{\‾}{u}i=(\frac{s_i\left(t_1\right)}{n_{u_i}}\mathrm{\ω}\left(t_1\right),\frac{s_i\left(t_2\right)}{n_{u_i}}\mathrm{\ω}\left(t_2\right),...,\frac{s_i\left(t_L\right)}{n_{u_i}}\mathrm{\ω}\left(t_L\right)),$ Determine i-th user characteristic model vector, wherein, L is the total quantity of label, described L be more than or equal to 2 integer, t _krepresent kth kind label, 1≤k≤L and be integer, u _irepresent i-th user, S _i(t _k) represent that i-th user adds the number of times of kth kind label, represent that i-th user adds the total degree of all labels, represent that a described M user adds the total degree of kth kind label.

4. method according to claim 3, it is characterized in that, the described number of times being added often kind of label according to a jth destination object, be added the total degree of all labels, the size of described M, described M user adds the total degree of often kind of label, determines the characteristic model vector of a jth destination object, comprising:

According to formula $\stackrel{\‾}{c}j=(\frac{s_j\left(t_1\right)}{m_{c_j}}\mathrm{\ω}\left(t_1\right),\frac{s_j\left(t_2\right)}{m_{c_j}}\mathrm{\ω}\left(t_2\right),...,\frac{s_j\left(t_L\right)}{m_{c_j}}\mathrm{\ω}\left(t_L\right)),$ Determine the characteristic model vector of a jth destination object, wherein, s _j(t _k) represent that a jth destination object is added the number of times of kth kind label, c _jrepresent a jth destination object, represent that a jth destination object is added the total degree of all labels.

5. method according to claim 4, is characterized in that, the characteristic model vector of a described jth destination object according to the characteristic model vector sum of described i-th user, determines that described i-th user is to the familiarity of a described jth destination object, comprising:

According to formula determine that described i-th user is to the familiarity of a described jth destination object, wherein, X is M × N number of in maximal value, α is default constant.

6. method according to claim 5, is characterized in that, described according to described first Interactive matrix and each user the familiarity to different destination objects determine the destination object recommended to each user, comprising:

The hidden eigenvectors matrix U of initialising subscriber and the hidden eigenvectors matrix V of destination object;

According to described each user to the familiarity of different destination objects and described first Interactive matrix, upgrade the hidden eigenvectors matrix U of described user and the hidden eigenvectors matrix V of described destination object, until the hidden eigenvectors matrix U of described user and the hidden eigenvectors matrix V of described destination object restrains;

According to the second Interactive matrix that the hidden eigenvectors matrix U of described user after renewal and the hidden eigenvectors matrix V dot product of described destination object obtain, determine the destination object recommended to each user.

7. method according to claim 6, it is characterized in that, described the second Interactive matrix obtained according to the hidden eigenvectors matrix U of described user after renewal and the hidden eigenvectors matrix V dot product of described destination object, determine the destination object recommended to each user, comprising:

According in described second Interactive matrix, each user, to the size order of the numerical value of the operation behavior of different target object, recommends at least one destination object to each user.

8. method according to claim 6, it is characterized in that, described according to described each user to the familiarity of different destination objects and described first Interactive matrix, upgrade the hidden eigenvectors matrix U of described user and the hidden eigenvectors matrix V of described destination object, comprising:

According to formula upgrade the hidden eigenvectors matrix U of described user, wherein, $\frac{\∂E}{\∂U_i}=\underset{j=1}{\overset{N}{\Σ}}{I}_{ij}(S({\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j)U_i^T{V}_j-r_{ij})\·S({\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j)V_j+{\mathrm{\λ}}_U{U}_i,$ I _ijrepresentation unit matrix, ${\mathrm{\λ}}_U=\frac{{{\mathrm{\σ}}_R}^2}{{\mathrm{\σ}}_U^2},$ R represents described first Interactive matrix, represent the variance of described first Interactive matrix, represent the variance of the hidden eigenvectors matrix of described user, U _irepresent the hidden proper vector of i-th user, η represents learning efficiency;

According to formula upgrade the hidden eigenvectors matrix V of described destination object, wherein, $\frac{\∂E}{\∂V_j}=\underset{i=1}{\overset{N}{\Σ}}{I}_{ij}(S({\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j)U_i^T{V}_j-r_{ij})\·S({\stackrel{\‾}{u}}_i,{\stackrel{\‾}{c}}_j)U_i+{\mathrm{\λ}}_V{V}_j,$ ${\mathrm{\λ}}_V=\frac{{{\mathrm{\σ}}_R}^2}{{\mathrm{\σ}}_V^2},$ represent the variance of the hidden eigenvectors matrix of described destination object, V _jrepresent the hidden proper vector of a jth destination object.

9. based on an information recommending apparatus for familiarity, it is characterized in that, comprising:

Set up module, for according to M user to the operation behavior of N number of destination object, set up the first Interactive matrix, the element r in described first Interactive matrix _ijrepresent that i-th user is to the operation behavior of a jth destination object, wherein, 1≤i≤M and be integer, 1≤j≤N and be integer, described M be more than or equal to 2 integer, described N be more than or equal to 2 integer;

First determination module, for the characteristic model vector of each destination object of characteristic model vector sum according to each user, determine the familiarity of each user to different destination objects, wherein, the each user of characteristic model vector representation of each user and the relation of different label, each destination object of characteristic model vector representation of each destination object and the relation of different label;

Second determination module, for according to described first Interactive matrix and each user the familiarity to different destination objects determine the destination object recommended to each user.

10. device according to claim 9, is characterized in that, described first determination module comprises:

First determining unit, for the number of times adding often kind of label according to i-th user, adds the total degree of all labels, the size of described M, and described M user adds the total degree of often kind of label, determines the characteristic model vector of i-th user;

Second determining unit, for being added the number of times of often kind of label according to a jth destination object, is added the total degree of all labels, the size of described M, and described M user adds the total degree of often kind of label, determines the characteristic model vector of a jth destination object;

3rd determining unit, for the characteristic model vector of a jth destination object according to the characteristic model vector sum of described i-th user, determines that described i-th user is to the familiarity of a described jth destination object.