CN104765825A - Method and device for predicting social network links based on cooperative fusion theory - Google Patents

Abstract

The invention provides a method and device for predicting social network links based on cooperative fusion principles. The method comprises the steps: acquiring a set A of nearest neighbor sets T(a), T(b) of a node pair (a, b) in a social network; acquiring a likelihood function P1 (Ai) and a trusting function Be1 (Ai) of a node Ai in the set A; cooperatively fusing the likelihood function P1 (Ai) and the trusting function Be1 (Ai) and acquiring connectivity p (a, b) and reliability d (a, b) of the node pair (a, b); acquiring missing edge judging rules of a hidden link in the social network and excavating a hidden relation of the social network. The above method is applicable to the computing requirement of the large-scale social network; while the order of magnitudes of the computing complexity is not changed, accuracy of predicting results can be improved greatly; the predicting link scale can be adjusted according to error tolerance, thus precise control for predicting large-scale social network links is realized.

Description

Based on social networks link prediction method and the device of collaborative fusion principle

Technical field

The present invention relates to extensive social network analysis technical field, particularly relate to a kind of social networks link prediction method based on collaborative fusion principle and device.

Background technology

In recent years, along with the development in data analysis application field, the social network analysis based on theory of probability and Transmission dynamic obtains certain development, and is applied in some social relationships analyses and social platform, obtains huge using value.But the continuous collecting of social network data is restricted, there is the problems such as a large amount of disappearances, noise data in image data, has had a strong impact on reliability and the accuracy of analysis result.

Link prediction method in association area has two kinds, and one is local similar sex index, and this area scholar proposes a series of measuring similarity mode, the index of similarity that the basis of scales-free network proposes, and more representational have: (wherein k _xfor the degree of node x, k _ydegree for node y); (wherein Γ (x) represents the set of the consecutive point composition of x node, k _zdegree for node z), these class methods are by calculating relevance index between points, and search local extremum carries out link prediction.Be the global link Forecasting Methodology calculated based on correlation matrix product, the method that prediction precision is higher has based on L ⁺cosine method $s_{\mathrm{xy}}^{{\cos }^{+}}=\cos {(x,y)}^{+}=\frac{v_x^T{v}_y}{\left|v_x\right|\·\left|v_x\right|}=\frac{l_{\mathrm{xy}}^{+}}{\sqrt{l_{\mathrm{xy}}^{+}\·l_{\mathrm{xy}}^{+}}}$ (wherein $l_{\mathrm{xy}}^{+}=v_x^T{v}_y,$ Λ=diag (λ _x)) these class methods need to carry out a large amount of correlation matrix products, eigenwert, the operations such as proper vector in computation process.Local prediction method, accuracy and degree of accuracy are not high, lack global coherency, and global prediction method, need the various parameters calculating correlation matrix, higher relative to little network exact degree, but it is helpless relative to large-scale (1,000,000 point) social network Luoque, for this present situation, some scholars propose the random sectional pattern of two kinds of disposal route compromises, as local path index (Local Path Index, be called for short LP) algorithm, local random walk (Local Random Walk, be called for short LRW) algorithm, but still need the correlation matrix asking piecemeal in essence, need in blocking process to be similar to network, change the intrinsic property of former network.

Given this, the computation requirement how meeting extensive social networks carries out network link prediction, improves the accuracy predicted the outcome and becomes the current technical issues that need to address.

Summary of the invention

For solving above-mentioned technical matters, the invention provides a kind of social networks link prediction method based on collaborative fusion principle and device, the computation requirement of extensive social networks can be adapted to, the accuracy predicted the outcome can be increased substantially while the order of magnitude not changing computation complexity, according to fault tolerance adjustment prediction link scale, the accurate control of extensive social networks link prediction can be realized.

First aspect, the invention provides a kind of social networks link prediction method based on collaborative fusion principle, comprising:

Obtain social networks interior joint to arest neighbors set Γ (a) of (a, b), the common factor A of Γ (b);

Obtain the node A in described common factor A _ilikelihood function Pl (A _i) and belief function Bel (A _i);

By described likelihood function Pl (A _i) and belief function Bel (A _i) carry out collaborative fusion, obtain node to the Connected degree p (a, b) of (a, b) and confidence level d (a, b);

According to described Connected degree p (a, b), confidence level d (a, b) and default tolerable error, obtain the disappearance limit judgment rule of the potential link of described social networks, and excavate the potential relation of described social networks.

Alternatively, described common factor A is obtained by the first formulae discovery,

Described first formula is:

A＝Γ(a)∩Γ(b)，

Wherein, the arest neighbors set of the node a that Γ (a) is social networks, the arest neighbors set of the node b that Γ (b) is social networks;

Described likelihood function Pl (A _i) obtained by the second formulae discovery,

Described second formula is:

$\mathrm{Pl}\left(A_i\right)=\frac{1}{D\left(A_i\right)},$

Wherein, the node A in described common factor A _ilikelihood function Pl (A _i) be the node A of node a through described common factor A from social networks _iflow to the energy of node b, D (A _i) be node A _idegree;

Described belief function Bel (A _i) obtained by the 3rd formulae discovery,

Described 3rd formula is:

$\mathrm{Bel}\left(A_i\right)=\frac{K\left(A_i\right)}{N\left(A\right)},$

Wherein, belief function Bel (A _i) be the inner tight degree of described common factor A, N (A) is the node number in described common factor A, Κ (A _i) be node A _ithe number on the limit be connected with the node in described common factor A.

Alternatively, described by described likelihood function Pl (A _i) and belief function Bel (A _i) carry out collaborative fusion, obtain node to the Connected degree p (a, b) of (a, b) and confidence level d (a, b), comprising:

According to the node A in described common factor A _ilikelihood function Pl (A _i) and belief function Bel (A _i), generate the information of each node to (Pl (A _i), Bel (A _i));

To described information to (Pl (A _i), Bel (A _i)) carry out fusion reasoning, to Κ (A _i) sort from high-order to low order, obtain reliability density function X _i;

To X _ibe normalized guarantee and according to described reliability density function X _i, obtain Connected degree p (a, b) by the 5th formulae discovery, obtain confidence level d (a, b) by the 6th formulae discovery;

Wherein, described 5th formula is:

$p(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Pl}\left(A_i\right)\×X_i;$

Described 6th formula is:

$d(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Bel}\left(A_i\right)\×X_i.$

Alternatively, described reliability density function X _iobtained by the 4th formulae discovery,

Described 4th formula is:

$X_i=1-(\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j-(1-\frac{K\left(A_i\right)}{N\left(A\right)}\×\frac{\mathrm{Bel}\left(A_i\right)}{\underset{A_j\∈A}{\mathrm{\Σ}}\mathrm{Bel}\left(A_j\right)}-\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j)\×r^{-(K\left(A_j\right)-1)}),$

Wherein, X _iinitial value is set to Pl (A _i), r is default penalty factor, from Κ (A _i) most high-order starts to calculate, until upgraded all Κ (A _ithe node of)>=1.

Alternatively, described according to described Connected degree p (a, b), confidence level d (a, b) and default tolerable error, obtain the disappearance limit judgment rule of the potential link of described social networks, and excavate the potential relation of described social networks, comprising:

According to nodes all in social networks between Connected degree p (a, b) with confidence level d (a, b), generate alternative set B, alternative set B is the set of all limits composition of the undirected full-mesh figure of all node compositions in social networks, N is the number of network node of social networks, and the element number of alternative set B is

The Connected degree p (a, b) right according to nodes that there is not limit all in social networks and confidence level d (a, b), generates alternative set C, describedly alternatively gathers the subset that C is alternative set B;

To gather in node sort from big to small to pressing p (a, b), and before selecting generate sample set D, wherein, for default tolerable error;

Node in set D is sorted, before selecting from big to small to pressing d (a, b) the lower bound generating set F, described set F is the judgment rule of the potential link of described social networks;

According to the judgment rule of the potential link of described social networks, in set C, excavate potential link.

Second aspect, the invention provides a kind of social networks link prediction device based on collaborative fusion principle, comprising:

First acquisition module, for obtaining arest neighbors set Γ (a) of social networks interior joint to (a, b), the common factor A of Γ (b);

Second acquisition module, for obtaining the node A in described common factor A _ilikelihood function Pl (A _i) and belief function Bel (A _i);

3rd acquisition module, for by described likelihood function Pl (A _i) and belief function Bel (A _i) carry out collaborative fusion, obtain node to the Connected degree p (a, b) of (a, b) and confidence level d (a, b);

Excavate module, for according to described Connected degree p (a, b), confidence level d (a, b) and default tolerable error, obtain the disappearance limit judgment rule of the potential link of described social networks, and excavate the potential relation of described social networks according to described disappearance limit judgment rule.

Alternatively, described common factor A is obtained by the first formulae discovery,

Described first formula is:

A＝Γ(a)∩Γ(b)，

Wherein, the arest neighbors set of the node a that Γ (a) is social networks, the arest neighbors set of the node b that Γ (b) is social networks;

Described likelihood function Pl (A _i) obtained by the second formulae discovery,

Described second formula is:

$\mathrm{Pl}\left(A_i\right)=\frac{1}{D\left(A_i\right)},$

Wherein, the node A in described common factor A _ilikelihood function Pl (A _i) be the node A of node a through described common factor A from social networks _iflow to the energy of node b, D (A _i) be node A _idegree;

Described belief function Bel (A _i) obtained by the 3rd formulae discovery,

Described 3rd formula is:

$\mathrm{Bel}\left(A_i\right)=\frac{K\left(A_i\right)}{N\left(A\right)},$

Wherein, belief function Bel (A _i) be the inner tight degree of described common factor A, N (A) is the node number in described common factor A, Κ (A _i) be node A _ithe number on the limit be connected with the node in described common factor A.

Alternatively, described 3rd acquisition module, comprising:

First generation unit, for according to the node A in described common factor A _ilikelihood function Pl (A _i) and belief function Bel (A _i), generate the information of each node to (Pl (A _i), Bel (A _i));

First acquiring unit, for described information to (Pl (A _i), Bel (A _i)) carry out fusion reasoning, to Κ (A _i) sort from high-order to low order, obtain reliability density function X _i;

Second acquisition unit, for X _ibe normalized guarantee and according to described reliability density function X _i, obtain Connected degree p (a, b) by the 5th formulae discovery, obtain confidence level d (a, b) by the 6th formulae discovery;

Wherein, described 5th formula is:

$p(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Pl}\left(A_i\right)\×X_i;$

Described 6th formula is:

$d(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Bel}\left(A_i\right)\×X_i.$

Alternatively, described reliability density function X _iobtained by the 4th formulae discovery,

Described 4th formula is:

$X_i=1-(\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j-(1-\frac{K\left(A_i\right)}{N\left(A\right)}\×\frac{\mathrm{Bel}\left(A_i\right)}{\underset{A_j\∈A}{\mathrm{\Σ}}\mathrm{Bel}\left(A_j\right)}-\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j)\×r^{-(K\left(A_j\right)-1)}),$

Wherein, X _iinitial value is set to Pl (A _i), r is default penalty factor, from Κ (A _i) most high-order starts to calculate, until upgraded all Κ (A _ithe node of)>=1.

Alternatively, described excavation module, comprising:

Second generation unit, for according to nodes all in social networks between Connected degree p (a, b) with confidence level d (a, b), generate alternative set B, alternative set B is the set of all limits composition of the undirected full-mesh figure of all node compositions in social networks, and N is the number of network node of social networks, and the element number of alternative set B is

3rd generation unit, for according to the right Connected degree p (a, b) of nodes that there is not limit all in social networks and confidence level d (a, b), generates alternative set C, describedly alternatively gathers the subset that C is alternative set B;

4th generation unit, for gathering in node sort from big to small to pressing p (a, b), and before selecting generate sample set D, wherein, for default tolerable error;

5th generation unit, for sorting the node in set D from big to small, before selecting to pressing d (a, b) the lower bound generating set F, described set F is the judgment rule of the potential link of described social networks;

Excavate unit, for the judgment rule according to the potential link of described social networks, in set C, excavate potential link.

As shown from the above technical solution, social networks link prediction method based on collaborative fusion principle of the present invention and device, by obtaining social networks interior joint to arest neighbors set Γ (a) of (a, b), the common factor A of Γ (b); Obtain the node A in described common factor A _ilikelihood function Pl (A _i) and belief function Bel (A _i); By likelihood function Pl (A _i) and belief function Bel (A _i) carry out collaborative fusion, obtain node to the Connected degree p (a, b) of (a, b) and confidence level d (a, b); According to Connected degree p (a, b), confidence level d (a, b) with default tolerable error, obtain the disappearance limit judgment rule of the potential link of described social networks, and excavate the potential relation of described social networks, thus, the computation requirement of extensive social networks can be adapted to, the accuracy predicted the outcome can be increased substantially while the order of magnitude not changing computation complexity, according to fault tolerance adjustment prediction link scale, the accurate control of extensive social networks link prediction can be realized.

Accompanying drawing explanation

The schematic flow sheet of the social networks link prediction method based on collaborative fusion principle that Fig. 1 provides for first embodiment of the invention;

The principle schematic of the social networks link prediction method based on collaborative fusion principle that Fig. 2 provides for the embodiment of the present invention;

The schematic diagram of the collaborative fusion reasoning process that Fig. 3 embodiment of the present invention provides;

The social network data source generation figure that Fig. 4 provides for second embodiment of the invention;

The node that Fig. 5 provides for second embodiment of the invention is to (a, b) arest neighbors set common factor graph of a relation;

The node that Fig. 6 provides for second embodiment of the invention is to the collaborative fusion results view of (a, b);

The social networks link prediction result view based on collaborative fusion principle that Fig. 7 second embodiment of the invention provides;

The structural representation of the social networks link prediction device based on collaborative fusion principle that Fig. 8 provides for third embodiment of the invention.

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, clear, complete description is carried out to the technical scheme in the embodiment of the present invention, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiments.Based on embodiments of the invention, those of ordinary skill in the art, not making the every other embodiment obtained under creative work prerequisite, belong to the scope of protection of the invention.

First embodiment

Fig. 1 shows the schematic flow sheet of the social networks link prediction method based on collaborative fusion principle that first embodiment of the invention provides, and as Fig. 1 shows, the social networks link prediction method based on collaborative fusion principle of the present embodiment is as described below.

101, social networks interior joint is obtained to arest neighbors set Γ (a) of (a, b), the common factor A of Γ (b).

In a particular application, the A that occurs simultaneously described in the present embodiment is obtained by the first formulae discovery,

Described first formula is:

A＝Γ(a)∩Γ(b)，

Wherein, the arest neighbors set of the node a that Γ (a) is social networks, the arest neighbors set of the node b that Γ (b) is social networks.

102, the node A in described common factor A is obtained _ilikelihood function Pl (A _i) and belief function Bel (A _i).

In a particular application, likelihood function Pl (A described in the present embodiment _i) obtained by the second formulae discovery,

Described second formula is:

$\mathrm{Pl}\left(A_i\right)=\frac{1}{D\left(A_i\right)},$

Wherein, the node A in described common factor A _ilikelihood function Pl (A _i) be the node A of node a through described common factor A from social networks _iflow to the energy of node b, D (A _i) be node A _idegree;

Belief function Bel (A described in the present embodiment _i) obtained by the 3rd formulae discovery,

Described 3rd formula is:

$\mathrm{Bel}\left(A_i\right)=\frac{K\left(A_i\right)}{N\left(A\right)},$

Wherein, belief function Bel (A _i) be the inner tight degree of described common factor A, N (A) is the node number in described common factor A, Κ (A _i) be node A _ithe number on the limit be connected with the node in described common factor A.

103, by described likelihood function Pl (A _i) and belief function Bel (A _i) carry out collaborative fusion, obtain node to the Connected degree p (a, b) of (a, b) and confidence level d (a, b).

In a particular application, step 103 can comprise not shown step 103a-103c:

103a, according to the node A in described common factor A _ilikelihood function Pl (A _i) and belief function Bel (A _i), generate the information of each node to (Pl (A _i), Bel (A _i)).

103b, to described information to (Pl (A _i), Bel (A _i)) carry out fusion reasoning, to Κ (A _i) sort from high-order to low order, obtain reliability density function X _i.

103c, to X _ibe normalized guarantee and according to described reliability density function X _i, obtain Connected degree p (a, b) by the 5th formulae discovery, obtain confidence level d (a, b) by the 6th formulae discovery.

Wherein, described 5th formula is:

$p(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Pl}\left(A_i\right)\×X_i;$

Described 6th formula is:

$d(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Bel}\left(A_i\right)\×X_i;$

Described reliability density function X _iobtained by the 4th formulae discovery,

Described 4th formula is:

$X_i=1-(\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j-(1-\frac{K\left(A_i\right)}{N\left(A\right)}\×\frac{\mathrm{Bel}\left(A_i\right)}{\underset{A_j\∈A}{\mathrm{\Σ}}\mathrm{Bel}\left(A_j\right)}-\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j)\×r^{-(K\left(A_j\right)-1)}),$

Wherein, X _iinitial value is set to Pl (A _i), r is default penalty factor, from Κ (A _i) most high-order starts to calculate, until upgraded all Κ (A _ithe node of)>=1.

The default penalty factor of the present embodiment preferably can be set to 2.

104, according to described Connected degree p (a, b), confidence level d (a, b) and default tolerable error, obtain the disappearance limit judgment rule of the potential link of described social networks, and excavate the potential relation of described social networks.

In a particular application, step 104 can comprise not shown step 104a-104e:

104a, according to nodes all in social networks between Connected degree p (a, b) with confidence level d (a, b), generate alternative set B, alternative set B is the set of all limits composition of the undirected full-mesh figure of all node compositions in social networks, N is the number of network node of social networks, and the element number of alternative set B is

104b, according to the right Connected degree p (a, b) of nodes that there is not limit all in social networks and confidence level d (a, b), generate alternative set C, describedly alternatively gather the subset that C is alternative set B.

104c, will to gather in node sort from big to small to pressing p (a, b), and before selecting generate sample set D, wherein, for default tolerable error.

Will be understood that the default tolerable error of the present embodiment can adjust according to prediction scale, precision of prediction, also can arrange according to the demand in statistics experience and this network data acquisition process.

104d, by set D in node sort from big to small, before selecting to pressing d (a, b) the lower bound generating set F, described set F is the judgment rule of the potential link of described social networks.

104e, judgment rule according to the potential link of described social networks, excavate potential link in set C.

Will be understood that, the potential link accuracy now calculated and confidence level are all higher than (100-n) %.

Will be understood that, the Connected degree of the present embodiment is link and there is probability, namely similarity.

Fig. 2 shows the principle schematic of the social networks link prediction method based on collaborative fusion principle that the embodiment of the present invention provides, the schematic diagram of the collaborative fusion reasoning process that Fig. 3 embodiment of the present invention provides.

The determination methods of existing link prediction arranges threshold value, just think to there is potential disappearance limit when similarity is greater than this threshold value, present invention adds the confidence level of this similarity, because in extensive social networks, the change of link prediction to network character of mistake is very risky, therefore adopt similarity and confidence level jointly to differentiate, misjudgement probability can be reduced.

The social networks link prediction method based on collaborative fusion principle of the present embodiment, propose on the basis of the random sectional pattern of social networks, prior art general local link Forecasting Methodology effect is poor, and Global Algorithm calculated amount is excessive, this method quantized node to similarity while, connect tightr according to arest neighbors set common factor internal node, node to be analyzed between to there is the possibility on limit larger, respectively define likelihood function and belief function, link prediction is carried out by collaborative fusion method, while guarantee predictablity rate, calculated amount comparatively global prediction algorithm has and significantly reduces, the computation requirement of extensive social networks can be adapted to, the accuracy predicted the outcome can be increased substantially while the order of magnitude not changing computation complexity, can according to fault tolerance adjustment prediction link scale, realize the accurate control of extensive social networks link prediction.

Second embodiment

The present embodiment is with certain enterprise on July 16th, 2014, social networks G (the V of 49 call-information compositions during 30 employee works on the 17th, E) (see table 1, do not consider repeated communications) to carry out link prediction be example, Fig. 4 shows the social networks structural representation of the present embodiment data.The present embodiment based on collaborative fusion principle social networks link prediction method following steps 201-204 described in.

Table 1

201, obtain with certain enterprise on July 16th, 2014, the social networks interior joint of 49 call-informations compositions during 30 employee works on the 17th to arest neighbors set Γ (a) of (a, b), the common factor A of Γ (b).

In a particular application, can calculate:

Γ(a)＝{n0,n2,n5,n6,n15,n18,n23,}，

Γ(b)＝{n2,n11,n15,n18,n20,n23,n26,n27}；

Node is occured simultaneously can be calculated by A=Γ (a) ∩ Γ (b) the arest neighbors set of (a, b), A=Γ (a) ∩ Γ (b)={ n2, n15, n18, n23}.

202, the node A in described common factor A is obtained _ilikelihood function Pl (A _i) and belief function Bel (A _i).

In a particular application, first, to node to (a, b) and arest neighbors common factor element thereof carry out analyzing separately can obtain as shown in Figure 5 node to (a, b) arest neighbors set common factor graph of a relation, according to can obtain, node A ₁degree D (A _i) be 8, so its likelihood function in like manner can 4 some likelihood scores of set of computations A be respectively

Then, node A ₁three limits are had to be connected with the point in set A, therefore its belief function in like manner can four some belief functions of set of computations A be respectively

203, by described likelihood function Pl (A _i) and belief function Bel (A _i) carry out collaborative fusion, obtain node to the Connected degree p (a, b) of (a, b) and confidence level d (a, b).

The step 203 of the present embodiment specifically comprises:

Node is generated collaborative fused data source to the likelihood function of four points of the common factor A of (a, b) and belief function as shown in Figure 6;

According to the principle of collaborative fusion

$X_i=1-(\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j-(1-\frac{K\left(A_i\right)}{N\left(A\right)}\×\frac{\mathrm{Bel}\left(A_i\right)}{\underset{A_j\∈A}{\mathrm{\Σ}}\mathrm{Bel}\left(A_j\right)}-\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j)\×r^{-(K\left(A_j\right)-1)})$

Can calculate $X=[\frac{3}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2}],$

Can obtain after normalization $X=[\frac{1}{2},\frac{1}{6},\frac{1}{6},\frac{1}{6}];$

Can the descriptor of computing node to (a, b) calculate according to fusion formula of the present invention

Connected degree is $p(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Pl}\left(A_i\right)\×X_i=\frac{13}{80},$

Confidence level is $d(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Bel}\left(A_i\right)\×X_i=\frac{2}{3},$

So the quantificational description information of node to (a, b) is

204, according to described Connected degree p (a, b), confidence level d (a, b) and default tolerable error, obtain the disappearance limit judgment rule of the potential link of described social networks, and excavate the potential relation of described social networks.

The step 204 of the present embodiment can specifically comprise:

In the social networks of 30 some compositions, the element in alternative set B is this 435 nodes pair are traveled through, the fusion value that computing node is right and confidence level by algorithm of the present invention.

In alternative set C, the number of element is N _c=N _b-49=386;

Present case is in order to improve the accuracy of link prediction, and default tolerable error is 1%, needs to select set ( 49 articles of limits that data acquisition obtains) in Connected degree and confidence level be respectively the front value of 10% (5th), be respectively can obtain gathering D (the link prediction method generally based on similarity judges according to set D),

D＝{(n23,n26),(n7,n28),(n27,n28),(n26,n28),(n15,n27),(n15,n26)…}，

D’＝{(n15,n26),(n7,n28),(n26,n27),(n2,n20),(n15,n23),(n18,n23)…}，

So, F=D ∩ D'={ (n15, n26), (n7, n28) } be the prediction link of the tolerable error 1% of present networks.After prediction, design sketch as shown in Figure 7.

The social networks link prediction method based on collaborative fusion principle of the present embodiment, the computation requirement of extensive social networks can be adapted to, the accuracy predicted the outcome can be increased substantially while the order of magnitude not changing computation complexity, according to fault tolerance adjustment prediction link scale, the accurate control of extensive social networks link prediction can be realized.

3rd embodiment

Fig. 8 shows the structural representation of the social networks link prediction device based on collaborative fusion principle that third embodiment of the invention provides, as shown in Figure 8, the social networks link prediction device based on collaborative fusion principle of the present embodiment, comprising: the first acquisition module 81, second acquisition module 82, the 3rd acquisition module 83 and excavation module 84;

First acquisition module 81, for obtaining arest neighbors set Γ (a) of social networks interior joint to (a, b), the common factor A of Γ (b);

Second acquisition module 82, for obtaining the node A in described common factor A _ilikelihood function Pl (A _i) and belief function Bel (A _i);

3rd acquisition module 83, for by described likelihood function Pl (A _i) and belief function Bel (A _i) carry out collaborative fusion, obtain node to the Connected degree p (a, b) of (a, b) and confidence level d (a, b);

Excavate module 84, for according to described Connected degree p (a, b), confidence level d (a, b) and default tolerable error, obtain the disappearance limit judgment rule of the potential link of described social networks, and excavate the potential relation of described social networks according to described disappearance limit judgment rule.

Wherein, described common factor A is obtained by the first formulae discovery,

Described first formula is:

A＝Γ(a)∩Γ(b)，

Wherein, the arest neighbors set of the node a that Γ (a) is social networks, the arest neighbors set of the node b that Γ (b) is social networks;

Described likelihood function Pl (A) is obtained by the second formulae discovery,

Described second formula is:

$\mathrm{Pl}\left(A_i\right)=\frac{1}{D\left(A_i\right)},$

Wherein, the node A in described common factor A _ilikelihood function Pl (A _i) be the node A of node a through described common factor A from social networks _iflow to the energy of node b, D (A _i) be node A _idegree;

Described belief function Bel (A) is obtained by the 3rd formulae discovery,

Described 3rd formula is:

$\mathrm{Bel}\left(A_i\right)=\frac{K\left(A_i\right)}{N\left(A\right)},$

Wherein, belief function Bel (A) is the inner tight degree of described common factor A, and N (A) is the node number in described common factor A, Κ (A _i) be node A _ithe number on the limit be connected with the node in described common factor A.

In a particular application, described in the present embodiment, the 3rd acquisition module 83 can comprise not shown:

First generation unit, for according to the node A in described common factor A _ilikelihood function Pl (A _i) and belief function Bel (A _i), generate the information of each node to (Pl (A _i), Bel (A _i));

First acquiring unit, for described information to (Pl (A _i), Bel (A _i)) carry out fusion reasoning, to Κ (A _i) sort from high-order to low order, obtain reliability density function X _i;

Second acquisition unit, for X _ibe normalized guarantee and according to described reliability density function X _i, obtain Connected degree p (a, b) by the 5th formulae discovery, obtain confidence level d (a, b) by the 6th formulae discovery;

Wherein, described 5th formula is:

$p(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Pl}\left(A_i\right)\×X_i;$

Described 6th formula is:

$d(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Bel}\left(A_i\right)\×X_i;$

Described reliability density function X _iobtained by the 4th formulae discovery,

Described 4th formula is:

$X_i=1-(\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j-(1-\frac{K\left(A_i\right)}{N\left(A\right)}\×\frac{\mathrm{Bel}\left(A_i\right)}{\underset{A_j\∈A}{\mathrm{\Σ}}\mathrm{Bel}\left(A_j\right)}-\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j)\×r^{-(K\left(A_j\right)-1)}),$

Wherein, X _iinitial value is set to Pl (A _i), r is default penalty factor, from Κ (A _i) most high-order starts to calculate, until upgraded all Κ (A _ithe node of)>=1.

In a particular application, excavating module 84 described in the present embodiment can comprise not shown:

Second generation unit, for according to nodes all in social networks between Connected degree p (a, b) with confidence level d (a, b), generate alternative set B, alternative set B is the set of all limits composition of the undirected full-mesh figure of all node compositions in social networks, and N is the number of network node of social networks, and the element number of alternative set B is

3rd generation unit, for according to the right Connected degree p (a, b) of nodes that there is not limit all in social networks and confidence level d (a, b), generates alternative set C, describedly alternatively gathers the subset that C is alternative set B;

4th generation unit, for gathering in node sort from big to small to pressing p (a, b), and before selecting generate sample set D, wherein, for default tolerable error;

5th generation unit, for sorting the node in set D from big to small, before selecting to pressing d (a, b) the lower bound generating set F, described set F is the judgment rule of the potential link of described social networks;

Excavate unit, for the judgment rule according to the potential link of described social networks, in set C, excavate potential link.

The social networks link prediction device based on collaborative fusion principle of the present embodiment, the computation requirement of extensive social networks can be adapted to, the accuracy predicted the outcome can be increased substantially while the order of magnitude not changing computation complexity, according to fault tolerance adjustment prediction link scale, the accurate control of extensive social networks link prediction can be realized.

The social networks link prediction device based on collaborative fusion principle of the present embodiment, may be used for the technical scheme performing embodiment of the method shown in earlier figures 1, it realizes principle and technique effect is similar, repeats no more herein.

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 a social networks link prediction method for collaborative fusion principle, it is characterized in that, comprising:

Obtain social networks interior joint to arest neighbors set Γ (a) of (a, b), the common factor A of Γ (b);

Obtain the node A in described common factor A _ilikelihood function Pl (A _i) and belief function Bel (A _i);

By described likelihood function Pl (A _i) and belief function Bel (A _i) carry out collaborative fusion, obtain node to the Connected degree p (a, b) of (a, b) and confidence level d (a, b);

According to described Connected degree p (a, b), confidence level d (a, b) and default tolerable error, obtain the disappearance limit judgment rule of the potential link of described social networks, and excavate the potential relation of described social networks.

2. method according to claim 1, is characterized in that, described common factor A is obtained by the first formulae discovery,

Described first formula is:

A＝Γ(a)∩Γ(b)，

Wherein, the arest neighbors set of the node a that Γ (a) is social networks, the arest neighbors set of the node b that Γ (b) is social networks;

Described likelihood function Pl (A _i) obtained by the second formulae discovery,

Described second formula is:

$\mathrm{Pl}\left(A_i\right)=\frac{1}{D\left(A_i\right)},$

Wherein, the node A in described common factor A _ilikelihood function Pl (A _i) be the node A of node a through described common factor A from social networks _iflow to the energy of node b, D (A _i) be node A _idegree;

Described belief function Bel (A _i) obtained by the 3rd formulae discovery,

Described 3rd formula is:

$\mathrm{Bel}\left(A_i\right)=\frac{K\left(A_i\right)}{N\left(A\right)},$

Wherein, belief function Bel (A _i) be the inner tight degree of described common factor A, N (A) is the node number in described common factor A, Κ (A _i) be node A _ithe number on the limit be connected with the node in described common factor A.

3. method according to claim 1, is characterized in that, described by described likelihood function Pl (A _i) and belief function Bel (A _i) carry out collaborative fusion, obtain node to the Connected degree p (a, b) of (a, b) and confidence level d (a, b), comprising:

According to the node A in described common factor A _ilikelihood function Pl (A _i) and belief function Bel (A _i), generate the information of each node to (Pl (A _i), Bel (A _i));

To described information to (Pl (A _i), Bel (A _i)) carry out fusion reasoning, to Κ (A _i) sort from high-order to low order, obtain reliability density function X _i;

To X _ibe normalized guarantee and according to described reliability density function X _i, obtain Connected degree p (a, b) by the 5th formulae discovery, obtain confidence level d (a, b) by the 6th formulae discovery;

Wherein, described 5th formula is:

$p(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Pl}\left(A_i\right)\×X_i;$

Described 6th formula is:

$d(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Bel}\left(A_i\right)\×X_i.$

4. method according to claim 3, is characterized in that, described reliability density function X _iobtained by the 4th formulae discovery,

Described 4th formula is:

$X_i=1-(\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j-(1-\frac{K\left(A_i\right)}{N\left(A\right)}\×\frac{\mathrm{Bel}}{\underset{A_j\∈A}{\mathrm{\Σ}}\mathrm{Bel}\left(A_j\right)}-\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j)\×r^{-(K\left(A_i\right)-1)}),$

Wherein, X _iinitial value is set to Pl (A _i), r is default penalty factor, from Κ (A _i) most high-order starts to calculate, until upgraded all Κ (A _ithe node of)>=1.

5. method according to claim 1, it is characterized in that, described according to described Connected degree p (a, b), confidence level d (a, b) with default tolerable error, obtain the disappearance limit judgment rule of the potential link of described social networks, and excavate the potential relation of described social networks, comprising:

According to nodes all in social networks between Connected degree p (a, b) with confidence level d (a, b), generate alternative set B, alternative set B is the set of all limits composition of the undirected full-mesh figure of all node compositions in social networks, N is the number of network node of social networks, and the element number of alternative set B is

The Connected degree p (a, b) right according to nodes that there is not limit all in social networks and confidence level d (a, b), generates alternative set C, describedly alternatively gathers the subset that C is alternative set B;

To gather in node sort from big to small to pressing p (a, b), and before selecting generate sample set D, wherein, for default tolerable error;

Node in set D is sorted, before selecting from big to small to pressing d (a, b) the lower bound generating set F, described set F is the judgment rule of the potential link of described social networks;

According to the judgment rule of the potential link of described social networks, in set C, excavate potential link.

6., based on a social networks link prediction device for collaborative fusion principle, it is characterized in that, comprising:

First acquisition module, for obtaining arest neighbors set Γ (a) of social networks interior joint to (a, b), the common factor A of Γ (b);

Second acquisition module, for obtaining the node A in described common factor A _ilikelihood function Pl (A _i) and belief function Bel (A _i);

3rd acquisition module, for by described likelihood function Pl (A _i) and belief function Bel (A _i) carry out collaborative fusion, obtain node to the Connected degree p (a, b) of (a, b) and confidence level d (a, b);

Excavate module, for according to described Connected degree p (a, b), confidence level d (a, b) and default tolerable error, obtain the disappearance limit judgment rule of the potential link of described social networks, and excavate the potential relation of described social networks according to described disappearance limit judgment rule.

7. device according to claim 6, is characterized in that, described common factor A is obtained by the first formulae discovery,

Described first formula is:

A＝Γ(a)∩Γ(b)，

Wherein, the arest neighbors set of the node a that Γ (a) is social networks, the arest neighbors set of the node b that Γ (b) is social networks;

Described likelihood function Pl (A _i) obtained by the second formulae discovery,

Described second formula is:

$\mathrm{Pl}\left(A_i\right)=\frac{1}{D\left(A_i\right)},$

Wherein, the node A in described common factor A _ilikelihood function Pl (A _i) be the node A of node a through described common factor A from social networks _iflow to the energy of node b, D (A _i) be node A _idegree;

Described belief function Bel (A _i) obtained by the 3rd formulae discovery,

Described 3rd formula is:

$\mathrm{Bel}\left(A_i\right)=\frac{K\left(A_i\right)}{N\left(A\right)},$

Wherein, belief function Bel (A _i) be the inner tight degree of described common factor A, N (A) is the node number in described common factor A, Κ (A _i) be node A _ithe number on the limit be connected with the node in described common factor A.

8. device according to claim 6, is characterized in that, described 3rd acquisition module, comprising:

First generation unit, for according to the node A in described common factor A _ilikelihood function Pl (A _i) and belief function Bel (A _i), generate the information of each node to (Pl (A _i), Bel (A _i));

First acquiring unit, for described information to (Pl (A _i), Bel (A _i)) carry out fusion reasoning, to Κ (A _i) sort from high-order to low order, obtain reliability density function X _i;

Second acquisition unit, for X _ibe normalized guarantee and according to described reliability density function X _i, obtain Connected degree p (a, b) by the 5th formulae discovery, obtain confidence level d (a, b) by the 6th formulae discovery;

Wherein, described 5th formula is:

$p(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Pl}\left(A_i\right)\×X_i;$

Described 6th formula is:

$d(a,b)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Bel}\left(A_i\right)\×X_i.$

9. device according to claim 8, is characterized in that, described reliability density function X _iobtained by the 4th formulae discovery,

Described 4th formula is:

$X_i=1-(\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j-(1-\frac{K\left(A_i\right)}{N\left(A\right)}\×\frac{\mathrm{Bel}}{\underset{A_j\∈A}{\mathrm{\Σ}}\mathrm{Bel}\left(A_j\right)}-\underset{K\left(A_i\right)>K\left(A_j\right)}{\mathrm{\Σ}}{X}_j)\×r^{-(K\left(A_i\right)-1)}),$

Wherein, X _iinitial value is set to Pl (A _i), r is default penalty factor, from Κ (A _i) most high-order starts to calculate, until upgraded all Κ (A _ithe node of)>=1.

10. method according to claim 1, is characterized in that, described excavation module, comprising:

Second generation unit, for according to nodes all in social networks between Connected degree p (a, b) with confidence level d (a, b), generate alternative set B, alternative set B is the set of all limits composition of the undirected full-mesh figure of all node compositions in social networks, and N is the number of network node of social networks, and the element number of alternative set B is

3rd generation unit, for according to the right Connected degree p (a, b) of nodes that there is not limit all in social networks and confidence level d (a, b), generates alternative set C, describedly alternatively gathers the subset that C is alternative set B;

4th generation unit, for gathering in node sort from big to small to pressing p (a, b), and before selecting generate sample set D, wherein, for default tolerable error;

5th generation unit, for sorting the node in set D from big to small, before selecting to pressing d (a, b) the lower bound generating set F, described set F is the judgment rule of the potential link of described social networks;

Excavate unit, for the judgment rule according to the potential link of described social networks, in set C, excavate potential link.