CN106301888A - Based on core node and the network community division method of community's convergence strategy - Google Patents

Abstract

The invention discloses a kind of based on core node with the network community division method of community's convergence strategy, mainly solve the resolution problem that in existing community blending algorithm, object function itself exists.Implementation step is: 1) reads in a width network S, and generates adjacency matrix A corresponding for network S；2) node degree of each node in network, and Network Search core node set C are calculated according to adjacency matrix A；3) network node tag set f ' is updated according to similarity function value F；4) Web Community tag set f is obtained according to network node tag set f '；5) based on improvement block density increment Delta D, the community in current network is iterated community to merge, exports final network node tag set f_z.The present invention has the advantage that nodal information utilization rate is high and community's resolution is high, can be used for community's detection.

Description

Based on core node and the network community division method of community's convergence strategy

Technical field

The invention belongs to network and divide field, further relate to a kind of network community division method, can be used for community's inspection Survey.

Background technology

In the modern life, there is the existence of network structure almost everywhere, and most complication system is all with net The form of network presents, such as metabolism net, the Internet, Email net and friends and family's net etc..To complex network The research of qualitative and quantitative characteristic, contribute to disclosing the different complication systems under complex network model represents generally exist Universal law, significant in the subjects such as bioscience, social sciences.

The structure that the relation being made up of the node of heterogeneity, type in network is abundant is referred to as community.Community's internal relations Dense, and the community structure that between different communities node, relation is sparse is one of feature of complex network.Community detection method Research contributes to understanding that topology of networks and network dynamics behavior are extremely important because of it.It has been proposed that A lot of methods carry out the division of community to network, and wherein one of study hotspot is Web Community based on community's convergence strategy Division methods.

Community's fusion method is broadly divided into following basic step: first, and each node in network is considered as an independence Community；Then, available target function value, usually localized mode lumpiness function after every pair of community merges is calculated；Finally, By relatively target function value obtained in the previous step, it may be judged whether corresponding Liang Ge community is merged.By following several times Ring iterative, it is thus achieved that community division result.The resolution problem that this method exists due to object function itself so that the knot of division Fruit is easily trapped into local optimum, inefficiency.

Summary of the invention

Present invention aims to above-mentioned the deficiencies in the prior art, propose one and merge plan based on core node and community Network community division method slightly, to improve object function resolution, reduces the situation being absorbed in local optimum, improves community and divides Efficiency.

For achieving the above object, the technical solution used in the present invention includes the following:

(1) read in a width network S, and generate adjacency matrix A corresponding for network S；

(2) node degree of each node in network is calculated according to adjacency matrix A, and according to the node of node each in network Degree Network Search core node set C；

(3) be assigned to the unique label i of i-th node one in network, i ∈ 1,2 ..., n}, n represent nodes Number, calculate similarity function value F of each core node neighbor node all with it in network core node set C, select this Functional value F is more than or equal to given threshold value F_tNeighbor node corresponding to=1, and by the tag update of neighbor node selected for The label of its core node connected, obtains the network node tag set f '={ f updated₁′,f₂′,...,f_i′,...,f_n', Wherein f_i' represent the label of i-th node in network；

(4) by network node tag set f '={ f₁′,f₂′,...,f_i′,...,f_n' in node corresponding to same label Form a community, obtain current network community tag set f (1,1)={ f₁(1,1),f₂(1,1),...,f_u(1,1),..., f_h(1)(1,1) }, and record the node comprised in each community, wherein, f_u(1,1) mark of u community in current network is represented Sign, u ∈ 1,2 ..., and h (1) }, h (1) represents the community's number in current network；

(5) it is iterated merging to the community in current network based on improvement block density increment Delta D:

5a) loop initialization number of times g=1；

5b) statistics current network community tag set f (g, 1)={ f₁(g,1),f₂(g,1),...,f_u(g,1),..., f_h(g)(g, 1) } in each community and extraneous linking number l_o(g, 1)={ l_o1(g,1),l_o2(g,1),...,l_ou(g, 1),...,l_oh(g)(g, 1) }, wherein f_u(g, 1) represents in network the u community label when circulating for the g time, l_ou(g, 1) table Show in network the u community when the g time circulation with extraneous linking number, u ∈ 1,2 ..., h (g), h (g) represents network In the g time circulation time community's number；

5c) by the community in current network community tag set f (g, 1) by the linking number l with the external world_o(g, 1) descending Arrangement, obtains matrix M (g) of h (g) row 2 row；

5d) initialize community and merge iterations t=1；

5e) respectively statistics current network community tag set f (g, t) in each community node degree sum d (g, t), society Linking number l between district's interior nodes_i(g, t), community and extraneous linking number l_o(g, t), between community u and its neighbours community v Linking number l_uv(g, t), wherein:

F (g, t)={ f₁(g,t),f₂(g,t),...,f_u(g,t),...,f_h(g)(g, t) }, f_u(g t) represents in network the U community label when the t time iteration of the g time circulation,

D (g, t)={ d₁(g,t),d₂(g,t),...,d_u(g,t),...,d_h(g)(g, t) }, d_u(g t) represents in network the U community node degree sum when the t time iteration of the g time circulation,

l_i(g, t)={ l_i1(g,t),l_i2(g,t),...,l_iu(g,t),...,l_ih(g)(g, t) }, l_iu(g t) represents network In the u community linking number between community's interior nodes when the t time iteration of the g time circulation,

l_o(g, t)={ l_o1(g,t),l_o2(g,t),...,l_ou(g,t),...,l_oh(g)(g, t) }, l_ou(g t) represents network In the u community when the t time iteration of the g time circulation with the linking number in the external world,

l_uv(g, t)={ l_uv1(g,t),l_uv2(g,t),...,l_uvk(g,t)(g, t) }, l_uv(g t) represents that in network, u is individual Between community and its neighbours community v the g time circulation the t time iteration time linking number, k (g, t) expression network in follow for the g time The number of interconnective community pair during the t time iteration of ring；

5f) calculate the improvement block density increment Delta of gained after community u and its neighbours community v in current network merges D_uv:

${\mathrm{\ΔD}}_{uv}=\frac{\[l_{iu}-l_{ou}-l_{ov}+3l_{uv}\]}{d_u+d_v}-\[\frac{l_{iu}-l_{ou}}{d_u}+\frac{l_{iv}-l_{ov}}{d_v}\],$

Wherein, l_iuRepresent the linking number between community's u interior nodes, l_ouRepresent community u and extraneous linking number, l_ovTable Show community v and extraneous linking number, l_uvRepresent the linking number between community u and v, d_uRepresent the node of community's u interior nodes Degree sum, d_vRepresent the node degree sum of community's v interior nodes, l_ivRepresent the linking number between community's v interior nodes；

5g) according to improving block density increment Delta D_uvObtain the t community p neighbours all with it community in matrix M (g) to melt Increment set Δ D is merged in the p community closing gained_p, and find set Δ D_pIn neighbours community q corresponding to maximum；

5h) according to improving block density increment Delta D_uvThe q community obtaining community q neighbours all with it community fusion gained melts Close increment set Δ D_q；

5i) increment set Δ D is merged in p community_pIncrement set Δ D is merged with q community_qCompare: if Δ D_pIn Big value is more than or equal to Δ D_qIn maximum, then community p and community q are merged, will current network community tag set f (g, T) in, the label of community p changes the label of neighbours community q into, otherwise, does not merges；

5j) iterations t is merged in given maximum community_max=h (g), it is judged that whether existing community merges iterations t Iterations t is merged in the community reaching maximum_maxIf reaching, then terminate iteration, and perform 5k), otherwise, t=t+1, return to 5e) carry out next iteration；

5k) given maximum cycle g_max=100, it is judged that whether current cycle time g reaches maximum cycle g_max, If reaching, then terminate circulation, and by final Web Community tag set f (g_max,t_maxCommunity in) is launched into final network joint Point tag set f_z={ f_z1,f_z2,...,f_zi,...,f_znOutput, wherein:

f(g_max,t_max)={ f₁(g_max,t_max),f₂(g_max,t_max),...,f_u(g_max,t_max),...,f_h(gmax)(g_max, t_max),

f_u(g_max,t_max) represent reach maximum cycle g_maxIterations t is merged with maximum community_maxTime network society The label of u community in district's tag set, u ∈ 1,2 ..., h (g_max), h (g_max) represent in network and reach largest loop Number of times g_maxTime community's number, f_ziRepresent and reach maximum cycle g_maxTime network node tag set in i-th node Label, i ∈ 1,2 ..., n}, n represent the interstitial content in network, otherwise, g=g+1, return to 5b) follow next time Ring.

Compared with prior art there is advantages below in the present invention:

1. the present invention is before community's fusion method, is primarily based on core node and network is divided by node similarity, Can more effectively utilize nodal information.

2. the present invention is in community's fusion process, propose one improve block density increment function as object function, change It is apt to the resolution problem that modularity function itself exists, has decreased the situation being absorbed in local optimum, improve community and divide effect Rate.

Accompanying drawing explanation

Fig. 1 is the karate instance graph that the present invention uses；

Fig. 2 is the flowchart of the present invention；

Fig. 3 is the present invention final division result figure to Fig. 1；

Fig. 4 is the existing community blending algorithm final division result figure to Fig. 1.

Detailed description of the invention

With reference to Fig. 1, the karate instance graph that the present invention uses is in karate club of research university by W.W.Zachary The network built during community relations between member.This club's network, due to the difference between club director and coach, divides It is cleaved into two disjoint corporations.Karate network is made up of 34 clubbites, as the node in network, and each member Between line have 78, as the line between nodes.In Fig. 1,1 to 34 represent the node in karate network respectively Numbering, rhombus and the circular Liang Lei community representing natural division respectively.

Below in conjunction with Fig. 2, the present invention is embodied as step to be described in further detail.

With reference to Fig. 2, a kind of based on core node with the network community division method of community's convergence strategy, comprise the steps:

Step 1, reads in a width network S, and generates adjacency matrix A corresponding for network S；

Adjacency matrix A is expressed as follows:

$A=\left[\begin{array}{ccccc}{a}_{11}& .& .& .& a_{1n}\\ .& .& .& .& .\\ .& .& a_{ij}& .& .\\ .& .& .& .& .\\ a_{n1}& .& .& .& a_{nn}\end{array}\right],$

Wherein, a_ijThe i-th row jth column element in expression adjacency matrix A, i ∈ 1,2 ..., and n}, represent appointing in network Meaning node, j ∈ 1,2 ..., and n}, represent the arbitrary node in network, n represents the number of nodes,

In an embodiment of the present invention, read in karate instance graph, the number n=34 of its nodes, generate empty-handed Road instance graph corresponding 34 × 34 adjacency matrix A as follows:

$A=\left[\begin{array}{cccccccc}0& 1& 1& .& .& .& .& 0\\ 1& 0& 1& .& .& .& .& 0\\ 1& 1& 0& .& .& .& .& 0\\ 0& 0& 0& .& .& .& .& 0\\ .& .& .& .& .& .& .& .\\ .& .& .& .& a_{ij}& .& .& .\\ .& .& .& .& .& .& .& .\\ 0& 0& 0& .& .& .& 0& 0\end{array}\right].$

Step 2, calculates the node degree of each node in network according to adjacency matrix A, and according to node each in network Node degree Network Search core node set C；

2a) calculate the node degree d of i-th node in network_i:

$d_i=\underset{j=\mathrm{1...}n}{\Σ}{a}_{ij},$

Wherein, a_ijRepresenting the element in adjacency matrix A, n represents the interstitial content in network；

2b) node degree of node neighbor node all with it each in network is compared, and by node degree higher than it The node of each neighbor node elects core node as, all core nodes composition core node set C selected.

In embodiments of the present invention, network core node set C found in karate network is by node 1 and node 34 groups Become.

Step 3, calculates similarity function value F of each core node neighbor node all with it in network core node set C, And according to the network node tag set f '={ f of similarity function value F renewal₁′,f₂′,...,f_i′,...,f_n', wherein f_i′ The label of i-th node in expression network, i ∈ 1,2 ..., n}, n represent the number of nodes.

3a) be assigned to the unique label i of i-th node one in network, i ∈ 1,2 ..., n}, n represent nodes Number；

3b) according to the phase of each core node in RA exponential function calculating network core node set C with its all neighbor nodes Like degree functional value F:

$F=\underset{c_o\∈N\left(i\right)\∩N\left(j\right)}{\Σ}\frac{1}{d\left(c_o\right)},$

Wherein, i, j represent the node that any two in network is different, c respectively_oRepresent i-th node and jth node Common neighbours, ∈ represents and belongs to symbol, and N (i) represents the neighborhood of i-th node, and ∩ represents and intersects operation, and N (j) represents The neighborhood of jth node, d (c_o) represent node c_oNode degree；

3c) given threshold value F_t=1, select functional value F more than or equal to given threshold value F_tCorresponding neighbor node, and will choosing The tag update of the neighbor node gone out is the label of connected core node, obtain update network node tag set f '= {f₁′,f₂′,...,f_i′,...,f_n′}。

In embodiments of the present invention, the node label set f '={ f of the karate network updated is obtained₁′,f₂′,..., f_i′,...,f_n', wherein, the tag update of node 2 and node 4 is 1, and the tag update of node 33 is 34.

Step 4, by network node tag set f '={ f₁′,f₂′,...,f_i′,...,f_n' in same label corresponding One community of node composition, obtains current network community tag set f (1,1)={ f₁(1,1),f₂(1,1),...,f_u(1, 1),...,f_h(1)(1,1) }, and record the node comprised in each community, wherein, f_u(1,1) the u society in current network is represented The label in district, u ∈ 1,2 ..., and h (1) }, h (1) represents the community's number in current network；

In embodiments of the present invention, in karate network, existing community number h (1) is 31.

Step 5, is iterated community's fusion based on improving block density increment Delta D, obtains the community in current network Final community division result.

Implement step as follows:

5a) loop initialization number of times g=1；

5b) statistics current network community tag set f (g, 1)={ f₁(g,1),f₂(g,1),...,f_u(g,1),..., f_h(g)(g, 1) } in each community and extraneous linking number l_o(g, 1)={ l_o1(g,1),l_o2(g,1),...,l_ou(g, 1),...,l_oh(g)(g, 1) }, wherein f_u(g, 1) represents in network the u community label when circulating for the g time, l_ou(g, 1) table Show in network the u community when the g time circulation with extraneous linking number, u ∈ 1,2 ..., h (g), h (g) represents network In the g time circulation time community's number；

5c) by the community in current network community tag set f (g, 1) by the linking number l with the external world_o(g, 1) descending Arrangement, obtains matrix M (g) of h (g) row 2 row；

Matrix M (g) is expressed as follows:

$M\left(g\right)=\left[\begin{array}{cc}{M}_{11}\left(g\right)& .\\ .& .\\ M_{u1}\left(g\right)& .\\ .& .\\ M_{h\left(g\right)1}\left(g\right)& M_{h\left(g\right)2}\left(g\right)\end{array}\right],$

Wherein, M_u1U row the 1st column element in (g) representing matrix M (g), u ∈ 1,2 ..., and h (g) }, h (g) represents network In the g time circulation time community's number；

In embodiments of the present invention, during the 1st circulation, the matrix M (1) of 31 row 2 row that karate network is corresponding is as follows:

$M\left(1\right)=\left[\begin{array}{cc}31& 27\\ 1& 25\\ .& .\\ .& .\\ .& .\\ 25& 2\\ 10& 1\end{array}\right].$

5d) initialize community and merge iterations t=1；

5e) respectively statistics current network community tag set f (g, t) in each community node degree sum d (g, t), society Linking number l between district's interior nodes_i(g, t), community and extraneous linking number l_o(g, t), between community u and its neighbours community v Linking number l_uv(g, t), wherein:

F (g, t)={ f₁(g,t),f₂(g,t),...,f_u(g,t),...,f_h(g)(g, t) }, f_u(g t) represents in network the U community label when the t time iteration of the g time circulation,

D (g, t)={ d₁(g,t),d₂(g,t),...,d_u(g,t),...,d_h(g)(g, t) }, d_u(g t) represents in network the U community node degree sum when the t time iteration of the g time circulation,

l_i(g, t)={ l_i1(g,t),l_i2(g,t),...,l_iu(g,t),...,l_ih(g)(g, t) }, l_iu(g t) represents network In the u community linking number between community's interior nodes when the t time iteration of the g time circulation,

l_o(g, t)={ l_o1(g,t),l_o2(g,t),...,l_ou(g,t),...,l_oh(g)(g, t) }, l_ou(g t) represents network In the u community when the t time iteration of the g time circulation with the linking number in the external world,

l_uv(g, t)={ l_uv1(g,t),l_uv2(g,t),...,l_uvk(g,t)(g, t) }, l_uv(g t) represents that in network, u is individual Between community and its neighbours community v the g time circulation the t time iteration time linking number, k (g, t) expression network in follow for the g time The number of interconnective community pair during the t time iteration of ring；

5f) calculate the improvement block density increment Delta of gained after community u and its neighbours community v in current network merges D_uv:

${\mathrm{\ΔD}}_{uv}=\frac{\[l_{iu}-l_{ou}-l_{ov}+3l_{uv}\]}{d_u+d_v}-\[\frac{l_{iu}-l_{ou}}{d_u}+\frac{l_{iv}-l_{ov}}{d_v}\],$

Wherein, l_iuRepresent the linking number between community's u interior nodes, l_ouRepresent community u and extraneous linking number, l_ovTable Show community v and extraneous linking number, l_uvRepresent the linking number between community u and v, d_uRepresent the node of community's u interior nodes Degree sum, d_vRepresent the node degree sum of community's v interior nodes, l_ivRepresent the linking number between community's v interior nodes；

5g) according to improving block density increment Delta D_uvObtain the t community p neighbours all with it community in matrix M (g) to melt Increment set Δ D is merged in the p community closing gained_p, and find set Δ D_pIn neighbours community q corresponding to maximum；

In embodiments of the present invention, as a example by the in karate network the 1st community 31, according to improving block density increment Delta D_uvObtain the community increment set Δ D obtained by neighbours all with it community of community 31 merges₃₁, find set Δ D₃₁In The neighbours community of big value correspondence is community 13.

5h) according to improving block density increment Delta D_uvThe q community obtaining community q neighbours all with it community fusion gained melts Close increment set Δ D_q；

In embodiments of the present invention, according to improving block density increment Delta D_uvObtain neighbours all with it community of community 13 to melt Close the community increment set Δ D of gained₁₃。

5i) increment set Δ D is merged in p community_pIncrement set Δ D is merged with q community_qCompare: if Δ D_pIn Big value is more than or equal to Δ D_qIn maximum, then community p and community q are merged, will current network community tag set f (g, T) in, the label of community p changes the label of neighbours community q into, otherwise, does not merges；

In embodiments of the present invention, set Δ D₃₁In maximum more than or equal to set Δ D₁₃In maximum, so, Community 31 is merged with community 13, will Web Community's division result f (g, t) in the label of community 31 change the mark of community 13 into Sign.

5j) iterations t is merged in given maximum community_max=h (g), it is judged that whether existing community merges iterations t Iterations t is merged in the community reaching maximum_maxIf reaching, then terminate iteration, and perform 5k), otherwise, t=t+1, return to 5e) carry out next iteration；

5k) given maximum cycle g_max=100, it is judged that whether current cycle time g reaches maximum cycle g_max, If reaching, then terminate circulation, and by final Web Community tag set f (g_max,t_maxCommunity in) is launched into final network joint Point tag set f_z={ f_z1,f_z2,...,f_zi,...,f_znOutput, wherein:

f(g_max,t_max)={ f₁(g_max,t_max),f₂(g_max,t_max),...,f_u(g_max,t_max),...,f_h(gmax)(g_max, t_max),

f_u(g_max,t_max) represent reach maximum cycle g_maxIterations t is merged with maximum community_maxTime network society The label of u community in district's tag set, u ∈ 1,2 ..., h (g_max), h (g_max) represent in network and reach largest loop Number of times g_maxTime community's number, f_ziRepresent and reach maximum cycle g_maxTime network node tag set in i-th node Label, i ∈ 1,2 ..., n}, n represent the interstitial content in network, otherwise, g=g+1, return to 5b) follow next time Ring.

The effect of the present invention can be further illustrated by following emulation experiment:

1. simulated conditions:

The present invention uses Matlab R2014a software being configured to core 2 2.4GHZ, internal memory 2G, WINDOWS 7 system Computer on carry out.

2. emulation content and result:

Emulation 1, carries out community's division by the present invention to the karate instance graph shown in Fig. 1, result such as Fig. 3.

Emulation 2, divides karate network with existing community blending algorithm, result such as Fig. 4.

Both approaches is all that Fig. 1 is divided into three classes, wherein rhombus, and circular and triangle represents three after dividing Classification.

From figure 3, it can be seen that the present invention is identical with community's classification of the self-assembling formation in Fig. 1, node 29 is correctly drawn Assign in the classification of circular representative.

From fig. 4, it can be seen that existing community blending algorithm is different, by node from the classification of the community of the self-assembling formation in Fig. 1 29 mistakes have been divided in the classification that rhombus represents.

Comparison diagram 3 and Fig. 4 is it can be seen that the community division method of the present invention is the most accurate to the division of community in network.

To sum up, the present invention propose based on core node and the network community division method of community's convergence strategy, first to net Network divides based on core node and node similarity, and proposes a kind of block density increment function that improves as object function, It is iterated community to merge, is effectively utilized nodal information, overcomes existing community blending algorithm and be easily trapped into local optimum Shortcoming, improves the resolution problem that modularity function itself exists.

Claims (5)

1., based on core node and a network community division method for community's convergence strategy, comprise the steps:

(1) read in a width network S, and generate adjacency matrix A corresponding for network S；

(2) calculate the node degree of each node in network according to adjacency matrix A, and look into according to the node degree of node each in network Look for network core node set C；

(3) be assigned to the unique label i of i-th node one in network, i ∈ 1,2 ..., n}, n represent the number of nodes Mesh, calculates similarity function value F of each core node neighbor node all with it in network core node set C, selects this function Value F is more than or equal to given threshold value F_tNeighbor node corresponding to=1, and by the tag update of the neighbor node selected for connecting with it The label of the core node connect, obtains the network node tag set f '={ f updated₁′,f₂′,...,f_i′,...,f_n', wherein f_i' represent the label of i-th node in network；

(4) by network node tag set f '={ f₁′,f₂′,...,f_i′,...,f_n' in same label corresponding node composition One community, obtains current network community tag set f (1,1)={ f₁(1,1),f₂(1,1),...,f_u(1,1),...,f_h(1) (1,1) }, and record the node comprised in each community, wherein, f_u(1,1) represents the label of u community, u in current network ∈ 1,2 ..., and h (1) }, h (1) represents the community's number in current network；

(5) it is iterated merging to the community in current network based on improvement block density increment Delta D:

5a) loop initialization number of times g=1；

5b) statistics current network community tag set f (g, 1)={ f₁(g,1),f₂(g,1),...,f_u(g,1),...,f_h(g)(g, 1) each community and extraneous linking number l in }_o(g, 1)={ l_o1(g,1),l_o2(g,1),...,l_ou(g,1),...,l_oh(g) (g, 1) }, wherein f_u(g, 1) represents in network the u community label when circulating for the g time, l_ou(g, 1) represents u in network Individual community when the g time circulation with extraneous linking number, u ∈ 1,2 ..., h (g), h (g) represents in network and circulates for the g time Time community's number；

5c) by the community in current network community tag set f (g, 1) by the linking number l with the external world_o(g, 1) descending, Obtain matrix M (g) of h (g) row 2 row；

5d) initialize community and merge iterations t=1；

5e) respectively statistics current network community tag set f (g, t) in node degree sum d of each community (g, t), in community Internodal linking number l_i(g, t), community and extraneous linking number l_o(g, company t), between community u and its neighbours community v Meet number l_uv(g, t), wherein:

F (g, t)={ f₁(g,t),f₂(g,t),...,f_u(g,t),...,f_h(g)(g, t) }, f_u(g t) represents that in network, u is individual Community's label when the t time iteration of the g time circulation,

D (g, t)={ d₁(g,t),d₂(g,t),...,d_u(g,t),...,d_h(g)(g, t) }, d_u(g t) represents that in network, u is individual Community's node degree sum when the t time iteration of the g time circulation,

l_i(g, t)={ l_i1(g,t),l_i2(g,t),...,l_iu(g,t),...,l_ih(g)(g, t) }, l_iu(g t) represents in network the U community linking number between community's interior nodes when the t time iteration of the g time circulation,

l_o(g, t)={ l_o1(g,t),l_o2(g,t),...,l_ou(g,t),...,l_oh(g)(g, t) }, l_ou(g t) represents in network the U community when the t time iteration of the g time circulation with the linking number in the external world,

l_uv(g, t)={ l_uv1(g,t),l_uv2(g,t),...,l_uvk(g,t)(g, t) }, l_uv(g t) represents the u community in network And linking number when the t time iteration of the g time circulation between its neighbours community v, (g t) represents and circulates the in network for the g time k The number of interconnective community pair during t iteration；

5f) calculate improvement block density increment Delta D of gained after community u and its neighbours community v in current network merges_uv:

${\mathrm{\ΔD}}_{uv}=\frac{\[l_{iu}-l_{ou}-l_{ov}+3l_{uv}\]}{d_u+d_v}-\[\frac{l_{iu}-l_{ou}}{d_u}+\frac{l_{iv}-l_{ov}}{d_v}\],$

Wherein, l_iuRepresent the linking number between community's u interior nodes, l_ouRepresent community u and extraneous linking number, l_ovRepresent society District v and extraneous linking number, l_uvRepresent the linking number between community u and v, d_uRepresent community u interior nodes node degree it With, d_vRepresent the node degree sum of community's v interior nodes, l_ivRepresent the linking number between community's v interior nodes；

5g) according to improving block density increment Delta D_uvObtain the t community p neighbours all with it community in matrix M (g) and merge institute Increment set Δ D is merged in the p community obtained_p, and find set Δ D_pIn neighbours community q corresponding to maximum；

5h) according to improving block density increment Delta D_uvObtain community q neighbours all with it community and merge the q community fusion increasing of gained Duration set Δ D_q；

5i) increment set Δ D is merged in p community_pIncrement set Δ D is merged with q community_qCompare: if Δ D_pIn maximum More than or equal to Δ D_qIn maximum, then community p and community q are merged, will current network community tag set f (g, t) in The label of community p changes the label of neighbours community q into, otherwise, does not merges；

5j) iterations t is merged in given maximum community_max=h (g), it is judged that existing community merges whether iterations t reaches Iterations t is merged in maximum community_maxIf reaching, then terminate iteration, and perform 5k), otherwise, t=t+1, return to 5e) enter Row next iteration；

5k) given maximum cycle g_max=100, it is judged that whether current cycle time g reaches maximum cycle g_maxIf reaching Arrive, then terminate circulation, and by final Web Community tag set f (g_max,t_maxCommunity in) is launched into final network node mark Sign set f_z={ f_z1,f_z2,...,f_zi,...,f_znOutput, wherein:

f(g_max,t_max)={ f₁(g_max,t_max),f₂(g_max,t_max),...,f_u(g_max,t_max),...,f_h(gmax)(g_max,t_max), f_u(g_max,t_max) represent reach maximum cycle g_maxIterations t is merged with maximum community_maxTime Web Community tally set The label of u community in conjunction, u ∈ 1,2 ..., h (g_max), h (g_max) represent in network and reach maximum cycle g_maxTime Community's number, f_ziRepresent and reach maximum cycle g_maxTime network node tag set in the label of i-th node, i ∈ 1,2 ..., n}, n represent the interstitial content in network, otherwise, g=g+1, return to 5b) circulate next time.

The most according to claim 1 based on core node with the network community division method of community's convergence strategy, wherein step (1) the adjacency matrix A in, is expressed as follows:

$A=\left[\begin{array}{ccccc}{a}_{11}& .& .& .& a_{1n}\\ .& .& .& .& .\\ .& .& a_{ij}& .& .\\ .& .& .& .& .\\ a_{n1}& .& .& .& a_{nn}\end{array}\right],$

Wherein, a_ijThe i-th row jth column element in expression adjacency matrix A, i ∈ 1,2 ..., and n}, represent any joint in network Point, j ∈ 1,2 ..., and n}, represent the arbitrary node in network, n represents the number of nodes,

The most according to claim 1 based on core node with the network community division method of community's convergence strategy, wherein step (2) Network Search core node set C in, is to be compared by the node degree of node neighbor node all with it each in network, And elect node degree as core node higher than the node of each of which neighbor node.

The most according to claim 1 based on core node with the network community division method of community's convergence strategy, wherein step (3) similarity function value F between interior joint, is to calculate according to RA exponential function, and its formula is as follows:

$F=\underset{c_o\∈N\left(i\right)\∩N\left(j\right)}{\Σ}\frac{1}{d\left(c_o\right)},$

Wherein, i, j represent the node that any two in network is different, c respectively_oExpression i-th node is total to jth node Same neighbours, ∈ represents and belongs to symbol, and N (i) represents the neighborhood of i-th node, and ∩ represents and intersects operation, and N (j) represents jth The neighborhood of individual node, d (c_o) represent node c_oNode degree.

The most according to claim 1 based on core node with the network community division method of community's convergence strategy, wherein step Matrix M (g) of h (g) row 2 row in 5c), is expressed as follows:

$M\left(g\right)=\left[\begin{array}{cc}{M}_{11}\left(g\right)& .\\ .& .\\ M_{u1}\left(g\right)& .\\ .& .\\ M_{h\left(g\right)1}\left(g\right)& M_{h\left(g\right)2}\left(g\right)\end{array}\right],$

Wherein, M_u1U row the 1st column element in (g) representing matrix M (g), u ∈ 1,2 ..., and h (g) }, h (g) represents in network the Community's number during g circulation.