CN106251230A - A kind of community discovery method propagated based on election label - Google Patents

Abstract

The present invention relates to a kind of community discovery method propagated based on election label, comprise the following steps: 1, for given community network, calculate the power of influence of each network node；2, utilize candidate to produce strategy and from the neighbor node of network node, select several nodes respectively as candidate；3, according to the more each candidate of power of influence size and its neighbor node, vie each other the generation person of winning；4, just respectively do not win person and its neighbours person of winning according to poll, vies each other and produces the person that newly do not wins；5, supporting rate is adjusted less than the setting value person of winning, selects, from its neighbours person of winning, the person that newly do not wins；6, carry out the community of network node according to network node label to divide and export, i.e. for all-network node, by the person that has same label belongs to the community structure of the method construct network of same community.The method can improve effectiveness and the stability of community discovery, effectively finds to be hidden in the community structure pattern of social networks.

Description

A kind of community discovery method propagated based on election label

Technical field

The present invention relates to Web Community's discovery technique field, a kind of be applied to community network based on election label The community discovery method propagated.

Background technology

The multiple natures such as such as WWW, social relation network and bio-networks or Complex Social System can be by complexity Network describes, the feature such as complex network degree of having power-law distribution, high convergence factor and modularity community.Substantially, network Community structure refers to such node set: the link between node in set is dense and gathers interior nodes and set exterior node Link sparse.Such as, having between the scholar of complex network research interest and contact relatively closely, they constitute a community, Often approach a subject at present together, and link up few with graph image researcher.Community structure can portray complication system function Topological relation between parts, excavate from complication system network hiding potential valuable community structure pattern be one very Meaningful but comprise the work of challenge.

In recent years, Web Community finds that research is simultaneously standby in the different field such as physics, sociology and computer science Concerned, the algorithm emerged is multifarious.Most existing community discovery algorithms suffer from higher computation complexity, Such as, the time complexity of GN algorithm is O (m²N), the time complexity of modularity optimization is O (mklogn).Too high meter Method cannot meet the needs of large-scale complex Web Community mode discovery to be counted as originally making these traditional communities find.Carry on the back at this Under scape, the label propagation algorithm (Label Propagation Algorithm, LPA) with approximately linear time complexity exists The research of large scale network community discovery gains great popularity.

First Raghavan et al. proposed community discovery method RAK based on label propagation algorithm early than 2007, should First each node initializing is unique community's label by method, then by iterative process by the label of each node All being updated to the label of its most of adjacent nodes, the node group of last intensive connection can progressively become from a unique label One community's node with common recognition thus form community structure.What RAK had followed label propagation algorithm efficiently calculates spy Property, but all using randomized policy due to it at initial labels, the selection aspect such as neighbor node and update sequence, this makes discovery Community structure has the biggest randomness, in some instances it may even be possible to produces all nodes and belongs to the singular solution of same community.For this, society District finds that LPA algorithm is improved by researcher from many aspects.

Random initial labels aspect, Subelj et al. proposes a kind of new label propagation algorithm DPA will be anti-with hierarchical fashion Imperial protection forms strategy combination with attack extension Liang Zhong community and gets up, and extracts community's core by recursive fashion and adjusts community's core The heart finds small community (whisker communities).The experiment that Leung et al. is found by Web Web Community, finds Node label jumps the tactful performance that can effectively promote LPA algorithm with node strength communication strategy of decay.In view of randomly selecting neighbour The more New Policy occupying node label can reduce the robustness of LPA algorithm, and Zhang et al. proposes the different marks of multiple optimums occurring Should select the label that there is local ring with current node to be updated that present node is updated during label.Lin et al. propose a kind of based on The label propagation algorithm CK-LPA of community's core, according to node importance in a network to its assignment and with this value to node label Carry out asynchronous refresh.In order to avoid producing numerous Sui little communities, Zhao Zhuoxiang et al. proposes a kind of society based on label power of influence District finds algorithm LIB, and first this algorithm is chosen a little vertex set and as subset and given each seed unique label Propagate as starting point using subset again, need same label institute accounting in opposite vertexes neighbours when node label is updated The many factors such as the weight on example, degree of vertex and limit consider.In order to avoid producing (institute in consolidated network of strange beast community Node is had all to be subordinate to same community), Barber et al. proposes a kind of modularity label propagation algorithm LPAm, by given one Individual object function so that label propagation algorithm suffers restraints during iterative diffusion, transforms into the problem of community discovery The problem finding object function optimal solution, defines an object function on the basis of neighbours' label number is identical and utilizes label to pass Broadcast the optimal value that algorithm finds the localization of function.Notice that LPAm is because being easily trapped into modularity local maximum and harmful consequences The accuracy of community, Liu et al. proposes LPAm+ algorithm, is combined with multistep greed polymerization (MSG) by LPAm algorithm, utilizes MSG merges multiple similar community to avoid being absorbed in local maximum simultaneously, it is achieved detect Web Community more accurately.Subelj Et al. propose BPA algorithm, first calculate the balance factor of each node, then chosen by the balance factor of cumulative same label Maximum neighborhood, it is to avoid the generation of singular solution.Xie et al. proposes community based on Speaker and Listener concept Find that algorithm SLPA, military will sky etc. propose based on balance ownership coefficient (Balanced belonging coefficient) community Find algorithm.

Although above-mentioned algorithm improves the quality of result community to a certain extent, but simple propagation rule cannot adapt to More complicated community network.For this, this paper presents a kind of community discovery method propagated based on election label (Voting-based Label Propagation algorithm for Non-Overlapping communities Detection, VLPNO).

Summary of the invention

It is an object of the invention to provide a kind of community discovery method propagated based on election label, the method can improve The effectiveness of community discovery and stability, effectively find to be hidden in the community structure pattern of social networks.

For achieving the above object, the technical scheme is that a kind of community discovery method propagated based on election label, Comprise the following steps:

Step S1: for given community network, calculate the power of influence of each network node；

Step S2: for each network node, utilizes candidate to produce strategy and selects several from its neighbor node respectively Node is as candidate；

Step S3: for each candidate, according to the more each candidate of power of influence size and its neighbor node, product of vying each other The raw person of winning；

Step S4: for the person that respectively do not wins, just respectively do not win person and its neighbours person of winning according to poll, product of vying each other The tissue regeneration promoting person of winning；

Step S5: supporting rate is adjusted less than the setting value person of winning, selects from its neighbours person of winning and newly win Person；

Step S6: the community carrying out network node according to network node label divides and exports, i.e. saves for all-network Point, by belonging to the community structure of the method construct network of same community by the person that has same label.

Further, in described step S1, for network node u, its power of influence p (u) is:

Wherein

Wherein, deg (u) is the degree of node u, and N (u) is the neighbor node set of node u.

Further, in described step S2, candidate is utilized to produce strategy respectively from the neighbor node of each network node Select several nodes as candidate, specifically include following steps:

Step S21: for network node u, the node in its neighbor node set N (u) is carried out the descending according to power of influence Sequence, according to power of influence select from N (u) K power of influence be more than p (u) node as candidate, if power of influence is big in N (u) In the node of p (u) less than K, then using in N (u) the powerful node more than p (u) as candidate；

Step S22: all candidates constitute set of candidates V_h, V_hIn each candidate ballot box Bbox in poll Initially it is 1, gives V simultaneously_hIn each candidate give unique community label.

Further, in described step S3, for candidate v, the node in v with N (v) is carried out power of influence and compares, if N V () exists the power of influence node more than p (v), the most therefrom select there is the node of maximum effect power as the person of winning, will simultaneously V reduces to voter and votes the described person of winning, and i.e. carries out adding 1 by the poll of the described person of winning and by community's label of v The operation of community's label of the person of winning described in being updated to, otherwise, candidate v becomes the person of winning, and keeps its community's label constant, The poll of v is added 1 simultaneously.

Further, in described step S4, it is repeated below competition process until poll is stable or reach greatest iteration number: For the person of winning w, w is carried out poll with its neighbours person of winning and compares, if there is poll in the neighbours person of winning of w higher than w's Node, the most therefrom selects the node with the highest poll as the person that newly do not wins, w reduces to voter simultaneously and is marked the community of w The community's label of voter signed and support w is all updated to community's label of the described person that newly do not wins, and otherwise, w does not still win person, and Keep its community's label constant.

Further, in described step S5, for the person of winning y, if its supporting rate Support (y) is less than setting value, From its neighbours person of winning, then select there is the node of the highest supporting rate as the person that newly do not wins, y is reduced to voter and by y simultaneously Label and the community's label of voter supporting y be all updated to community's label of the described person that newly do not wins, wherein l_yWith l_vIt is respectively community's label of node y Yu v, δ (l_y,l_v) it is Kronecker function, if l_y =l_vThen functional value is 1, is otherwise 0.

The invention has the beneficial effects as follows and provide a kind of community discovery method propagated based on election label, compared to tradition Network community discovery method, the inventive method based on real-life voting pattern, is come by the way of ballot and lobbying Choose the core node in network, form the community centered by core node, can significantly improve the effectiveness of community discovery with Stability, effectively finds to be hidden in the community structure pattern of social networks, can be widely applied to micro blog network, mail network, BBS The various social platform such as Forum network, can promote information actively service quality, strengthens the Internet culture safely etc..

Accompanying drawing explanation

Fig. 1 is the flowchart of the embodiment of the present invention.

Fig. 2 is the relation schematic diagram of voter in the embodiment of the present invention, candidate and the person of winning.

Fig. 3 is the schematic diagram that in the embodiment of the present invention, candidate produces strategy.

Fig. 4 is the schematic diagram that in the embodiment of the present invention, voter updates label.

Fig. 5 is the schematic diagram of Karate network in the embodiment of the present invention.

Fig. 6 is the schematic diagram of the Q-value convergence of artificial network Net_1 in the embodiment of the present invention.

Fig. 7 is the schematic diagram of the Q-value convergence of artificial network Net_2 in the embodiment of the present invention.

Fig. 8 is the schematic diagram of the Q-value convergence of artificial network Net_3 in the embodiment of the present invention.

Detailed description of the invention

The present invention provides a kind of community discovery method propagated based on election label, as it is shown in figure 1, comprise the following steps:

Step S1: for given community network, calculate the power of influence of each network node.For network node u, its impact Power p (u) is:

Wherein

Wherein, deg (u) is the degree of node u, and N (u) is the neighbor node set of node u.

Step S2: for each network node, utilizes candidate to produce strategy and selects several from its neighbor node respectively Node is as candidate.Specifically include following steps:

Step S21: for network node u, the node in its neighbor node set N (u) is carried out the descending according to power of influence Sequence, according to power of influence select from N (u) K power of influence be more than p (u) node as candidate, if power of influence is big in N (u) In the node of p (u) less than K, then using in N (u) the powerful node more than p (u) as candidate；

Step S22: all candidates constitute set of candidates V_h, V_hIn each candidate ballot box Bbox in poll Initially it is 1, gives V simultaneously_hIn each candidate give unique community label.

Step S3: for each candidate, according to the more each candidate of power of influence size and its neighbor node, product of vying each other The raw person of winning.Method particularly includes: for candidate v, the node in v with N (v) is carried out power of influence and compares, if N (v) exists The power of influence node more than p (v), the most therefrom selects have the node of maximum effect power as the person of winning, v reduces to ballot simultaneously The described person of winning also is voted by person, i.e. carries out adding 1 by the poll of the described person of winning and be described by community's tag update of v The operation of community's label of the person of winning, otherwise, candidate v becomes the person of winning, and keeps its community's label constant, simultaneously by v's Poll adds 1.

Step S4: for the person that respectively do not wins, just respectively do not win person and its neighbours person of winning according to poll, product of vying each other The tissue regeneration promoting person of winning.Method particularly includes: it is repeated below competition process until poll is stablized or reaches greatest iteration number: for winning Person w, carries out poll by w with its neighbours person of winning and compares, if there is poll in the neighbours person of winning of w higher than the node of w, then from Middle selection has the node of the highest poll as the person that newly do not wins, and w reduces to voter simultaneously and by community's label of w and support w Community's label of voter be all updated to community's label of the described person that newly do not wins, otherwise, w does not still win person, and keeps its society District's label is constant.

Step S5: supporting rate is adjusted less than the setting value person of winning, selects from its neighbours person of winning and newly win Person.Method particularly includes: for the person of winning y, if its supporting rate Support (y) is less than setting value, then select from its neighbours person of winning Select the node with the highest supporting rate as the person that newly do not wins, y is reduced to voter and by the label of y and the ballot supporting y simultaneously Community's label of person is all updated to community's label of the described person that newly do not wins, whereinl_yWith l_vRespectively For community's label of node y Yu v, δ (l_y,l_v) it is Kronecker function, if l_y=l_vThen functional value is 1, is otherwise 0.

Step S6: the community carrying out network node according to network node label divides and exports, i.e. saves for all-network Point, by belonging to the community structure of the method construct network of same community by the person that has same label.

Below in conjunction with the accompanying drawings and specific embodiment the invention will be further described.Elaborate the present invention for convenience, First related definition is illustrated, then analyze the deficiency of the label propagation algorithm of existing half synchronized update.

Related definition

The given Undirected networks G=(V, E) of definition 1 (community structure), set of network nodes is V, and network edge collection is combined into E.For each node u, u ∈ V, making its neighbor node tag set is N (u), and the degree of node u is expressed as deg (u), node u's Label is l_u.Community structure corresponding to node set V is set P={P₁,P₂,…,P_k, wherein AndK value can be specified by user or algorithm automatically determines, ifThen P is called the non-overlapped community of G, otherwise ifMakeSet up, then P is called the overlapping community of G.

Given network G=(V, the E) of definition 2 (node power of influence), the power of influence of node u ∈ V show as u and other nodes it Between Connected degree, usually, the highest then its power of influence of degree of node is the biggest, and the importance of this node is the most notable.The impact of node u Dynamics metering method form can turn to equation below:

$Infl\left(u\right)=\underset{v\∈N\left(u\right)}{\Σ}\frac{\mathrm{deg}\left(u\right)}{\underset{w\∈N\left(v\right)}{\Σ}\mathrm{deg}\left(w\right)}---\left(1\right)$

Wherein Infl (u) represents the power of influence of node u, and deg (u) represents that the degree of node u, N (u) and N (v) represent joint respectively The neighborhood of some u and node v.

Definition 3 (candidate) election contest incipient stage, first elect some nodes with certain power of influence as candidate, often Individual candidate all has a unique initial labels with one for adding up the enumerator BCounter of self obtained poll, order All being elected as the node set of candidate is V_h.One ballot box of each candidate original allocation is used for adding up poll, no matter Candidate is to promote or degradation, and ballot box is all the time for adding up the poll that this node obtains, and participates in the tired of similar label poll Adding, i.e. two ballot box labels are identical, and poll carries out accumulating operation.

Realize the survival of the fittest by competition mechanism between definition 4 (person of winning) candidate, compete failed candidate and reduce to Voter, and compete successful candidate and be upgraded to the person of winning, the person of winning keeps original label, the order person of winning to gather V_s。

Definition 5 (voter) node that power of influence is relatively low during whole election contest does not have the right to be elected, this category node Being referred to as voter, voter only has the right of ballot, selects connected candidate or the person of winning by the way of ballot Label as self label, make voter collect and be combined into V_t。

Network node is in the label decision making process that democracy is voted, and its role is not unalterable, can by competition, Produce and the mechanism such as adjustment realizes voter, role transforming between candidate and the person of winning, this transformational relation such as Fig. 2 institute Show.Voter can be realized by candidate's generation mechanism based on node power of influence and be transformed into candidate, and Regulation mechanism can be by low The person of winning of supporting rate is changed into voter, and similarly, competition mechanism can make candidate have an opportunity to become the person of winning.

Candidate produces strategy

As the vote by ballot scene in reality democratic life, first had to according to certain before carrying out vote by ballot Mode produces candidate target (candidate), and in the starting stage of VLPNO, each network node is initialized to voter, and counts Calculating the node power of influence of each network node, the most each node elects K according to itself affect power from its neighbor node set The individual node bigger than its power of influence, as candidate, if node bigger than its power of influence in neighbours is less than K, is elected the most as far as possible Meeting condition neighbours, K is constant, can be adjusted according to the consistency of network.As it is shown on figure 3, node u is initial For voter, electing 3 power of influence (arrow points to the big person of the power of influence) node bigger than it is candidate.All candidates are constituted Set of candidates V_h, V_hIn the ballot box Bbox of all candidates be initially 1, give V simultaneously_hIn each candidate give only The label of one.

It is worthy of note, VLPNO uses candidate based on node power of influence and elects strategy rather than random plan Slightly, this is based primarily upon such consideration: tradition LPA algorithm research shows, simple random initializtion strategy is not owing to accounting for The relation of nodal properties and neighbours thereof frequently can lead to occur a large amount of scattered isolated little society in follow-up label communication process District and " countercurrently " phenomenon, i.e. make more meaningful big community cannot form the node less with power of influence in label communication process Middle meeting affects the node that some power of influence are bigger in turn.

Campaign strategy

Often wheel election contest is divided into two processes, competes, then carried out according to result by voter between first candidate and the person of winning Ballot.During democracy ballot decision-making, candidate must show the characteristic and advantage of oneself by different approaches, with energy Enough competitions between candidate obtain wins.In VLPNO, compete between candidate and mainly include electing poll to compare and oneself Figure rings force rate relatively two ways.Specifically, candidate when competing first by compare power of influence determine the first run election contest knot Really, it may be assumed that the big person of power of influence is promoted as the person of winning, and in follow-up competition process, candidate is by comparing it at voter The poll obtained determines election, it may be assumed that the many persons of poll are promoted as the person of winning.The person of winning keeps self label constant, failure Person is then downgraded to voter, changes self label, and throws ticket to victor.Candidate competition process relates generally to following 2 rank Section:

Stage 1: each candidate selects a power of influence maximum node as winning from the set of candidates that it is adjacent Person, and be the person's of winning label by self tag update, i.e. oneself campaign for and unsuccessfully reduce to voter, and the ticket of oneself is thrown to this choosing Act person.If there is no the neighbours candidate that power of influence is bigger than it, then self become the person of winning, and keep label constant.For given Candidate u, the stage 1, the generation person of winning of competition was such as the Vic (u) of formula (2).

Wherein Infl (v) represents the power of influence of node v, and N (u) represents the neighbours of node u.

Stage 2: be repeated below competition until poll is stablized or reaches greatest iteration number: the person of winning is from its adjacent winning Person's set selects the most person of winning of aggregate votes as the new person of winning, and by self tag update for the person's of winning label (i.e.: Oneself campaign for and unsuccessfully reduce to voter, and the ticket of oneself is thrown to this person of winning), meanwhile support the voter of this loser Being also required to the tag update of oneself is the person's of winning label, if not having the neighbours constituent people that poll is bigger than it, then self becomes Constituent, and keep label constant.In the similar stage 1, for resulting from the person of the winning u in stage 1, it is in the iterative process in stage 2 In generate the new person of winning by formula (3).

$u.BCounter=u.Bbox+\underset{v\∈N(u\]}{\Σ}v.Bbox\mathrm{\δ}(l_u,l_v)---\left(4\right)$

Wherein u.BCounter represents the obtained poll of node u, often after wheel poll closing, needs the obtained ticket of the statistics person of winning Number participates in next round election contest.U.Bbox represents the poll in node u ballot box Bbox.

Voter mainly includes 2 part of nodes, and the first is not the node of candidate by election by the starting stage, its Two is to compete failed candidate node in candidate competition process.Voter changes the label of oneself with table by ballot Show and its selected person of winning is followed.During ballot selects, each voter for once vote chance, and every time It is that the person of winning that poll in its neighbor node set is most is voted, if its neighbor node exists multiple maximum poll The person of winning, then wait next round ballot.As shown in Figure 4, group's label that voter u has poll most in selecting neighbours comes more The newly label of oneself, in figure, the poll of yellow label is (P₁₁+P₁₂), the poll of red-ticket is (P₂₁+P₂₂), blue label Poll is (P₃₁), if red poll maximum, poll is thrown to node v₃Or node v₄(voter u only has 1 ticket, in group only A poll is needed to increase).If there is multiple poll as many, then wait next round ballot.

Later stage adjustable strategies

After too much wheel election contest and ballot, election contest can tend towards stability, and does not i.e. have voting behavior.Now, in addition it is also necessary to some Supporting rate carries out degraded operation than the relatively low person of winning, and such as Fig. 3-4 (color corresponding label), campaigns for after terminating, although the person of winning (five-pointed star) comes out top, but its supporting rate is the lowest, and neighbours' interior joint great majority broadly fall into different community, for this little society District needs to merge.The supporting rate person of winning less than 50% is forced to carry out degraded operation.The person's of winning supporting rate Support U () computing formula is as follows:

$Support\left(u\right)=\frac{\underset{v\∈N\left(u\right)}{\Σ}\mathrm{\δ}(l_u,l_v)}{\mathrm{deg}\left(u\right)}---(3-5)$

Wherein,Represent with node u with the nodes of label.l_uWith l_vIt is respectively the label of node u Yu v, δ (l_u, l_v) it is Kronecker function, if l_u=l_vThen this functional value is 1, is otherwise 0.

It addition, election contest during, campaigning for the person of winning unsuccessfully by self tag update is the label of victor, and before with It is also required to update the label of oneself with the voter of this loser, will tag update be the new label of loser.

Performance evaluating

In order to analyze the performance of VLPNO algorithm quantitatively, we are by label propagation algorithm conventional to VLPNO algorithm and 3 (LPA, SLPA, BMLPA) generates carry out experiment ratio on program data set at 3 benchmark live network data set and 3 LFR networks Relatively with analysis.Experimental situation is, and: CPU isCore^TM2i5-3230M, internal memory 4G；OS is Win7.3 live networks are respectively It is: Karate network that this network is the mutual social relations between the karate clubbite in one university of the U.S., has 34 Individual node, 78 limits；Dolphins network, has 62 nodes, 159 limits；Football network, this network is Newman etc. The network that the playing conditions of 2000 racing seasons of American college football league is analyzed arranging and sets up by people, it comprises 115 nodes and 616 limits.3 artificial networks are mainly generated by LFR network program by arranging different parameters, and network is joined Number and meaning thereof are as shown in table 1, generate three kinds of different types of artificial networks, small community net respectively according to the parameter in table 1 Network Net_1, medium-sized community network Net_2 and large-scale community network Net_3, its design parameter such as table 2.

Table 1 manually generated network parameter list

Evaluation criterion

This experiment uses modularity Q and normalised mutual information NMI (Normalized Mutual information) two The effectiveness of ISLPA algorithm is evaluated by index with robustness.Modularity Q is that currently used most commonly used community quality refers to Mark, its value deducts in another random network the ratio shared by community's internal edges equal to ratio shared by community's internal edges in network Example, specifically can formalization as follows:

$Q=\frac{1}{2m}\underset{ij}{\Σ}(A_{ij}-\frac{k_i{k}_j}{2m})\mathrm{\δ}(g_i,g_j)---\left(5\right)$

Wherein, the limit number during m represents network；When node i is connected with node j, A_ijEqual to 1, otherwise it is 0；k_iRepresent joint The number of degrees of some i；g_iRepresent community belonging to node i；δ(g_i,g_j) it is Kronecker function, work as g_i=g_jTime, δ (g_i,g_j)=1, no Then, δ (g_i,g_j)=0.Web Community's degree of modularity and Q are positively related, and i.e. when community structure is obvious, Q is close to 1；Work as society When plot structure is inconspicuous, Q is close to 0.

NMI is a kind of community quality evaluation index theoretical based on theory of information, it by calculate known community structure and by Similarity between the community structure that algorithm obtains is to realize estimating of community structure quality, and its value the biggest explanation algorithm is tied The degree that really community is consistent with community content structure is the highest, and concrete formula is as follows:

$NMI(P^{res},P^{true})=\frac{-2\underset{V_m\∈P^{res}}{\Σ}\underset{V_n\∈P^{true}}{\Σ}\frac{|V_m\∩V_n|}{\left|V\right|}log\left(\frac{\left|V\right|\left|V_m{\∩}_n\right|}{\left|V_m\right|\left|V_n\right|}\right)}{\underset{V_m\∈P^{res}}{\Σ}\frac{\left|V_m\right|}{\left|V\right|}\log \left(\frac{\left|V_m\right|}{\left|V\right|}\right)+\underset{V_n\∈P^{true}}{\Σ}\frac{\left|V_n\right|}{\left|V\right|}log\left(\frac{\left|V_n\right|}{\left|V\right|}\right)}---\left(6\right)$

Wherein P^resFor algorithm partition set, P^trueTruly divide set for network, m and n is corresponding division community number, V_m Node set for community m.

Efficiency analysis

For the effectiveness of verification algorithm, have chosen four kinds of algorithms run on Network data set the Q-value of 50 gained with The meansigma methods of NMI value, is analyzed.The division of live network, VLPNO algorithm in this paper is can be seen that by table 3 Relatively contrast algorithms have higher modularity with other.Especially in tae kwon do and dolphin net, divide effect particularly evident.Study carefully it Reason, is primarily due to this two classes network and becomes apparent from relative to football club's network structure, there is the core that power of influence is bigger Heart node, week mid-side node all formation communities centered by core node.And football net, it is equal that corresponding community is typically all power of influence The football team of weighing apparatus, so VLPNO algorithm advantage is not clearly in the unconspicuous network of structure.

Table 3 live network Q-value contrasts

Can be observed by table 4, VLPNO close to 1.0 in the division of tae kwon do network, is equivalent to live network and divides, far Much larger than other three classes algorithms.In conjunction with Fig. 6, it can be seen that this Karate club is mainly to be responsible for principal two as core Network structure.Network surveying finds, forms the reason of Liang Ge group and be club supervisor and whether principal is raising Disagreement is created, so network splits into these two groups as core in the problem of club's charge.The shape of this network Become the election being similar in life, by lobbying and ballot, form the group with campaigner as core.Therefore, election mould is introduced The propagation algorithm of formula, it is possible to effectively find the community network with the big person of power of influence as core in life.

Table 3-4 live network NMI value contrasts

Robust analysis

In order to evaluate the robustness of VLPNO algorithm, we are by regulation hybrid parameter μ (the least community structure of hybrid parameter μ The most obvious) construct the different artificial networks that community module degree is different.Fig. 6,7 and 8 are respectively 4 kinds of LPA algorithms at heterogeneous networks In division result (NMI value).From Fig. 6,7 and 8, in 3 artificial networks, all algorithms are all along with chaotic coefficient μ's Increase and reduce, but the fall of VLPNO is less than other contrast algorithms, illustrates that algorithm has preferable stability.

Being above presently preferred embodiments of the present invention, all changes made according to technical solution of the present invention, produced function is made With during without departing from the scope of technical solution of the present invention, belong to protection scope of the present invention.

Claims (6)

1. the community discovery method propagated based on election label, it is characterised in that comprise the following steps:

Step S1: for given community network, calculate the power of influence of each network node；

Step S2: for each network node, utilizes candidate to produce strategy and selects several nodes respectively from its neighbor node As candidate；

Step S3: for each candidate, according to the more each candidate of power of influence size and its neighbor node, generation victory of vying each other The person of going out；

Step S4: for the person that respectively do not wins, just respectively do not win person and its neighbours person of winning according to poll, and generation of vying each other is new The person of winning；

Step S5: supporting rate is adjusted less than the setting value person of winning, selects, from its neighbours person of winning, the person that newly do not wins；

Step S6: the community carrying out network node according to network node label divides and exports, i.e. for all-network node, logical Cross the community structure of the method construct network that the person that has same label is belonged to same community.

A kind of community discovery method propagated based on election label the most according to claim 1, it is characterised in that described step In rapid S1, for network node u, its power of influence p (u) is:

Wherein

Wherein, deg (u) is the degree of node u, and N (u) is the neighbor node set of node u.

A kind of community discovery method propagated based on election label the most according to claim 2, it is characterised in that described step In rapid S2, utilize candidate to produce strategy and from the neighbor node of each network node, select several nodes respectively as candidate Person, specifically includes following steps:

Step S21: for network node u, the node in its neighbor node set N (u) is carried out the descending sort according to power of influence, Select K power of influence more than the node of p (u) as candidate from N (u) according to power of influence, if power of influence is more than p in N (u) The node of (u) less than K, then using in N (u) the powerful node more than p (u) as candidate；

Step S22: all candidates constitute set of candidates V_h, V_hIn each candidate ballot box Bbox in poll initial It is 1, gives V simultaneously_hIn each candidate give unique community label.

A kind of community discovery method propagated based on election label the most according to claim 3, it is characterised in that described step In rapid S3, for candidate v, the node in v with N (v) is carried out power of influence and compares, if N (v) existing power of influence more than p (v) Node, the most therefrom select there is the node of maximum effect power as the person of winning, v is reduced to voter simultaneously and wins to described Person votes, i.e. carry out adding 1 by the poll of the described person of winning with by community's tag update of v be described in the person of winning community mark The operation signed, otherwise, candidate v becomes the person of winning, and keeps its community's label constant, adds 1 by the poll of v simultaneously.

A kind of community discovery method propagated based on election label the most according to claim 4, it is characterised in that described step In rapid S4, it is repeated below competition process until greatest iteration number is stablized or reached to poll: for the person of winning w, by w and its neighbours The person of winning carries out poll and compares, if there is the poll node higher than w in the neighbours person of winning of w, the most therefrom selection has the highest W, as the person that newly do not wins, is reduced to voter the community by community's label of w with the voter supporting w by the node of poll simultaneously Label is all updated to community's label of the described person that newly do not wins, and otherwise, w does not still win person, and keeps its community's label constant.

A kind of community discovery method propagated based on election label the most according to claim 5, it is characterised in that described step In rapid S5, for the person of winning y, if its supporting rate Support (y) is less than setting value, then select to have from its neighbours person of winning Y, as the person that newly do not wins, is reduced to voter the society by the label of y with the voter supporting y by the node of the highest supporting rate simultaneously District's label is all updated to community's label of the described person that newly do not wins, whereinl_yWith l_vIt is respectively node y With community's label of v, δ (l_y,l_v) it is Kronecker function, if l_y=l_vThen functional value is 1, is otherwise 0.