CN107169871A - It is a kind of to optimize many relation community discovery methods expanded with seed based on composition of relations - Google Patents
It is a kind of to optimize many relation community discovery methods expanded with seed based on composition of relations Download PDFInfo
- Publication number
- CN107169871A CN107169871A CN201710260510.6A CN201710260510A CN107169871A CN 107169871 A CN107169871 A CN 107169871A CN 201710260510 A CN201710260510 A CN 201710260510A CN 107169871 A CN107169871 A CN 107169871A
- Authority
- CN
- China
- Prior art keywords
- mrow
- community
- msub
- seed
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Abstract
The invention belongs to community network and Computer Applied Technology field, disclose a kind of many relation community discovery methods for optimizing based on composition of relations and being expanded with seed, by optimize various relations in network weight match by many relational networks be fused into one can effectively integrate each relation community information while the low single relational network of noise, then the community's division information for integrating each relation in many relational networks searches out the crowd all in same community in each relation to come, seed community is used as using the small community that these crowds constitute, community mining is carried out using a kind of seed Expansion strategies many-many relationship community network, the higher community structure of accuracy rate is obtained to divide.Experiment shows that the present invention has result accuracy rate high compared with conventional method, the strong advantage of Noise Resistance Ability.
Description
Technical field
The invention belongs to community network and Computer Applied Technology field, more particularly to it is a kind of based on composition of relations optimization and
Many relation community discovery methods of seed expansion.
Background technology
Community network (Social Network) is the network structure being made up of many nodes and Lian Bian, and node can be corresponded to
Company side between real-life people and various community organizations, node can correspond to the various contacts in daily life
Relation, such as the social relationships or wechat, phone, mail etc. such as interpersonal friend, household, classmate contact is closed
System.Community discovery is an important research direction of the complex network including community network, in ecommerce, public peace
Entirely, the field such as biology suffers from huge application value.Community refers to that some in network possess more common spy each other
Levy, contact more node clusterings.Network is opened up between benefit structure is embodied in the node of same community and contacts relatively closer,
And be not belonging to contact than sparse between the node of same community.Community discovery is a basic task of social network analysis,
It is that community structure present in network is excavated, studies the community of network to understanding that the 26S Proteasome Structure and Function of whole network has
Vital effect.General is mostly that a kind of relation in network is studied to researching and analysing for community network, and is showed
Grow directly from seeds it is living in community network individual between all there is polytype relation, such as exist between men friend, classmate,
The social relationships such as household.And the arrival in Web2.0 epoch more makes the exchange contact method existed between people become varied,
Except traditional phone and short message, also wechat, microblogging, everybody Facebook, Twitter, Youtube etc. is a variety of based on interconnection
The exchange contact also generally existing in community network, or even there is also a variety of exchange sides in same media of communication of net
Exchange can be produced by multiple channels such as comment, forwarding, thumb ups between people in formula, such as microblogging.The society of many relations
Network is prevalent in actual life.Grinding based on a kind of relation is substantially to the community discovery method of community network at present
Study carefully, and because the exchange way between Social Individual has certain subjectivity, individual can not be fully grasped only with a kind of relation
Between exchange of information, it is likely that can not corresponded to reality because information not cause community division result final comprehensively.While certain
Exchange way may produce more social noise because of being easier to occur to contact with stranger or usually unfamiliar people, this
A little noises can also cover real community structure in actual life.Existing a part of many relation community discovery methods are to many passes
When being that network carries out community discovery, the network that various relations are constituted all is made no exception, but in actual many tie society networks, by
In the difference of exchange way, the social noise of each relation is different.Although such method can make each to be related to it
Between community information complement each other, but can also introduce the noise of each social networks simultaneously, it is larger in some social networks noise ratios
In the case of be possible to cause to consider a variety of relations and carry out the result of community discoveries on the contrary not as only considering single relation
Carry out the result of community discovery.
In summary, the problem of prior art is present be：It has ignored when at present to the community discovery method of community network each
The otherness of relation, the information that each relation is brought is made no exception, therefore there is the community information of comprehensive each relational network
When have also been introduced social noise entrained by each relational network, cause community discovery accuracy rate not high enough, some relation noises compared with
The result for considering a variety of relations progress community discoveries can be even caused when big on the contrary not as only considering that single relation carries out society
The result that area is found.
The content of the invention
The problem of existing for prior art, the invention provides it is a kind of based on composition of relations optimize and seed expand it is many
Relation community discovery method.
The present invention is achieved in that a kind of many relation community discovery sides for optimizing based on composition of relations and being expanded with seed
Method, it is described to be optimized based on many relation community discovery methods that composition of relations optimizes and seed is expanded by multiple-objection optimization in network
Various relations weight proportioning by many relational networks be fused into one can effectively integrate each relation community information while noise it is low
Single relational network, then integrate in many relational networks that community's division information of each relation will be in each relation all in same
The crowd of individual community, which searches out, to come, and the small community using these crowds composition is as seed community, using seed Expansion strategies to many
Tie society network carries out community mining, obtains the higher community structure of accuracy rate and divides；
The weight ratio optimization sub-goal one is used as weighing result community structure intensity and reliability using modularity
Index, result of calculation CaModularity Qa, modularity is calculated as follows：
Wherein AI, jIt is the adjacency matrix of whole network, m is the number on the side included in network, kiAnd kjNode is represented respectively
I and node the j number of degrees, δ (Ci, Cj) value depend on i and j whether in the same community, δ (C if beingi, Cj) value
For it is 0 otherwise to talk about its value；
The weight ratio optimization sub-goal two uses NMI as community structure information similarity metric, by as follows
Calculate：
Wherein confusion matrix H row represents real community division result, and H row represent the community that partitioning algorithm is drawn
As a result.CAAnd CBA, community's number that two results of B are included, H are represented respectivelyijRepresentative should but appear in community j in community i
In node number, Hi.And H.jRepresent the nodes sum of the community of ith row and jth column respectively, N is contained in whole network
Interstitial content；As NMI=1, the community structure for representing two results of A and B is identical, NMI=0, and interval scale is completely not
Equally；
The seed Expansion strategies choose the most community S of nodesiIt is used as seed community；Calculate seed community SiWith it
Remaining candidate seed community and the similarity of remaining node, calculate formula as follows：
U and v are community S respectivelyi, SjIn node, niAnd njIt is the quantity of Liang Ge communities interior joint respectively,WithIt is
Node serial number set.
Further, many relation community discovery methods expanded with seed that optimized based on composition of relations include following step
Suddenly：
1) each relation of many relational networks is directly configured into weight according to same ratio and many relational networks is merged into one
Single relational network A of individual cum rights；
2) community's division, the community division result of gained are carried out to network A using a kind of community division method of single relation
It is designated as Ca, then using a kind of index for weighing community discovery result reliability and structural strength to result CaWeighed, weighed
The value gone out is designated as Qa；
3) with the weight of each relation with decision variable is used for, using a kind of Multipurpose Optimal Method simultaneously to following two
Sub-goal optimizes to obtain one group of optimizing decision variables set；
Optimize sub-goal one：Single relational network B and the knot of the community division result of network A that new weight proportioning is merged into
Structure measurement index difference is as big as possible；
Optimize sub-goal two：The single relational network B and the community division result of network A that new weight proportioning is merged into show
The community structure information similitude shown is as big as possible.
4) in step 3) in the optimizing decision variables set that obtains selection make the maximum weight proportioning of sub-goal one by many relations
The network integration carries out community discovery into single relational network M, and using a kind of single relation community discovery method to M, by Suo Huo communities
Number is designated as K；
5) community's division is carried out to the network that each relation is constituted using a kind of single relation community discovery method respectively；
6) iteration mark L=0 is set, to step 5) result of each relation community discovery counts, and will be in all relations
In as seed community be put into seed community set C all in the node of a communityLIn；
7) by single relational network M by a kind of similarity calculating method come similar between each node in calculating network
Degree；
If 8) L=0, L=L+1, set C are madeL=CL-1, carry out next step；If L>0, detection set CLWith CL-1Whether one
Sample, is to go to step 11), otherwise makes L=L+1, set CL=CL-1Carry out next step；
9) by seed community candidate collection CLIn the number of nodes that is included according to community of community for not yet participating in merging
It is ranked up from big to small, chooses sequence highest community SiIt is used as seed community；
10) by step 7) the Similarity Measure seed community S that calculatesiWith remaining candidate seed community and residue
The similarity of node is designated as Sim；
11) choose and SiSimilarity exceedes given threshold β community and node is put into candidate regions, passes through a kind of local adaptation
Spend computational methods and calculate SiThe current fitness F and current fitness F of each community of candidate regionsh, candidate is then calculated respectively to be merged
The community in area and node and SiFitness value after merging is designated as Fnew, then calculate FnewRelative to SiThe increasing of fitness F originally
Long rate VsAnd candidate community fitness FnewGrowth rate Vh；
12) in VsAnd VhOn the premise of both greater than growth rate threshold value δ, choosing makes Vs+VhMaximum candidate community SjWith SiMerge
Into a new communities, S is updatediWith seed community set CL, return to step 10)；If without qualified, checking set CL
In existing community whether all merged, it is no if return to step 9), will set C if beingLIn all communities merging
Record is reset, return to step 8)；
13) community's number in network is checked, is waited and K if being less than, output result, EP (end of program), output set CL；If big
In K, then following dispelling tactics are performed；Using calculated comprising the most preceding K communities of interstitial content as target, successively residue community and
Node and their similarity, remaining community and node merger are entered with K community before their similarity highests, set is updated
CL, finally export community's set CLIt is used as final division result.
Further, community stroke is carried out to the network of four relation compositions using single relation community discovery algorithm BGLL respectively
Point, statistics division result is according to formulaObtain cohesion matrix I, wherein IijFor node i, j is in a matrix
Corresponding value, N is 4, m represents which dimension, C here for the sum of dimensionmWhether (i, j) represents at 2 points and is drawn in m dimensions
Divide in same community, if 2 points of C if m-th of dimension is divided into same communitym(i, j) value 1, otherwise being 0；Ergodic Matrices
I, is all set to zero by the number that all values are less than 4, obtains matrix of areas F；Depth-first traversal is carried out to matrix F, it is obtained
All communicated subareas, the region by subregion interstitial content more than 1 is stored in candidate seed community's set CL, L=0.
Further, by single relational network M come the similarity Sim between each node in calculating network:
Wherein K represents that i, j are divided into the dimension number of identical community, N=4, node i when Δ gain factor is 1, W (i, j)
With node j connection side right weight, two nodes neighbors nodes of com (i, j) occur simultaneously, N (i, j) be the neighbor node of two nodes simultaneously
Collection.
Further, choose and SiSimilarity exceedes the community of given threshold γ=1 and node is put into candidate regions, calculates SiWhen
The current fitness F of preceding each community of fitness F and candidate regionsh, calculation formula is as follows：
Wherein formulaIt is twice of the weight sum of community's internal edges,Be with the weight of community's external edge it
With the parameter of, α for regulation and control community's scale, value is set to one；
Then community and node and the S of candidate assembly section are calculated respectivelyiFitness value after merging is designated as, and is then calculated
FnewRelative to SiThe growth rate V of fitness F originallysAnd candidate community fitness FhGrowth rate Vh；
In VsAnd VhBoth greater than on the premise of growth rate threshold value δ=0.1, V is chosens+VhThat maximum community and SiMerge
Into a new communities, S is updatediWith seed community set CL；If without qualified, checking set CLIn existing community be
It is no all to have merged；Will set C if beingLIn all communities merging record reset；
Community's number is checked, is waited and K if being less than, output result, EP (end of program), output set CL；If being more than K, with
Be target comprising the most preceding K communities of interstitial content, remaining community and their similarity calculated successively, by remaining community with
Node merger enters with K community before their similarity highests, updates set CL, finally export community's set CLDrawn as final
Divide result.
Another object of the present invention is to provide many passes expanded described in a kind of application based on composition of relations optimization and seed
It is the community network of community discovery method.
Another object of the present invention is to provide many passes expanded described in a kind of application based on composition of relations optimization and seed
It is the computer of community discovery method.
Advantages of the present invention and good effect are：
1. the various relations in many-many relationship network of the present invention are combined optimization by weight proportioning, can be in synthesis
The social noise for preferably overcoming different relational networks to bring while each relational network effective information, it is larger in each relation noise
When, still result in more accurately community division result.The community that the present invention passes through each relation in comprehensive many relational networks
Division information searches out the crowd all in same community in each relation to come, these unusual fixed social circles one
As all in place community core position, using them as seed community carry out expansion be more beneficial for finding out many relational networks
In imply community content.Artificial network's experiment shows that the accuracy that community discovery result of the invention has is more flat than conventional method
More 9% is improved, specific effect is shown in Fig. 4.The noise ratio gap carried in each relation is up under 50% harsh conditions, is passed
Many relational approaches of uniting are reduced to 71% not as individually being closed using noise minimum by noise effect community discovery result accuracy
The 82% of system, and the accuracy of the present invention is 86.3% in this case, specific effect is shown in Fig. 5.
Brief description of the drawings
Fig. 1 is many relation community discovery methods provided in an embodiment of the present invention for being optimized based on composition of relations and being expanded with seed
Flow chart.
Fig. 2 is that many relation community discoveries provided in an embodiment of the present invention that expanded with seed that optimized based on composition of relations are realized
Flow chart.
Fig. 3 is that result provided in an embodiment of the present invention is shown with carrying out the Comparative result of community discovery only with a kind of relation
It is intended to.
Fig. 4 is the result provided in an embodiment of the present invention from traditional many relation community discovery method PMM under different noises
Contrast schematic diagram.
Fig. 5 is the specific effect diagram of the present invention provided in an embodiment of the present invention.
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to embodiments, to the present invention
It is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to
Limit the present invention.
The application principle of the present invention is explained in detail below in conjunction with the accompanying drawings.
The many relation community discovery methods provided in an embodiment of the present invention that expanded with seed that optimized based on composition of relations are passed through
Many relational networks are fused into one by the weight proportioning of various relations in optimization network can effectively integrate each relation community information
While the low single relational network of noise, the community's division information for then integrating each relation in many relational networks will be in each relation
Search out all in the crowd of same community, the small community using these crowds composition is as seed community, using a kind
Sub- Expansion strategies many-many relationship community network carries out community mining, is divided so as to obtain the higher community structure of accuracy rate.
As shown in figure 1, provided in an embodiment of the present invention optimize many relation communities hair expanded with seed based on composition of relations
Existing method comprises the following steps：
S101：By each relation of many relational networks directly according to 1:Many relational networks are merged into by 1 ratio to configure weight
Single relational network of one Weight；
S102：Community's division is carried out to the network using a kind of single relation community partitioning algorithm, using certain measurement index
Computation partition result community structure intensity；
S103：Optimization aim is optimized to the weight combination matching of each dimensional relationships using a kind of multi-objective Evolutionary Algorithm
The community division result for the network that the one new weight for being is merged into is with most starting 1:The society of the 1 Web Community's division result being merged into
The difference of plot structure intensity index is as big as possible；Optimization aim two weighs the two community division result similarity using a kind of index,
Make similarity as high as possible；
S104：Matched with the weight after optimization and many relational networks are changed into single relational network；
S105：Community's division is carried out to the network that each relation is constituted using a kind of single relation community discovery algorithm respectively,
The result of statistical relationship community discovery, regard the node all in a community in all relations as candidate seed community；
S106：By selected community's similarity function and local fitness function between seed community and remaining node
Merge；
S107：Detection community's number decides whether to use dispelling tactics after merging terminates, and exports final community division result.
Many relation community discovery methods provided in an embodiment of the present invention based on composition of relations optimization and seed expansion are specific
Comprise the following steps：
1) each relation of many relational networks is directly configured into weight according to same ratio and many relational networks is merged into one
Single relational network A of individual cum rights.
2) community's division, the community division result of gained are carried out to network A using a kind of community division method of single relation
It is designated as Ca, then using a kind of index for weighing community discovery result reliability and structural strength to result CaWeighed, weighed
The value gone out is designated as Qa。
3) with the weight of each relation with decision variable is used for, using a kind of Multipurpose Optimal Method simultaneously to following two
Sub-goal optimizes to obtain one group of optimizing decision variables set；
Optimize sub-goal one：Single relational network B and the knot of the community division result of network A that new weight proportioning is merged into
Structure measurement index difference is as big as possible；
Optimize sub-goal two：The single relational network B and the community division result of network A that new weight proportioning is merged into show
The community structure information similitude shown is as big as possible, and the two community structure information similitude can use some conventional indexs, than
Such as normalised mutual information NMI.
4) in step 3) in the optimizing decision variables set that obtains selection make the maximum weight proportioning of sub-goal one by many relations
The network integration carries out community discovery into single relational network M, and using a kind of single relation community discovery method to M, by Suo Huo communities
Number is designated as K.
5) community's division is carried out to the network that each relation is constituted using a kind of single relation community discovery method respectively.
6) iteration mark L=0 is set, to step 5) result of each relation community discovery counts, and will be in all relations
In as seed community be put into seed community set C all in the node of a communityLIn.
7) by single relational network M by a kind of similarity calculating method come similar between each node in calculating network
Degree.
If 8) L=0, L=L+1, set C are madeL=CL-1, carry out next step；If L>0, detection set CLWith CL-1Whether one
Sample, is to go to step 11), otherwise makes L=L+1, set CL=CL-1Carry out next step.
9) by seed community candidate collection CLIn the number of nodes that is included according to community of community for not yet participating in merging
It is ranked up from big to small, chooses sequence highest community SiIt is used as seed community.
10) by step 7) the Similarity Measure seed community S that calculatesiWith remaining candidate seed community and residue
The similarity of node is designated as Sim.
11) choose and SiSimilarity exceedes given threshold β community and node is put into candidate regions, passes through a kind of local adaptation
Spend computational methods and calculate SiThe current fitness F and current fitness F of each community of candidate regionsh, candidate is then calculated respectively to be merged
The community in area and node and SiFitness value after merging is designated as Fnew, then calculate FnewRelative to SiThe increasing of fitness F originally
Long rate VsAnd candidate community fitness FnewGrowth rate Vh。
12) in VsAnd VhOn the premise of both greater than growth rate threshold value δ, choosing makes Vs+VhMaximum candidate community SjWith SiMerge
Into a new communities, S is updatediWith seed community set CL, return to step 10)；If without qualified, checking set CL
In existing community whether all merged, it is no if return to step 9), will set C if beingLIn all communities merging
Record is reset, return to step 8).
13) community's number in network is checked, is waited and K if being less than, output result, EP (end of program), output set CL；If big
In K, then following dispelling tactics are performed；
So that remaining community and node and their phase are calculated as target, successively comprising the most preceding K communities of interstitial content
Like spending, remaining community and node merger are entered with K community before their similarity highests, set C is updatedL, finally export society
Area's set CLIt is used as final division result.
The application principle of the present invention is further described below in conjunction with the accompanying drawings.
The many relation community discovery methods provided in an embodiment of the present invention that expanded with seed that optimized based on composition of relations are included
Following steps：
Step one：Many relational networks are inputted, and each relational network is processed into adjacency matrix, input similarity threshold σ=1
And growth rate threshold value beta=0.1；
The many relational networks used is manually generated networks, and the size of artificial network is 350 nodes, includes three societies
Area, each community includes 50,100 and 200 nodes respectively.Exist between node and exist between the relation of four dimensions, i.e. node
Four kinds of connected modes.Node between same community is attached with probability u, and the connection being not belonging between the node of same community is general
Rate is according to dimension variation, i.e., the connection probability v of different community's intermediate nodes under different relations is different.Added simultaneously into network
Noise, the method that noise is produced is connected at random with probability r between all nodes in network.
Step 2：By four relations of many relational networks directly according to 1:Many networks of personal connections are complexed by 1 ratio to configure weight
And into single relational network A of a Weight, the embodiment of the present invention is entered from the community partitioning algorithm BGLL of single relation to network A
Row community is divided, and the community division result of gained is designated as Ca, the embodiment of the present invention using modularity be used as weighing result community knot
The index of structure intensity and reliability, result of calculation CaModularity Qa, modularity is calculated as follows：
Wherein AI, jIt is the adjacency matrix of whole network, m is the number on the side included in network, kiAnd kjNode is represented respectively
I and node the j number of degrees, δ (Ci, Cj) value depend on i and j whether in the same community, δ (C if beingi, Cj) value
For it is 0 otherwise to talk about its value.
Step 3：With the weight of each relation with decision variable is used for, using a kind of Multipurpose Optimal Method while under
Two sub-goals of row optimize to obtain one group of optimizing decision variables set；
Optimize sub-goal one：Single relational network B and the mould of the community division result of network A that new weight proportioning is merged into
Lumpiness difference is as big as possible, optimizes sub-goal two：The single relational network B and the community of network A that new weight proportioning is merged into are drawn
Divide the community structure information similitude shown by result as big as possible, the implementation case uses normalised mutual information NMI, by as follows
Calculate：
Wherein confusion matrix H row represents real community division result, and H row represent the community that partitioning algorithm is drawn
As a result.CAAnd CBA, community's number that two results of B are included, H are represented respectivelyijRepresentative should but appear in community j in community i
In node number, Hi.And H.jRepresent the nodes sum of the community of ith row and jth column respectively, N is contained in whole network
Interstitial content.As NMI=1, the community structure for representing two results of A and B is identical, NMI=0, and interval scale is completely not
Equally.Multi-objective genetic algorithm initial population size is 50, crossover probability 0.8, mutation probability 0.2, iterations 300, coding
Mode is real coding.
Step 4：Selection makes the maximum solution of module angle value as last solution in the optimal solution set that step 3 is obtained, according to
Many relational networks are merged into single relational network M by the weight proportioning that last solution is provided, and are calculated using a kind of single relation community discovery
Method carries out community discovery to M, Suo Huo communities number is designated as into K, here K=3.
Step 5：Community stroke is carried out to the network of four relation compositions using single relation community discovery algorithm BGLL respectively
Point, statistics division result is according to formulaObtain cohesion matrix I, wherein IijFor node i, j is in a matrix
Corresponding value, N is 4, m represents which dimension, C here for the sum of dimensionmWhether (i, j) represents at 2 points and is drawn in m dimensions
Divide in same community, if 2 points of C if m-th of dimension is divided into same communitym(i, j) value 1, otherwise being 0；Ergodic Matrices
I, is all set to zero by the number that all values are less than 4, obtains matrix of areas F；Depth-first traversal is carried out to matrix F, it is obtained
All communicated subareas, the region by subregion interstitial content more than 1 is stored in candidate seed community's set CL, L=0.
Step 6：By single relational network M come the similarity Sim between each node in calculating network:
Wherein K represents that i, j are divided into the dimension number of identical community, N=4, node i when Δ gain factor is 1, W (i, j)
With node j connection side right weight, two nodes neighbors nodes of com (i, j) occur simultaneously, N (i, j) be the neighbor node of two nodes simultaneously
Collection.
Step 7：If L=0, L=L+1, set C are madeL=CL-1, carry out next step；Set C is detected if noLWith CL-1It is
It is no the same, 11 are gone to step if being, otherwise, L=L+1, set C is madeL=CL-1Carry out next step.
Step 8：By seed community candidate collection CLIn the node that is included according to community of community for not yet participating in merging
How much number sorts, and chooses the most community S of nodesiIt is used as seed community.
Step 9：Calculate seed community SiWith remaining candidate seed community and the similarity of remaining node, formula is calculated
It is as follows：
U and v are community S respectivelyi, SjIn node, niAnd njIt is the quantity of Liang Ge communities interior joint respectively,WithIt is
Node serial number set.
Step 10：Choose and SiSimilarity exceedes the community of given threshold γ=1 and node is put into candidate regions, calculates SiWhen
The current fitness F of preceding each community of fitness F and candidate regionsh, calculation formula is as follows：
Wherein formulaIt is twice of the weight sum of community's internal edges,Be with the weight of community's external edge it
With the parameter of, α for regulation and control community's scale, here value be set to one；
Then community and node and the S of candidate assembly section are calculated respectivelyiFitness value after merging is designated as, and is then calculated
FnewRelative to SiThe growth rate V of fitness F originallysAnd candidate community fitness FhGrowth rate Vh。
Step 11：In VsAnd VhBoth greater than on the premise of growth rate threshold value δ=0.1, V is chosens+VhThat maximum society
Area and SiA new communities are merged into, S is updatediWith seed community set CL, return to step ten；If without qualified, examining
Look into set CLIn existing community whether all merged, it is no if return to step nine, will set C if beingLIn all communities
Merging record reset, return to step eight.
Step 12：Community's number is checked, is waited and K if being less than, output result, EP (end of program), output set CL；If big
, then, will be surplus to calculate remaining community and their similarity as target, successively comprising the most preceding K communities of interstitial content in K
Remaining community and node merger enter with K community before their similarity highests, update set CL, finally export community's set CLMake
For final division result.
The application effect of the present invention is explained in detail with reference to emulation.
Emulation one is u=0.5 in case study on implementation of the present invention, and the present invention only with a kind of with closing in the case of noise r=0.1
System carries out the Comparative result of community discovery with BGLL algorithms, using NMI values come the phase of confirmatory experiment result and community content structure
Like spending, Fig. 3 is seen.It can be seen that the result for carrying out community discovery using a variety of relations is better than only with a kind of result of relation.
When emulation two is continues to increase noise in case study on implementation of the present invention (increase to 0.4 from 0.1, increase every time 0.05),
The contrast of the testing result of the present invention and traditional PMM methods, is shown in Fig. 4.It can be seen that the noise resistance effect of the method for the present invention
It is stronger.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
Any modifications, equivalent substitutions and improvements made within refreshing and principle etc., should be included in the scope of the protection.
Claims (8)
1. a kind of optimize many relation community discovery methods expanded with seed based on composition of relations, it is characterised in that described to be based on
Many relation community discovery methods that composition of relations optimizes and seed is expanded optimize various relations in network by multiple-objection optimization
Weight proportioning by many relational networks be fused into one can effectively integrate each relation community information while the low single network of personal connections of noise
Network, then integrates community's division information of each relation in many relational networks by the people in each relation all in same community
Group, which searches out, to come, and the small community using these crowds composition is as seed community, using seed Expansion strategies many-many relationship social network
Network carries out community mining, obtains the higher community structure of accuracy rate and divides；
The weight ratio optimization sub-goal one uses modularity as weighing result community structure intensity and the index of reliability,
Result of calculation CaModularity Qa, modularity is calculated as follows：
<mrow>
<mi>Q</mi>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mrow>
<mn>2</mn>
<mi>m</mi>
</mrow>
</mfrac>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</munder>
<mo>&lsqb;</mo>
<msub>
<mi>A</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>-</mo>
<mfrac>
<mrow>
<msub>
<mi>k</mi>
<mi>i</mi>
</msub>
<mo>*</mo>
<msub>
<mi>k</mi>
<mi>j</mi>
</msub>
</mrow>
<mrow>
<mn>2</mn>
<mi>m</mi>
</mrow>
</mfrac>
<mo>&rsqb;</mo>
<mi>&delta;</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>C</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>C</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
Wherein AI, jIt is the adjacency matrix of whole network, m is the number on the side included in network, kiAnd kjRespectively represent node i and
The node j number of degrees, δ (Ci, Cj) value depend on i and j whether in the same community, δ (C if beingi, Cj) value is,
Otherwise it is 0 to talk about its value；
The weight ratio optimization sub-goal two, as community structure information similarity metric, is calculated as follows using NMI：
<mrow>
<mi>N</mi>
<mi>M</mi>
<mi>I</mi>
<mrow>
<mo>(</mo>
<mi>A</mi>
<mo>,</mo>
<mi>B</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<mo>-</mo>
<mn>2</mn>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<msub>
<mi>C</mi>
<mi>A</mi>
</msub>
</msubsup>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<msub>
<mi>C</mi>
<mi>B</mi>
</msub>
</msubsup>
<msub>
<mi>H</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<msub>
<mi>H</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mi>N</mi>
</mrow>
<mrow>
<msub>
<mi>H</mi>
<mrow>
<mi>i</mi>
<mo>.</mo>
</mrow>
</msub>
<msub>
<mi>M</mi>
<mrow>
<mo>.</mo>
<mi>j</mi>
</mrow>
</msub>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<msub>
<mi>C</mi>
<mi>A</mi>
</msub>
</msubsup>
<msub>
<mi>H</mi>
<mrow>
<mi>i</mi>
<mo>.</mo>
</mrow>
</msub>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>H</mi>
<mrow>
<mi>i</mi>
<mo>.</mo>
</mrow>
</msub>
<mo>/</mo>
<mi>N</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<msub>
<mi>C</mi>
<mi>B</mi>
</msub>
</msubsup>
<msub>
<mi>H</mi>
<mrow>
<mo>.</mo>
<mi>j</mi>
</mrow>
</msub>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>H</mi>
<mrow>
<mo>.</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>/</mo>
<mi>N</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
Wherein confusion matrix H row represents real community division result, and H row represent community's knot that partitioning algorithm is drawn
Really.A is represented respectively, and community's number that two results of B are included, representative should but appear in the node in community j in community i
Number, and represent the nodes sum of the community of ith row and jth column respectively, N is contained interstitial content in whole network；When
During NMI=1, the community structure for representing two results of A and B is identical, and NMI=0, interval scale is completely different；
The seed Expansion strategies choose the most community S of nodesiIt is used as seed community；Calculate seed community SiWith remaining time
Choose seeds the similarity of sub- community and remaining node, calculate formula as follows：
<mrow>
<mi>C</mi>
<mi>O</mi>
<mi>S</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>S</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>S</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mi>u</mi>
<msub>
<mi>n</mi>
<msub>
<mi>s</mi>
<mi>i</mi>
</msub>
</msub>
</msubsup>
<msubsup>
<mi>&Sigma;</mi>
<mi>v</mi>
<msub>
<mi>n</mi>
<msub>
<mi>s</mi>
<mi>j</mi>
</msub>
</msub>
</msubsup>
<mi>S</mi>
<mi>i</mi>
<mi>m</mi>
<mrow>
<mo>(</mo>
<mi>u</mi>
<mo>,</mo>
<mi>v</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mi>n</mi>
<mi>i</mi>
</msub>
<mo>*</mo>
<msub>
<mi>n</mi>
<mi>j</mi>
</msub>
</mrow>
</mfrac>
<mo>;</mo>
</mrow>
U and v are community S respectivelyi, SjIn node, niAnd njIt is the quantity of Liang Ge communities interior joint respectively,WithIt is node
Numbering set.
2. as claimed in claim 1 optimize many relation community discovery methods expanded with seed, its feature based on composition of relations
It is, many relation community discovery methods expanded with seed that optimized based on composition of relations are comprised the following steps：
1) each relation of many relational networks is directly configured into weight according to same ratio and many relational networks is merged into a band
Single relational network A of power；
2) community's division is carried out to network A using a kind of community division method of single relation, the community division result of gained is designated as
Ca, then using a kind of index for weighing community discovery result reliability and structural strength to result CaWeighed, weighed out
Value is designated as Qa；
3) with the weight of each relation with decision variable is used for, using a kind of Multipurpose Optimal Method simultaneously to following two specific item
Mark optimizes to obtain one group of optimizing decision variables set；
4) in step 3) in the optimizing decision variables set that obtains selection make the maximum weight proportioning of sub-goal one by many relational networks
Single relational network M is fused into, and community discovery is carried out to M using a kind of single relation community discovery method, by Suo Huo communities number
It is designated as K；
5) community's division is carried out to the network that each relation is constituted using a kind of single relation community discovery method respectively；
6) iteration mark L=0 is set, to step 5) result of each relation community discovery counts, and will be in all relations all
Node in a community is put into seed community set C as seed communityLIn；
7) by single relational network M by a kind of similarity calculating method come the similarity between each node in calculating network；
If 8) L=0, L=L+1, set C are madeL=CL-1, carry out next step；If L>0, detection set CLWith CL-1Whether, it is
Then go to step 11), otherwise make L=L+1, set CL=CL-1Carry out next step；
9) by seed community candidate collection CLIn the number of nodes that includes according to community of the community for not yet participating in merging from greatly to
It is small to be ranked up, choose sequence highest community SiIt is used as seed community；
10) by step 7) the Similarity Measure seed community S that calculatesiWith remaining candidate seed community and remaining node
Similarity be designated as Sim；
11) choose and SiSimilarity exceedes given threshold β community and node is put into candidate regions, passes through a kind of local adaptation's degree meter
Calculation method calculates SiThe current fitness F and current fitness F of each community of candidate regionsh, candidate assembly section is then calculated respectively
Community and node and SiFitness value after merging is designated as Fnew, then calculate FnewRelative to SiThe growth rate of fitness F originally
VsAnd candidate community fitness FnewGrowth rate Vh；
12) in VsAnd VhOn the premise of both greater than growth rate threshold value δ, choosing makes Vs+VhMaximum candidate community SjWith SiIt is merged into one
Individual new communities, update SiWith seed community set CL, return to step 10)；If without qualified, checking set CLIn it is existing
Have whether community had all merged, it is no if return to step 9), will set C if beingLIn all communities merging record
Reset, return to step 8)；
13) community's number in network is checked, is waited and K if being less than, output result, EP (end of program), output set CL；If more than K,
Then perform following dispelling tactics；So that remaining community and node are calculated as target, successively comprising the most preceding K communities of interstitial content
With their similarity, remaining community and node merger are entered with K community before their similarity highests, set C is updatedL,
Finally export community's set CLIt is used as final division result.
3. as claimed in claim 2 optimize many relation community discovery methods expanded with seed, its feature based on composition of relations
It is, using normalised mutual information NMI, is calculated as follows：
<mrow>
<mi>N</mi>
<mi>M</mi>
<mi>I</mi>
<mrow>
<mo>(</mo>
<mi>A</mi>
<mo>,</mo>
<mi>B</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<mo>-</mo>
<mn>2</mn>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<msub>
<mi>C</mi>
<mi>A</mi>
</msub>
</msubsup>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<msub>
<mi>C</mi>
<mi>B</mi>
</msub>
</msubsup>
<msub>
<mi>H</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<msub>
<mi>H</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mi>N</mi>
</mrow>
<mrow>
<msub>
<mi>H</mi>
<mrow>
<mi>i</mi>
<mo>.</mo>
</mrow>
</msub>
<msub>
<mi>M</mi>
<mrow>
<mo>.</mo>
<mi>j</mi>
</mrow>
</msub>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<msub>
<mi>C</mi>
<mi>A</mi>
</msub>
</msubsup>
<msub>
<mi>H</mi>
<mrow>
<mi>i</mi>
<mo>.</mo>
</mrow>
</msub>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>H</mi>
<mrow>
<mi>i</mi>
<mo>.</mo>
</mrow>
</msub>
<mo>/</mo>
<mi>N</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<msub>
<mi>C</mi>
<mi>B</mi>
</msub>
</msubsup>
<msub>
<mi>H</mi>
<mrow>
<mo>.</mo>
<mi>j</mi>
</mrow>
</msub>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>H</mi>
<mrow>
<mo>.</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>/</mo>
<mi>N</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>;</mo>
</mrow>
Wherein confusion matrix H row represents real community division result, and H row represent community's knot that partitioning algorithm is drawn
Really；CAAnd CBA, community's number that two results of B are included, H are represented respectivelyijRepresentative should be but appeared in community j in community i
Node number, HiAnd HjRepresent the nodes sum of the community of ith row and jth column respectively, N is contained section in whole network
Count out；As NMI=1, the community structure for representing two results of A and B is identical, and NMI=0, interval scale is completely different.
4. as claimed in claim 2 optimize many relation community discovery methods expanded with seed, its feature based on composition of relations
It is, community's division is carried out to the network of four relation compositions using single relation community discovery algorithm BGLL respectively, statistics is divided
As a result according to formulaObtain cohesion matrix I, wherein IijFor node i, j corresponding values in a matrix, N is
The sum of dimension is 4, m represents which dimension, C herem(i, j) represents at 2 points and whether is divided in same community in m dimensions,
If 2 points of C if m-th of dimension is divided into same communitym(i, j) value 1, otherwise being 0；Ergodic Matrices I is small by all values
Number in 4 is all set to zero, obtains matrix of areas F；Depth-first traversal is carried out to matrix F, its all connection is obtained
Region, the region by subregion interstitial content more than 1 is stored in candidate seed community's set CL, L=0.
5. as claimed in claim 2 optimize many relation community discovery methods expanded with seed, its feature based on composition of relations
It is, by single relational network M come the similarity sim between each node in calculating network:
<mrow>
<mi>S</mi>
<mi>i</mi>
<mi>m</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mi>K</mi>
<mi>N</mi>
</mfrac>
<mo>*</mo>
<mfrac>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>l</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>c</mi>
<mi>o</mi>
<mi>m</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
</mrow>
</msubsup>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<mi>W</mi>
<mo>(</mo>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>l</mi>
</mrow>
<mo>)</mo>
<mo>,</mo>
<mi>W</mi>
<mo>(</mo>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>l</mi>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>l</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>N</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
</mrow>
</msubsup>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
<mrow>
<mo>(</mo>
<mi>W</mi>
<mo>(</mo>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>l</mi>
</mrow>
<mo>)</mo>
<mo>,</mo>
<mi>W</mi>
<mo>(</mo>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>l</mi>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>+</mo>
<mi>W</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>*</mo>
<mi>&Delta;</mi>
<mo>;</mo>
</mrow>
Wherein K represents that i, j are divided into the dimension number of identical community, N=4, node i and section when Δ gain factor is 1, W (i, j)
Point j connection side right weight, two nodes neighbors nodes of com (i, j) occur simultaneously, and N (i, j) is the neighbor node union of two nodes.
6. as claimed in claim 2 optimize many relation community discovery methods expanded with seed, its feature based on composition of relations
It is, chooses and SiSimilarity exceedes the community of given threshold γ=1 and node is put into candidate regions, calculates SiCurrent fitness F with
And the current fitness F of each community of candidate regionsh, calculation formula is as follows：
<mrow>
<mi>F</mi>
<mrow>
<mo>(</mo>
<mi>c</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<msubsup>
<mi>w</mi>
<mrow>
<mi>i</mi>
<mi>n</mi>
</mrow>
<mi>c</mi>
</msubsup>
<msup>
<mrow>
<mo>(</mo>
<msubsup>
<mi>w</mi>
<mrow>
<mi>i</mi>
<mi>n</mi>
</mrow>
<mi>c</mi>
</msubsup>
<mo>+</mo>
<msubsup>
<mi>w</mi>
<mrow>
<mi>o</mi>
<mi>u</mi>
<mi>t</mi>
</mrow>
<mi>c</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mi>&alpha;</mi>
</msup>
</mfrac>
<mo>;</mo>
</mrow>
Wherein formulaIt is twice of the weight sum of community's internal edges,It is the weight sum with community's external edge, α is
Regulate and control the parameter of community's scale, value is set to one；
Then community and node and the S of candidate assembly section are calculated respectivelyiFitness value after merging is designated as, and then calculates FnewRelatively
In SiThe growth rate V of fitness F originallysAnd candidate community fitness FhGrowth rate Vh；
In VsAnd VhBoth greater than on the premise of growth rate threshold value δ=0.1, V is chosens+VhThat maximum community and SiIt is merged into one
Individual new communities, update SiWith seed community set CL；If without qualified, checking set CLIn existing community whether all
Merged；Will set C if beingLIn all communities merging record reset；
Community's number is checked, is waited and K if being less than, output result, EP (end of program), output set CL；If more than K, to include section
Most preceding K communities count out for target, remaining community and their similarity are calculated successively, remaining community and node are returned
It is incorporated to K community before their similarity highests, updates set CL, finally export community's set CLIt is used as final division result.
7. optimize many relation communities expanded with seed described in a kind of application claim 1~6 any one based on composition of relations
It was found that the community network of method.
8. optimize many relation communities expanded with seed described in a kind of application claim 1~6 any one based on composition of relations
It was found that the computer of method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710260510.6A CN107169871B (en) | 2017-04-20 | 2017-04-20 | Multi-relationship community discovery method based on relationship combination optimization and seed expansion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710260510.6A CN107169871B (en) | 2017-04-20 | 2017-04-20 | Multi-relationship community discovery method based on relationship combination optimization and seed expansion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107169871A true CN107169871A (en) | 2017-09-15 |
| CN107169871B CN107169871B (en) | 2020-08-28 |
Family
ID=59813344
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710260510.6A Active CN107169871B (en) | 2017-04-20 | 2017-04-20 | Multi-relationship community discovery method based on relationship combination optimization and seed expansion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107169871B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106776781A (en) * | 2016-11-11 | 2017-05-31 | 深圳云天励飞技术有限公司 | A kind of human relation network analysis method and device |
| CN108648094A (en) * | 2018-05-08 | 2018-10-12 | 阿里巴巴集团控股有限公司 | A kind of community discovery method, device and equipment |
| CN108876116A (en) * | 2018-05-30 | 2018-11-23 | 西北工业大学 | A kind of creation effect knowledge recommendation method of Facing to Manufacturing technical optimization |
| CN112528166A (en) * | 2020-12-16 | 2021-03-19 | 平安养老保险股份有限公司 | User relationship analysis method and device, computer equipment and storage medium |
| CN113395172A (en) * | 2021-05-18 | 2021-09-14 | 中国电子科技集团公司第五十四研究所 | Important user discovery and behavior prediction method based on communication network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663499A (en) * | 2012-03-11 | 2012-09-12 | 西安电子科技大学 | Network community division method based on simulated annealing genetic algorithm |
| CN102768735A (en) * | 2012-07-04 | 2012-11-07 | 西安电子科技大学 | Network community partitioning method based on immune clone multi-objective optimization |
| CN104200272A (en) * | 2014-08-28 | 2014-12-10 | 北京工业大学 | Complex network community mining method based on improved genetic algorithm |
| CN104268629A (en) * | 2014-09-15 | 2015-01-07 | 西安电子科技大学 | Complex network community detecting method based on prior information and network inherent information |
-
2017
- 2017-04-20 CN CN201710260510.6A patent/CN107169871B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663499A (en) * | 2012-03-11 | 2012-09-12 | 西安电子科技大学 | Network community division method based on simulated annealing genetic algorithm |
| CN102768735A (en) * | 2012-07-04 | 2012-11-07 | 西安电子科技大学 | Network community partitioning method based on immune clone multi-objective optimization |
| CN104200272A (en) * | 2014-08-28 | 2014-12-10 | 北京工业大学 | Complex network community mining method based on improved genetic algorithm |
| CN104268629A (en) * | 2014-09-15 | 2015-01-07 | 西安电子科技大学 | Complex network community detecting method based on prior information and network inherent information |
Non-Patent Citations (1)
| Title |
|---|
| 韩忠明 等: "NCSS: 一种快速有效的复杂网络社团划分算法", 《中国科学》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106776781A (en) * | 2016-11-11 | 2017-05-31 | 深圳云天励飞技术有限公司 | A kind of human relation network analysis method and device |
| CN106776781B (en) * | 2016-11-11 | 2018-08-24 | 深圳云天励飞技术有限公司 | A kind of human relation network analysis method and device |
| CN108648094A (en) * | 2018-05-08 | 2018-10-12 | 阿里巴巴集团控股有限公司 | A kind of community discovery method, device and equipment |
| CN108876116A (en) * | 2018-05-30 | 2018-11-23 | 西北工业大学 | A kind of creation effect knowledge recommendation method of Facing to Manufacturing technical optimization |
| CN108876116B (en) * | 2018-05-30 | 2021-08-24 | 西北工业大学 | Manufacturing effect knowledge recommendation method oriented to manufacturing technology optimization |
| CN112528166A (en) * | 2020-12-16 | 2021-03-19 | 平安养老保险股份有限公司 | User relationship analysis method and device, computer equipment and storage medium |
| CN113395172A (en) * | 2021-05-18 | 2021-09-14 | 中国电子科技集团公司第五十四研究所 | Important user discovery and behavior prediction method based on communication network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107169871B (en) | 2020-08-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107169871A (en) | It is a kind of to optimize many relation community discovery methods expanded with seed based on composition of relations | |
| CN110532436B (en) | Cross-social network user identity recognition method based on community structure | |
| Sun et al. | Identification of overlapping and non-overlapping community structure by fuzzy clustering in complex networks | |
| CN103559407B (en) | A kind of commending system for measuring direct graph with weight interior joint cohesion and method | |
| CN104102745B (en) | Complex network community method for digging based on Local Minimum side | |
| Wang et al. | Intelligent multivariable air-quality forecasting system based on feature selection and modified evolving interval type-2 quantum fuzzy neural network | |
| CN112036445B (en) | User identity recognition method across social networks based on neural tensor network | |
| CN102413029A (en) | Method for partitioning communities in complex dynamic network by virtue of multi-objective local search based on decomposition | |
| CN110232434A (en) | A kind of neural network framework appraisal procedure based on attributed graph optimization | |
| CN110473592A (en) | The multi-angle of view mankind for having supervision based on figure convolutional network cooperate with lethal gene prediction technique | |
| CN103020163A (en) | Node-similarity-based network community division method in network | |
| CN108038746A (en) | Method is recommended based on the bigraph (bipartite graph) of key user and time context | |
| CN108765180A (en) | The overlapping community discovery method extended with seed based on influence power | |
| CN107945037A (en) | A kind of social networks based on node structure feature goes de-identification method | |
| CN107784327A (en) | A kind of personalized community discovery method based on GN | |
| Babers et al. | A nature-inspired metaheuristic lion optimization algorithm for community detection | |
| CN110442800A (en) | A kind of semi-supervised community discovery method of aggregators attribute and graph structure | |
| Azizifard | Social network clustering | |
| CN103051476B (en) | Topology analysis-based network community discovery method | |
| Kibanov et al. | On the evolution of contacts and communities in networks of face-to-face proximity | |
| CN105719190A (en) | Social network node influence measuring method based on triangle structures | |
| CN109886313A (en) | A kind of Dynamic Graph clustering method based on density peak | |
| CN111147311A (en) | Network structure difference quantification method based on graph embedding | |
| Zhou et al. | Hgena: A hyperbolic graph embedding approach for network alignment | |
| Sikarwar et al. | A Review on Social Network Analysis Methods and Algorithms |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |