CN104090957A - Heterogeneous network interactive visualization method - Google Patents
- ️Wed Oct 08 2014
CN104090957A - Heterogeneous network interactive visualization method - Google Patents
Heterogeneous network interactive visualization method Download PDFInfo
-
Publication number
- CN104090957A CN104090957A CN201410327034.1A CN201410327034A CN104090957A CN 104090957 A CN104090957 A CN 104090957A CN 201410327034 A CN201410327034 A CN 201410327034A CN 104090957 A CN104090957 A CN 104090957A Authority
- CN
- China Prior art keywords
- node
- cluster
- network
- clustering
- neighbor Prior art date
- 2014-03-10 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/904—Browsing; Visualisation therefor
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
本发明公开了一种异构网络可交互可视化方法。本方法为：1)对异构网络中的节点按照节点属性的取值进行聚类，并生成相应的聚类网络可视化图；2)对所选聚类结果中的每一节点，计算该节点的邻居节点集合；然后按照邻居节点集合节点属性值对所选聚类结果中的每一节点进行聚类；并生成聚类可视化图作为下一级可视图；3)选取步骤2)的若干聚类结果，对所选聚类结果中的每一节点根据其邻居节点集合进行聚类，将具有相同邻居节点集合的节点划分到同一聚类中；然后将此次聚类结果生成聚类可视化图并将其作为步骤2)所生成聚类网络可视化图的下一级可视图。本发明结合了拓扑和属性信息，使用户能够用更细的粒度进行更低层级的查看。
The invention discloses a heterogeneous network interactive visualization method. The method is as follows: 1) cluster the nodes in the heterogeneous network according to the value of the node attribute, and generate a corresponding clustering network visualization diagram; 2) calculate the node The set of neighbor nodes; then each node in the selected clustering result is clustered according to the node attribute value of the set of neighbor nodes; and a cluster visualization diagram is generated as a next-level visual view; Class results, cluster each node in the selected clustering result according to its neighbor node set, and divide the nodes with the same neighbor node set into the same cluster; then generate a cluster visualization graph for this clustering result And use it as the next-level visualization of the clustering network visualization generated in step 2). The invention combines topology and attribute information, enabling users to view at a lower level with finer granularity.
Description
Technical field
The present invention relates to the fields such as cluster analysis, heterogeneous network, data mining, Analysis of Topological Structure, the analysis of large data network visualization, proposed a kind of method of the large-scale heterogeneous network data of processing that combine based on heterogeneous network nodal community and network topology structure.The method is applicable to the typical information network datas such as social network, computer network, sensor network and knowledge network.It is a kind of visual presentation method that can carry out interaction analysis.
Background technology
The arrival of large data age, has produced large data network complicated and changeable at short notice, and a lot of people start the network data large for this tittle, matter is assorted and analyze and further investigate.In these data networks, some are the network types on network node with different attribute feature, and some are network types of topological structure and incidence relation complexity.The heterogeneous network data of indication of the present invention mainly refer to have the network of above two kinds of features simultaneously, by some, have the comparatively complicated relationship/association network that the back end of different attribute, type forms.Current, more existing researchs are to carry out cluster analysis for heterogeneous network data.Be mainly the different attribute having according to node in network data, carry out cluster visual analyzing.Meanwhile, network topology structure mainly refers to because of certain incidence relation, connect each other between back end and back end the network of formation, and the cluster analysis of topological structure Network Based is also comparatively common, as figure cluster, Spectral Clustering.
In visual research field, most network data visual research all concentrates on the visual of network topology structure.Along with the arrival in web2.0 epoch, data volume constantly increases, and very large change has occurred the form of network data.Node and their limit that more existing data networks are a lot of have all had attribute, visual for this heterogeneous network with attribute, still seldom have now research.And known method all can not be carried out Conjoint Analysis in conjunction with network node attribute and topological structure, can not based on this two category information, realize visual presentation and analysis simultaneously.For example, Wattenberg is studied the visual analyzing of heterogeneous network.But it is visual that he has only utilized network node attribute to realize, for above two kinds of heterogeneous network informations, do not carry out the visual research that the two combines.
Particularly, as follows for studying a question of the large-scale heterogeneous network data visualization of this class:
1) existing heterogeneous network data visualization cluster analysis research, having some is cluster analyses of nodal community data Network Based, some are cluster analyses of topology data Network Based, clustering method is the single source of Adoption Network topological structure or nodal community only, can not realize in conjunction with two class data types and carry out cluster analysis simultaneously.. for example, in the academic heterogeneous network that comprises " paper, author, meeting/periodical " three category nodes, existing visual analysis method can not intuitively be shown heuristic problems such as " which author's co-worker have delivered the more high/medium/low paper of quoting, and the paper which author delivers has approximate citation rate to distribute ".
Existing large-scale heterogeneous network data visualization cluster analysis research, can, according to network node attribute or topological structure, generate the network node cluster of some.Yet, the cluster result granularity of these two kinds of methods or too coarse to such an extent as to lose a large amount of network details (if the method based on nodal community is by the several node clusterings of whole network boil down to, or too fine and closely woven so be difficult to layout and displaying (as the cluster of structure Network Based) several values of corresponding node attribute).Although existing certain methods can clusters number be input (as figure dividing method), yet user is difficult to control and understand cluster result, and can not support user-defined top-down storage optimization analytic process.
Summary of the invention
For the problem analysis of above-mentioned large-scale heterogeneous network data, the object of the invention is to propose a kind of heterogeneous network can interactive visual method.The invention provides the viewdata analytical approach that can address the above problem with and the Visual Implementation form---onion diagram.This be a kind of can be according to the attribute of arbitrary node itself, topological relation, or the large-scale heterogeneous network data analysing method of cluster analysis and visual presentation is carried out in the mixing of two category informations.Each network node or node clustering can carry out secondary fractionation or cluster according to attribute or topological relation flexibly.On visual representing, network node, according to the difference of the cluster level at place, shows with a plurality of concentric circless, image as the same onion being cut open.Therefore, our method is named with onion diagram.
The present invention is based on the support of national 973 science and technology items (Supportedby the National Basic Research Program of China under Grant No.2014CB340301) and state natural sciences fund (the National Science Foundation of China (NSFC) under Grant No.61379088), by analyzing attributive character and the topological relation of network data, by the network node cluster of large-scale heterogeneous network data compression composition level, the network node cluster that provides visual pattern (onion diagram) intuitively to show to calculate and the relation between cluster, and by simple, easy-to-use interactive form guides user stratification to browse the also cluster result of the large-scale heterogeneous network data of visual analyzing.Especially, this method supports user according to the analysis demand of oneself, select specific heterogeneous network local data, and from by the network node cluster of selected certain level, continue successively to amplify (to lower level cluster) or merge (to lower level cluster) according to analysis demand.By show the cluster granularity of different levels in same heterogeneous network visualization view kind, realize the detailed association analysis of the local heterogeneous network data of user's concern.
In specific implementation, this method realizes above-mentioned mutual by button and the common mouse action at interface, support user to browse as required and analyze large-scale heterogeneous network data: user both can see under higher level cluster state apparent according to a few node set of the different value clusters of nodal community, also can distribute and incidence relation by heterogeneous network data clusters visualization view being expanded under thousands of more oligomeric class hierarchy state alternately.
To achieve these goals, the technical solution adopted in the present invention is:
Heterogeneous network can an interactive visual method, the steps include:
1) node in pending heterogeneous network is carried out to cluster according to the value of the nodal community of choosing or setting, and generate the visual figure of corresponding clustering network;
2) selecting step 1) some cluster results, and to each node in selected cluster result, according to the topological structure of this heterogeneous network, calculate the neighbor node set of this node; Then according to the neighbor node collector node property value of choosing or setting, each node in selected cluster result is carried out to cluster; In the cluster result obtaining, it is identical and similar at the topology location of heterogeneous network that the node that is arranged in same cluster has node attribute values; Then this cluster result is generated to the visual figure of cluster and using it as step 1) the next stage Visual Graph of the visual figure of clustering network;
3) selecting step 2) some cluster results, and to each node in selected cluster result, according to its neighbor node set, carry out cluster, by the node division with identical neighbor node set in same cluster; In the cluster result obtaining, it is identical and identical at the topology location of heterogeneous network that the node that is arranged in same cluster has node attribute values; Then this cluster result is generated to the visual figure of cluster and using it as step 2) the next stage Visual Graph of the visual figure of clustering network that generates.
Further, described heterogeneous network is oriented heterogeneous network G=(V, E); Wherein, V={v 1..., v nexpression set of network nodes, E={e 1..., e mthe set of expression network edge; The adjacency matrix of heterogeneous network is W, element w wherein ijrepresent connected node v ito node v jlimit; For each node v i, N +(v i)={ v j| w ij=1} representation node v igo out to neighbor node set, N -(v i)={ v j| w ji=1} is node v ienter to neighbor node set.
Further, described step 2) or step 3) in clustering method be:
31) by node v ineighbor node set expression be neighbour vector R (v i)={ c i1..., c it, c 1i..., c ti; Wherein, c itfor node v it go out to neighbor node, c tifor node v it enter to neighbor node;
32) for all network nodes of need dividing, according to the vector calculation of its neighbor node set node degree of approximation distance between node between two;
33) according to similarity distance between the node calculating and expectation clusters number, by node division, be k cluster.
Further, each node is adopted and unifies a line vector representation; Described unified row vector comprises: the neighbor information of node attribute information and value, node and neighbours' attribute value; During each cluster, the cryptographic hash of the unified row vector of computing node, is same cluster by the identical node division of cryptographic hash; Wherein, carry out step 1) cluster time, the unified row vector of node comprises node attribute information and value, other values are empty; Carry out step 2) cluster time, neighbours' attribute value that the unified row vector of node comprises node attribute information and value, node; During cluster step 3), the neighbor information that the unified row vector of node comprises node attribute information and value, node.
Further, this pending heterogeneous network being generated to visual figure as first order Visual Graph, step 1) the visual figure of clustering network that generates is the next stage Visual Graph of this first order Visual Graph; By step 3) node in each cluster result generate a visual figure and using it as step 3) the next stage Visual Graph of the visual figure of clustering network that generates, i.e. minimum particle size Visual Graph.
Further, different according to the property value that carries out cluster of choosing or setting, the visual figure of next stage that comprises one or more granularities in the visual figure of same one-level.
The sum of the original heterogeneous network relationships between nodes that further, between this cluster of mark, relation comprises in the annexation between each cluster is as the strength of joint of annexation between this cluster.
(1) principle summary of realizing of the present invention
The main heterogeneous network back end attribute that adopts is in conjunction with the hierarchical cluster analysis method of topological structure: at higher level, and by nodal community value cluster, the main macroscopic view association situation of supporting to check from the overall situation whole large-scale heterogeneous network; At lower level, on nodal community basis, add network topological information, and by this mixed information, existing cluster is carried out to secondary cluster, each original cluster can be divided into some sub-clusters.Meanwhile, utilize visual Cross support user to be transitioned into lower level cluster from lower level cluster, to check more network associate detailed information.The main target of the hierarchy relating to is here to guarantee that the information that each lower level is shown all comprises more information than its a upper level, and the content of simultaneously showing is also more complicated.
Existing network data method for visualizing also can be realized certain hierarchical cluster effect.But, their only Adoption Network topology data or only Adoption Network nodal community, and our method can be in conjunction with these the two kinds of data types in heterogeneous network.And existing hierarchical cluster method is all to have presupposed a hierarchical structure, user can only browse according to existing cluster path, and can not, according to user's particular demands, realize online secondary cluster and complete visual analyzing.Existing hierarchical clustering algorithm is very responsive to input parameter, is unfavorable for checking alternately, and user is the real meaning of the visualization result of indigestion interaction analysis operation gained also.
By contrast, onion diagram hierarchy and method for visualizing that we propose have utilized more reasonable, comprehensive clustering algorithm, can be in conjunction with network topology structure and nodal community, and the result of each hierarchy has been carried out to definition clearly, algorithm does not need core parameter input substantially.User can know the implication that each the step interactive operation corresponding data of oneself is analyzed very clearly.User also can carry out layering visual presentation according to different heterogeneous network data characteristics and demand.
Onion diagram method is built-in, and localized network launches and two kinds of data manipulation methods of global data filtration, can utilize localized network method, and hierarchical network data are deployed into a lower rank level from high-level.By overall filter operation, user can, according to the different attribute value on network node, filter out the details network data that is not very important.Like this, just can represent more clearly the associated trend map of analyzing whole network data.
(2) onion diagram hierarchical cluster structure
Onion diagram structure main thought, for large-scale heterogeneous network data are carried out to cluster by many algorithms, is processed into the data model with five layer network data clusters structures, as shown in Figure 1.Wherein, every layer data structures can be calculated in real time in reciprocal process, also can pass through off-line pre-service calculated in advance, to accelerate visual analyzing response speed, improves user's experience.
Above-mentioned five layers of heterogeneous network data hierarchy structure and method are as follows, and concrete implementation algorithm sees below literary composition:
Ground floor: first large-scale heterogeneous network data are imported, as data source and basic data.At ground floor, whole network is gathered into a start node, as the starting point of mutual visual analysis.
The second layer: nodes is carried out to cluster according to node attribute values by algorithm.First select an attribute, for example the sex of node (nature person) in community network, carries out cluster by the node on this attribute with identical value, and generates the visual figure of corresponding clustering network.Meanwhile, support that the cluster based on a plurality of attributes is shown, be about to catenet according to selected a plurality of nodal community values, as sex and the age of node in community network (nature person), carry out more fine-grained cluster.By localized network, launch, be also supported in the cluster that comprises multiple granularity in same view visual, if some node is according to a property value cluster, other node carries out cluster according to the value on a plurality of attributes.
The 3rd layer: on according to the cluster result of node attribute values, further according to the property value of the neighbor node set of node, complete secondary cluster, with the local heterogeneous network of further refinement, show.The neighbor node set is here calculated according to network topology structure, refers to all set that are comprised of the neighbor node of annexation of certain node.Cluster attribute used can generate automatically, adopts last layer to be used for the nodal community of cluster; Also can according to analyzing again, be selected by user.The property value cluster detailed method of pressing neighbor node set is as follows: for any one node, find its neighbor node set, then according to the value of all nodes on selected attribute in this set, the neighbor node property value that calculates them distributes.For example, be set in community network with naturally artificial node, selected attribute is sex, and the neighbor node property value of certain node is distributed as male sex neighbours 3 people, women neighbours 2 people.The distribution of neighbor node property value equates or similar node is classified as same cluster.Especially, the neighbor node property value here distributes and equates not consider the neighbor node number on certain property value, only considers whether to have on certain property value neighbor node.In other words, adopt neighbor node property value set to realize secondary cluster here.The method of consideration neighbor node number on certain property value realizes by approximate data, sees below civilian fuzzy algorithm.Here it needs to be noted, the node clustering of this level combines attribute information and the topology of networks thereof on heterogeneous network node.Meanwhile, on the node of some scope that the cluster analysis of this level also can be chosen user, complete, and other subnetworks still keep the more cluster result of coarseness of upper strata.When practical operation, the upper strata cluster node that user can choose certain or certain several hope to check in detail, by above-mentioned algorithm, the upper strata cluster node of choosing is refined as to the while according to the cluster result of network node attribute and its neighbor node attribute value, thereby realizes the demand that user checks more detailed localized network.In the cluster result generating according to this algorithm, different nodes in same cluster are nodal community value identical (on the selected attribute of user) not only, its topology location in heterogeneous network is also identical/similar, is connected with the neighbours of identical/like attribute.
The 4th layer: a similar upper method, user can, by mutual visual analysis, continue segmentation to the cluster result having generated, to check in more detail localized network structure.This hierarchical clustering is inferior more careful than last layer, and the node that requires the to be divided into same cluster not only property value of neighbor node is identical/similar, and its neighbor node set is also consistent.Due in real network, seldom, so most of fuzzy algorithm that adopts of this hierarchical method, calculates the similar cluster of neighbor node set to the identical case of neighbor node set.This cluster is at computer network, and especially network safety filed has very important application.In DDOS (distributed denial of service) Attack Scenarios, the computing machine of being handled is in a large number by client's target machine of the designated attack of access to netwoks.Although the computing machine of being handled belongs to different segment, physically interval is ten thousand li, the client computer set connecting by them, i.e. and neighbor node in computer network, can by the computing machine of being handled by the same attack analysis that gathers together.Meanwhile, client's target machine of being attacked also can adopt identical method to gather cluster according to attacking starter.
Layer 5: user can further segment to single node granularity node clustering by the method for visual analyzing, and node clustering is fully expanded, each node is independent demonstration all.Similarly, in same view, for guaranteeing that whole vision complexity does not exceed the intelligible scope of user, most of situation selection portion subnetwork is realized the expansion of layer 5.
It should be noted that, above layered approach is some clusters by the node division of large-scale heterogeneous network, its incidence relation is also attached between cluster accordingly, it is each incidence relation of initial heterogeneous network data, to be attached between cluster corresponding to its source, destination node, if any relating heading, direction is constant.
In superincumbent onion diagram hierarchy, at third and fourth level, this method is supported to carry out fuzzy clustering (by similarity cluster) according to the user-selected granularity of browsing simultaneously, and allows the clusters number in designated result, thereby has guaranteed the Clustering Effect that user is controlled.
(3) visualization scheme
Visualization scheme is also the pith of this method.We have selected, and " " concept is carried out the heterogeneous network hierarchy of corresponding our method to concentric circles to onion " this intention form, extracted onion metaphor ", intuitively visual to realize.In specific design, when onion metaphor coherent element has all been chosen in the aspects such as form and color, for the node of different attribute value, we have also designed significantly different visualized graphs on pattern.We also more innovate easy-to-use visualization interface for this visualization scheme has designed specially.Meanwhile, on onion diagram, also support a series of statistical graphs to show, be convenient to query analysis association attributes and related information better.User only need to carry out shirtsleeve operation and can carry out visual screening and check.
(4) interaction schemes
Onion diagram provides the interactive operation of high availability and has controlled interface, and user, when successively browsing large-scale heterogeneous network data message, can operate by hierarchical clustering algorithm and local expansion, realizes mutual visual analysis.These exchange methods comprise:
1) user is according to the required granularity representing, and input is to the expansion level of heterogeneous network (level 1～5).This method is carried out cluster according to input to heterogeneous network, and result is represented by visualization view, is switched to the view launching according to selected level.This exchange method comes into force to the whole heterogeneous network overall situation, and, after interactive operation, in view, the cluster granularity of heterogeneous network is identical.
2) user is according to the required granularity representing and the heterogeneous network scope paid close attention to, and input is to the expansion level of heterogeneous network (level 1～5) and need the subrange of the heterogeneous network that launches.This method is carried out cluster according to input to the part of choosing of heterogeneous network, and result is represented by visualization view.Unchecked network others part is constant.Incidence relation is still according to being related to addition method displaying between cluster.
3) result that user obtains according to the required granularity representing, the heterogeneous network scope of paying close attention to and hope, the subrange of the heterogeneous network that input launches the expansion level of heterogeneous network (level 1～5), needs and the clusters number that hope obtains.This method be similar to cluster according to input to the part of choosing of heterogeneous network, guarantees the clusters number that the localized network of the choosing generation user of institute inputs, and result is represented by visualization view.Unchecked network others part is constant.Incidence relation is still according to being related to addition method displaying between cluster.
Especially, above user's input is all inputted by visualization interface, the hierarchy selection of cluster granularity can be set by mutual slide block, mouse frame choosing is selected to adopt in the part of heterogeneous network, mutual execution also can trigger by simple mouse action (as double-clicked mouse, will choose heterogeneous network to expand one deck to fine granularity more, or promote one deck to coarseness more).These interactive operation assisted users are level and smooth browses large-scale heterogeneous network data.Fig. 1 has provided the schematic diagram of this kind of interaction results.
In existing visual research, user can only see the information of global network association, or adopts data filtering to check the internal correlation information of a certain generic attribute value.And in our method, user can be by above interactive operation, large-scale heterogeneous network is divided into a plurality of localized networks, for different localized networks according to different levels and initial conditions cluster, to check the different local isomery incidence relations after a plurality of different levels clusters of network simultaneously.What user also can go forward one by one again launches one of them cluster or a plurality of cluster in same interface.This has great convenience for the user a plurality of categorical attribute value condition to be checked simultaneously, and compare analysis between a plurality of grouped data value.The functions such as meanwhile, our method is also supported initialization (get back to initial clustering level set), historical rollback (getting back to last cluster level sets), facilitate user to carry out iteration, mutual visual analysis operation.Meanwhile, the instrument that this method for visualizing is realized also provides the Query List of relevant information, facilitates the more careful every terms of information of comprehensively checking cluster node in these heterogeneous network data of user.
Compare with existing large-scale heterogeneous network data analysis technique, innovative point of the present invention is:
(1) this method supports top-down hierarchy to carry out cluster visual analysis.Cluster algorithm combines heterogeneous network topological sum network node attribute two parts information, and this is that method in the past is not all supported.
(2) this method support is refined as fine granularity cluster from overall coarseness cluster, and by algorithm, supports the deployment analysis that goes forward one by one of localized network.This operation can realize in localized network, to heterogeneous network different piece according to different grain size cluster, and the incidence relation between checking.
(3) visual concept of this method introducing " onion " represents different hierarchical relationships, represents the hierarchical information at place by drawing the concentric circles of some.By the corresponding relation of the concentrically ringed number of balance and level, reach the visual complexity of the abstract view of controlling large-scale heterogeneous network data, the more friendly browse mode for user is provided.
Accompanying drawing explanation
Fig. 1 is onion diagram hierarchy principal diagram.
Fig. 2 is that the second level is by network node hierarchical cluster attribute result schematic diagram.
Fig. 3 is that tri-layer is by the relative normal equivalent cluster result of network topology structure schematic diagram.
Fig. 4 is that the 4th level is by network topology structure absolute structure cluster result schematic diagram of equal value.
Fig. 5 is that the node in onion diagram hierarchical clustering algorithm performing step is unified row vector design.
Fig. 6 is onion diagram method for visualizing schematic diagram.
Fig. 7 utilizes the academic network data in the visual field of onion diagram visual analysis-launch (by nodal community) according to heterogeneous nodes type.
Fig. 8 utilizes the academic network data in the visual field of onion diagram visual analysis-further launch (by nodal community) according to meeting/periodical type.
Fig. 9 utilizes the academic network data in the visual field of onion diagram visual analysis-further launch (by nodal community) according to papers quoted type.
Figure 10 utilizes the academic network data in the visual field of onion diagram visual analysis-further to launch (by neighbor node attribute) according to author's situation of quoting that publishes thesis.
Figure 11 is used statistical graph to show neighbor node property distribution on Figure 10 basis.
Embodiment
Divide the concrete elaboration of four parts main contents of the present invention below.
1. algorithm design
The basic symbol definition of the following arthmetic statement of given first.
We represent an oriented heterogeneous network with G=(V, E).Wherein, V={v 1..., v nexpression set of network nodes, E={e 1..., e mexpression network edge set (being incidence relation).W represents the adjacency matrix that network is corresponding, each element w ij=1 represents a connected node v ito node v jlimit.For each node v i, N +(v i)={ v j| w ij=1}, N -(v i)={ v j| w ji=1} is representation node v respectively igo out to neighbor node set and enter to neighbor node set.Make D={d 1..., d srepresent the nodal community set of this heterogeneous network G to comprise altogether s nodal community.D(v i)={ d 1(v i) ..., d s(v i) expression node v ivalue on all s nodal community, d k(v i) expression node v ivalue on k nodal community.
The concept that our Adoption Network is divided is described the clustering method to heterogeneous network data.Be that each clustering algorithm is to be all some clusters by all node division in selected network, each node is present in and exists only in a cluster of same layer.Another network division result be described as P:V → 1,2 ..., t}, P represents a mapping, corresponding cluster numbering after each network node is mapped to network and divides.All-network node is divided into t cluster, P (v altogether i) be node v icluster corresponding after division is numbered.
According to the description of front portion, in the five-layer structure of onion diagram, all nodes of ground floor are same cluster, layer 5 for each node be a cluster.This two-layer particular algorithm that do not need.Below we mainly describe second and third, the algorithm design of four strata classes.
(1) network node attribute clustering algorithm (second layer)
According to network node attribute, heterogeneous network is divided, and generated cluster.
Algorithmic descriptions:
According to the community set for cluster of appointment network node is divided into a plurality of clusters, and the node in each cluster is in set
dvalue on built-in attribute is all identical.Make this layer of clustering algorithm map network division result P, for any two the node V that are divided into same cluster iand V j, its nodal community value is also identical, and algorithm is equivalent to:
P ( v i ) = P ( v j ) ⇔ D ‾ ( v i ) = D ‾ ( v j )
Fig. 2 has provided the arithmetic result example according to network node hierarchical cluster attribute.In figure, five, left side node is at selected attribute
dupper value is I, and on the node of five, right side, value is II, therefore according to this method, at this level, is divided into two clusters, respectively the solid node group in corresponding left side and the hollow node group in right side.Same, when user selects a plurality of network node attribute, still the value on a plurality of attributes according to network node, determines its cluster result.
(2) the relative normal equivalent clustering algorithm of network topology structure (the 3rd layer)
According to heterogeneous network topological structure and nodal community, carry out Conjoint Analysis, computational grid is divided, and generates cluster.
Algorithmic descriptions:
According to the nodal community set of upper strata Clustering and selection
dand map network node division P 0, adopt relative normal equivalent clustering algorithm, calculate this layer of new network and divide P.P meets, for any two node V iand V j:
P ( v i ) = P ( v j ) ⇔ P 0 ( v i ) = P 0 ( v j ) and P 0 ( N + ( v i ) ) = P 0 ( N + ( v j ) ) and P 0 ( N - ( v i ) ) = P 0 ( N - ( v j ) )
The node that to be cluster divide to consolidated network meets upper strata and divides the upper strata of identical and its neighbor node set and divide identical.Fig. 3 has provided this layer by the arithmetic result example of the relative normal equivalent cluster of network topology structure.Fig. 3 is the division on the basis of Fig. 2, and original two clusters are refined as to four clusters, by node shape and filling sign.Its attribute of the node of same cluster and neighbor node community set are all identical.
(3) network topology structure absolute structure clustering algorithm (the 4th layer) of equal value
According to heterogeneous network topological structure and nodal community, carry out Conjoint Analysis, computational grid is divided, and generates cluster.This level is divided with last layer similar, but stricter, and not only neighbor node attribute is identical to require to belong to the node of same cluster, and neighbor node itself is also identical.
Algorithmic descriptions:
According to the nodal community set of upper strata Clustering and selection
dand map network node division P 0, adopt absolute structure clustering algorithm of equal value, calculate this layer of new network and divide P.P meets, for any two node V iand V j:
P ( v i ) = P ( v j ) ⇔ P 0 ( v i ) = P 0 ( v j ) and N + ( v i ) = N + ( v j ) and N - ( v i ) = N - ( v j )
The node that to be cluster divide to consolidated network meets upper strata and divides identical and its neighbor node set is identical.Fig. 4 has provided this layer by the arithmetic result example of network topology structure absolute structure cluster of equal value.Fig. 4 is the division on the basis of Fig. 2, and than the more refinement of the cluster of Fig. 2 and Fig. 3.Fig. 4 is divided into six clusters, by node shape and filling sign.Its attribute of the node of same cluster and neighbor node set are all identical.
(4) fuzzy clustering algorithm
In some cases, above-mentioned third and fourth layer is too strict according to the clustering method of the relative normal equivalent of network structure and absolute structure equivalence, causes too refinement of cluster result, as is divided into tens clusters, is difficult to for user's intuitivism apprehension and for visual analysis.Our method has been introduced fuzzy algorithm simultaneously, the clusters number that needs division is specified in permission when third and fourth layer of clustering, according to the optimized mode of the inner degree of approximation of cluster, realize clustering, user can, by mutual, control cluster granularity and check deeper related information like this.
Fuzzy algorithm is divided three step designs:
The first step, by each node v ineighbor node set expression be neighbour vector R (v i)={ c i1..., c it, c 1i..., c ti.The corresponding v of vector first half igo out to neighbours, the corresponding v of latter half ienter to neighbours, total length 2t.For the 3rd layer of clustering, t represents the span counting of current selected nodal community.For example, select sex attribute, this is counted as 2 (masculinity and femininities).C i1expression is from v iset out and connect the number that is related to of node that other attribute value is first property value (t property value altogether).For the 4th layer of clustering, t represents the interstitial content in network, c i1expression is from v inode sets out and connects v 1 node be related to number.
Second step, for all network nodes that need division, according to the vector representation R (v of its neighborhood i) calculate between two the node degree of approximation between node apart from d (v i, v j).This method acquiescence adopts vectorial Euclidean distance, and following formula also supports to adopt other vector distance or similarity calculating method (as the cosine distance of calculating by vectorial cosine similarity).
d(v i,v j)＝{∑ s＝1,2,…2t[(R s(v i)-R s(v j)] 2} 0.5
The 3rd step, inputs expectation clusters number according to similarity distance and user between the node calculating, and application k-mean algorithm, is k cluster by node division.This step also can adopt other clustering method, as spectral clustering.
2. algorithm specific implementation
In order to realize above-mentioned onion diagram hierarchical algorithm, we have introduced the final clustering that a new method is calculated heterogeneous network node fast.The advantage of this implementation method is to have realized the unitized cluster of five-layer structure.The core thinking of the method is the expansion row vector of design node, is called the unified row vector of node.As shown in Figure 5, this row vector relates to three parts, the neighbor information of the identifier of node, node attribute information and value, node and neighbours' attribute value.At different cluster levels, this unifies row vector and comprises different information: second and third, four layers comprise node attribute information and value; The 3rd layer of neighbor node aggregate attribute value that additionally comprises node; The 4th layer of neighbor node aggregate information that additionally comprises node; Layer 5 only comprises node label symbol, for network being refined as to single node granularity.Each node in heterogeneous network is according to the unified row vector of current cluster level computing node, and the node division that this row vector is identical is same cluster.
Concrete final clustering adopts hash algorithm to realize, and the unified row vector to each node, utilizes certain hash function, calculates cryptographic hash, and the node division that cryptographic hash is identical is same cluster.Realization can adopt the HashMap data structure in Java language.
3. algorithm performance analysis
According to theory, calculate, the time complexity of above hierarchical cluster implementation algorithm (not comprising fuzzy algorithm) is O (m+dn).Wherein, n, m, d represents respectively the interstitial content of heterogeneous network data, limit (relation) number, the attribute number on node.Be that this algorithm increases and linear growth with network size computing time, be therefore highly suitable for the cluster analysis of ultra-large heterogeneous network.We are applied to (interstitial content several thousand is to up to a million) on four group network data sets by a prototype realization of this algorithm.Upper at a main flow desktop desktop computer (four core I ntelCPU, 3.3GHz dominant frequency, 6GB internal memory), the computing time that test obtains above-mentioned algorithm is as table 1～3.Wherein, table 1 has provided the computing time that the second layer needs by the analysis of network node hierarchical cluster attribute.From result, can find out, the growth of computing time is slightly larger than linear growth, and contrast number of network node object increases.And, on 1,000,000 node data collection, only need 10 seconds left and right computing times.Table 2 has provided respectively third and fourth layer and has calculated by the relative normal equivalent algorithm of network topology structure and absolute structure equivalence algorithm the time that cluster analysis needs with table 3.Result can find out, the growth of computing time still maintains the trend that is slightly larger than linear growth.And on 1,000,000 node data collection, required time is slightly longer than second layer cluster analysis, but still is no more than 30 seconds.This time scale is effective in the real-time analysis of large scale scale heterogeneous network.
For adoptable fuzzy algorithm in third and fourth strata alanysis, the known its time complexity of theoretical analysis is O (k*n*d*l), wherein, n, d, k, l represents that respectively the interstitial content of heterogeneous network data, the attribute number on node, cluster number and calculating need the number of times of iteration.Known according to the result of table 2 and table 3, adopt fuzzy algorithm (Fuzzy), when the 3rd strata alanysis, still can support 1,000,000 magnitude interstitial contents, be no more than 60 seconds computing time.For the 4th strata alanysis, computation complexity increases very fast, and usage range reaches the heterogeneous network of 100,000 magnitude limit (relation) numbers.
4. case analysis
We illustrate the concrete application of onion diagram here.
The data that present case adopts are data of extracting from ArnetMiner database.It has comprised 9 large visual meetings and periodical whole papers of nearly 30 years.Data set is according to paper information storage, and every piece of paper comprises: thesis topic, author, meeting-place, deliver the time, quote situation, the attribute such as keyword, summary.The heterogeneous network that we set up based on these data mainly comprises three kinds of important node types: 11049 authors, 9557 pieces of papers and 9 meetings; Amount to 106316 annexations.About the basic onion diagram of these data, show and see Fig. 6.Use and data analysis that the visual analyzing instrument that we have invited several senior visual research personnel to apply us is correlated with, all found the indiscoverable linked character of some past methods.Here certain researcher's the analytic process of take is example.First, the initial network data display view that our visual analysis method provides is as Fig. 7.Whole network is divided into three clusters according to node ground type attribute, is respectively author's cluster, paper cluster and meeting/periodical cluster.He is by selecting meeting/periodical cluster, and this subnetwork is gone forward one by one to the second layer, according to meeting/journal title, divides, and further obtains Fig. 8.He has found a feature in data at once: meeting/periodical that paper number is maximum is CG & A and CGF, and the paper of these two meeting/periodicals has all surpassed 2000 pieces (seeing the label numeral in annexation).By selecting another cluster node, paper cluster node, and divide at the second layer according to paper citation times, can further obtain Fig. 9.By switching the mapping relations of label numeral in annexation, be mapped as average strength of joint, he continues to find: although the paper of delivering at CG & A and CGF meeting/periodical is a lot, wherein the low ratio of quoting paper is very large.To recently saying, the middle highly cited papers ratio of SciVis and TVCG meeting/periodical is relative higher.Further, he has selected another cluster node, author's cluster, and divide according to the attribute of the 3rd layer of neighbor node (by the paper cluster node of quoting situation division), as Figure 10.By filtering out to publish thesis, be less than the author of 10 pieces, he finds that all high yield authors (publishing thesis at least 10 pieces) can be divided into four groups of clusters: in first group of cluster, have 309 authors, they delivered a lot of in, the paper of low citation rate, on average only have the one piece high article of quoting.In second group of cluster, comprised 75 authors, they have delivered more article (doubling first group of author), but in the low article of quoting situation still account for the overwhelming majority.In the 3rd group of cluster, comprise 7 authors, be the senior fellow in visual field.They average everyone delivered the 41 pieces low papers of quoting, quote paper and 5 pieces of highly cited papers in 48 pieces, deliver highly cited papers number far away higher than front two groups of clusters.Last group cluster is the most interesting, and it has comprised three authors, and they have delivered 223 pieces of papers altogether, but wherein only have 1 piece of paper of being quoted by height.For showing more intuitively the distribution of neighbor node attribute in the different clusters that the 3rd strata alanysis obtains, this method also supports to adopt statistical graph to show neighbor node property distribution.Figure 11 for example, author's cluster node adopt histograms show the situation that is cited corresponding to different authors cluster distribute.
By adopting onion diagram hierarchical cluster analysis method, user can realize the academic network of above-mentioned large-scale isomery go forward one by one, iteration, mutual visual analysis, find fast special Clustering and the feature thereof of indiscoverable macroscopical incidence relation and microcosmic in the past.
Table 1. network node attribute clustering algorithm (the second hierarchical clustering) performance index
The relative normal equivalent clustering algorithm of table 2. network topology structure (tri-layer cluster) performance index
Table 3. network topology structure absolute structure clustering algorithm (the 4th hierarchical clustering) performance index of equal value
Claims (7)
1.一种异构网络可交互可视化方法，其步骤为：1. A heterogeneous network interactive visualization method, the steps of which are: 1)对待处理异构网络中的节点按照选取或设定的节点属性的取值进行聚类，并生成相应的聚类网络可视化图；1) The nodes in the heterogeneous network to be processed are clustered according to the value of the selected or set node attributes, and a corresponding clustering network visualization diagram is generated; 2)选取步骤1)的若干聚类结果，并对所选聚类结果中的每一节点，根据该异构网络的拓扑结构计算该节点的邻居节点集合；然后按照选取或设定的邻居节点集合节点属性值对所选聚类结果中的每一节点进行聚类；得到的聚类结果中，位于同一聚类中的节点具有节点属性值相同且在异构网络中的拓扑位置相似；然后将此次聚类结果生成聚类可视化图并将其作为步骤1)聚类网络可视化图的下一级可视图；2) Select some clustering results of step 1), and for each node in the selected clustering results, calculate the neighbor node set of the node according to the topology of the heterogeneous network; then according to the selected or set neighbor nodes Collect node attribute values to cluster each node in the selected clustering results; in the obtained clustering results, the nodes in the same cluster have the same node attribute values and similar topological positions in the heterogeneous network; then The clustering result is used to generate a clustering visualization diagram and use it as the next-level visualization diagram of the step 1) clustering network visualization diagram; 3)选取步骤2)的若干聚类结果，并对所选聚类结果中的每一节点，根据其邻居节点集合进行聚类，将具有相同邻居节点集合的节点划分到同一聚类中；得到的聚类结果中，位于同一聚类中的节点具有节点属性值相同且在异构网络中的拓扑位置相同；然后将此次聚类结果生成聚类可视化图并将其作为步骤2)所生成聚类网络可视化图的下一级可视图。3) Select some clustering results in step 2), and cluster each node in the selected clustering results according to its neighbor node set, and divide the nodes with the same neighbor node set into the same cluster; get In the clustering results of , nodes in the same cluster have the same node attribute value and the same topological position in the heterogeneous network; The next level of visualization for the clustering network visualization. 2.如权利要求1所述的方法，其特征在于所述异构网络为有向异构网络G＝(V,E)；其中，V＝{v1,...,vn}表示网络节点集合，E＝{e1,...,em}表示网络边集合；异构网络的邻接矩阵为W，其中的元素wij代表连接节点vi到节点vj的边；对于每个节点vi，N+(vi)＝{vj|wij＝1}代表节点vi的出向邻居节点集合，N-(vi)＝{vj|wji＝1}为节点vi的入向邻居节点集合。2. The method according to claim 1, wherein the heterogeneous network is a directed heterogeneous network G=(V,E); wherein, V={v 1 ,...,v n } means that the network The node set, E={e 1 ,...,e m } represents the network edge set; the adjacency matrix of the heterogeneous network is W, where the element w ij represents the edge connecting node v i to node v j ; for each Node v i , N + (v i )={v j |w ij ＝1} represents the set of outgoing neighbor nodes of node v i , N - (v i )={v j |w ji ＝1} is node v i The set of incoming neighbor nodes. 3.如权利要求1或2所述的方法，其特征在于所述步骤2)或步骤3)中的聚类方法为：3. The method according to claim 1 or 2, characterized in that the clustering method in said step 2) or step 3) is: 31)将节点vi的邻居节点集合表示为一个邻居向量R(vi)＝{ci1,...,cit,c1i,...,cti}；其中，cit为节点vi的第t个出向邻居节点，cti为节点vi的第t个入向邻居节点；31) Express the set of neighbor nodes of node v i as a neighbor vector R(v i )={c i1 ,...,c it ,c 1i ,...,c ti }; where c it is node v The tth outgoing neighbor node of i , c ti is the tth incoming neighbor node of node v i ; 32)对于所有需要划分的网络节点，按照其邻居节点集合的向量计算两两节点之间的节点近似度距离；32) For all network nodes that need to be divided, calculate the node proximity distance between any two nodes according to the vector of its neighbor node set; 33)根据计算出的节点间相似度距离及期望聚类数目，将节点划分为k个聚类。33) According to the calculated similarity distance between nodes and the expected number of clusters, divide the nodes into k clusters. 4.如权利要求1所述的方法，其特征在于将每一节点采用一统一行向量表示；所述统一行向量包括：节点属性信息及取值、节点的邻居信息及邻居属性取值；每次聚类时计算节点的统一行向量的哈希值，将哈希值相同的节点划分为同一聚类；其中，进行步骤1)的聚类时，节点的统一行向量包含节点属性信息及取值，其他取值为空；进行步骤2)的聚类时，节点的统一行向量包含节点属性信息及取值、节点的邻居属性取值；步骤3)的聚类时，节点的统一行向量包含节点属性信息及取值、节点的邻居信息。4. The method according to claim 1, wherein each node is represented by a unified row vector; the unified row vector includes: node attribute information and values, neighbor information and neighbor attribute values of nodes; Calculate the hash value of the unified row vector of the node during secondary clustering, and divide the nodes with the same hash value into the same cluster; wherein, when performing the clustering in step 1), the unified row vector of the node contains node attribute information and value, other values are empty; when clustering in step 2), the unified row vector of the node contains node attribute information and values, and the value of the neighbor attribute of the node; when clustering in step 3), the unified row vector of the node Contains node attribute information and values, and node neighbor information. 5.如权利要求1所述的方法，其特征在于将待处理的该异构网络生成可视化图作为第一级可视图，步骤1)所生成的聚类网络可视化图为该第一级可视图的下一级可视图；将步骤3)每一聚类结果中的节点生成一可视化图并将其作为步骤3)所生成聚类网络可视化图的下一级可视图，即最小粒度可视图。5. The method according to claim 1, characterized in that the heterogeneous network to be processed generates a visualized graph as a first-level visualized graph, and the clustering network visualized graph generated by step 1) is the first-level visualized graph The next-level visualization of the node in each clustering result in step 3) is used to generate a visualization diagram and used as the next-level visualization of the clustering network visualization generated in step 3), that is, the minimum granularity visualization. 6.如权利要求1所述的方法，其特征在于根据选取或设定的进行聚类的属性值不同，在同一级可视化图内包含一种或多种粒度的下一级可视化图。6. The method according to claim 1, characterized in that according to the selected or set attribute values for clustering, one or more next-level visualization diagrams of granularity are included in the same-level visualization diagram. 7.如权利要求1所述的方法，其特征在于在每一聚类之间的连接关系上标记该聚类间关系所包含的原始异构网络节点间关系的总数作为该聚类间连接关系的连接强度。7. The method according to claim 1, characterized in that the total number of original heterogeneous network node relationships included in the inter-cluster relationship is marked on the connection relationship between each cluster as the inter-cluster connection relationship connection strength.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410327034.1A CN104090957A (en) | 2014-03-10 | 2014-07-10 | Heterogeneous network interactive visualization method |
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410085487.8 | 2014-03-10 | ||
| CN201410085487 | 2014-03-10 | ||
| CN201410327034.1A CN104090957A (en) | 2014-03-10 | 2014-07-10 | Heterogeneous network interactive visualization method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104090957A true CN104090957A (en) | 2014-10-08 |
Family
ID=51638673
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410327034.1A Pending CN104090957A (en) | 2014-03-10 | 2014-07-10 | Heterogeneous network interactive visualization method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104090957A (en) |
Cited By (23)
* Cited by examiner, † Cited by third party| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104408153A (en) * | 2014-12-03 | 2015-03-11 | 中国科学院自动化研究所 | Short text hash learning method based on multi-granularity topic models |
| CN105208076A (en) * | 2015-08-13 | 2015-12-30 | 清华大学 | Multi-target service composition method based on correlation |
| CN105260611A (en) * | 2015-10-26 | 2016-01-20 | 浙江工业大学 | Map layout method based on node attribute transfer functions |
| CN106126523A (en) * | 2016-06-12 | 2016-11-16 | 中国科学院软件研究所 | A kind of counterfeit money Crime Information analyzes system and the method for analysis |
| CN106203516A (en) * | 2016-07-13 | 2016-12-07 | 中南大学 | A kind of subspace clustering visual analysis method based on dimension dependency |
| CN106571947A (en) * | 2015-11-16 | 2017-04-19 | 中国人民解放军理工大学 | Network element model implementation method supporting complex multi-network construction |
| CN106934422A (en) * | 2017-03-16 | 2017-07-07 | 浙江工业大学 | Hierarchical visual abstraction method based on improved force guide diagram layout |
| CN108156011A (en) * | 2016-12-02 | 2018-06-12 | 上海掌门科技有限公司 | A kind of method and apparatus for carrying out wireless access point cluster |
| CN108363797A (en) * | 2018-01-04 | 2018-08-03 | 北京工商大学 | A kind of associated diagram visual analysis method and its system based on transformation |
| CN108600022A (en) * | 2018-04-28 | 2018-09-28 | 中国人民解放军国防科技大学 | Dynamic network layout accelerating method |
| CN108595659A (en) * | 2018-04-28 | 2018-09-28 | 中国人民解放军国防科技大学 | A method for network multi-granularity organization |
| CN109426458A (en) * | 2017-09-04 | 2019-03-05 | 阿里巴巴集团控股有限公司 | A kind of Method of printing and device of relational graph |
| CN109859204A (en) * | 2019-02-22 | 2019-06-07 | 厦门美图之家科技有限公司 | Convolutional neural networks Model Checking and device |
| CN110032603A (en) * | 2019-01-22 | 2019-07-19 | 阿里巴巴集团控股有限公司 | The method and device that node in a kind of pair of relational network figure is clustered |
| CN110059227A (en) * | 2019-01-22 | 2019-07-26 | 阿里巴巴集团控股有限公司 | A kind of method and device determining the network structure between multiple samples |
| CN110225006A (en) * | 2019-05-27 | 2019-09-10 | 国家计算机网络与信息安全管理中心 | Network security data method for visualizing, controller and medium |
| CN111126510A (en) * | 2020-01-02 | 2020-05-08 | 深圳计算科学研究院 | Method for calculating similarity in heterogeneous network and related components thereof |
| CN111309917A (en) * | 2020-03-11 | 2020-06-19 | 上海交通大学 | Method and system for visualization of ultra-large-scale academic network based on galaxy map of conference journals |
| CN112183179A (en) * | 2019-07-01 | 2021-01-05 | 是德科技股份有限公司 | Method of analyzing a plurality of EDSs and computer readable medium |
| CN112953825A (en) * | 2021-01-26 | 2021-06-11 | 中山大学 | Attribute heterogeneous network embedding method, device, equipment and medium |
| CN115061760A (en) * | 2022-05-30 | 2022-09-16 | 四川大学 | State perception element visualization method oriented to analysis process |
| CN115241976A (en) * | 2022-07-27 | 2022-10-25 | 国网江苏省电力有限公司电力科学研究院 | A kind of distribution network monitoring data visualization method, computer equipment and storage medium |
| CN117994007A (en) * | 2024-04-03 | 2024-05-07 | 山东科技大学 | Social recommendation method based on multi-view fusion heterogeneous graph neural network |
Citations (2)
* Cited by examiner, † Cited by third party| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1443396A2 (en) * | 2003-01-30 | 2004-08-04 | Agilent Technologies, Inc. | Systems and methods for providing visualizations of network diagrams |
| CN102281154A (en) * | 2011-07-12 | 2011-12-14 | 广东宜通世纪科技股份有限公司 | Display method and system of network topology graphing |
-
2014
- 2014-07-10 CN CN201410327034.1A patent/CN104090957A/en active Pending
Patent Citations (2)
* Cited by examiner, † Cited by third party| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1443396A2 (en) * | 2003-01-30 | 2004-08-04 | Agilent Technologies, Inc. | Systems and methods for providing visualizations of network diagrams |
| CN102281154A (en) * | 2011-07-12 | 2011-12-14 | 广东宜通世纪科技股份有限公司 | Display method and system of network topology graphing |
Non-Patent Citations (4)
* Cited by examiner, † Cited by third party| Title |
|---|
| A TANG 等: "Community Detection based on Structural and Attribute Similarities", 《ACHI》 * |
| LEISHI 等: "Hierarchical Focus+Context Heterogeneous Network Visualization", 《IEEE PACIFIC VISUALIZATION SYMPOSIUM》 * |
| 张健沛 等: "一种基于节点相似性的链接预测算法", 《中国科技论文》 * |
| 时磊 等: "基于变换的大图点边可视化综述", 《计算机辅助设计与图形学学报》 * |
Cited By (36)
* Cited by examiner, † Cited by third party| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104408153B (en) * | 2014-12-03 | 2018-07-31 | 中国科学院自动化研究所 | A kind of short text Hash learning method based on more granularity topic models |
| CN104408153A (en) * | 2014-12-03 | 2015-03-11 | 中国科学院自动化研究所 | Short text hash learning method based on multi-granularity topic models |
| CN105208076A (en) * | 2015-08-13 | 2015-12-30 | 清华大学 | Multi-target service composition method based on correlation |
| CN105208076B (en) * | 2015-08-13 | 2018-06-15 | 清华大学 | A kind of multiple target service combining method perceived based on correlation |
| CN105260611B (en) * | 2015-10-26 | 2018-11-13 | 浙江工业大学 | Figure layout method based on nodal community transmission function |
| CN105260611A (en) * | 2015-10-26 | 2016-01-20 | 浙江工业大学 | Map layout method based on node attribute transfer functions |
| CN106571947A (en) * | 2015-11-16 | 2017-04-19 | 中国人民解放军理工大学 | Network element model implementation method supporting complex multi-network construction |
| CN106571947B (en) * | 2015-11-16 | 2019-08-23 | 中国人民解放军理工大学 | A kind of network element model implementation method for supporting complicated pluralism net structure |
| CN106126523A (en) * | 2016-06-12 | 2016-11-16 | 中国科学院软件研究所 | A kind of counterfeit money Crime Information analyzes system and the method for analysis |
| CN106203516A (en) * | 2016-07-13 | 2016-12-07 | 中南大学 | A kind of subspace clustering visual analysis method based on dimension dependency |
| CN106203516B (en) * | 2016-07-13 | 2019-04-09 | 中南大学 | A Visual Analysis Method of Subspace Clustering Based on Dimension Correlation |
| CN108156011A (en) * | 2016-12-02 | 2018-06-12 | 上海掌门科技有限公司 | A kind of method and apparatus for carrying out wireless access point cluster |
| CN106934422B (en) * | 2017-03-16 | 2019-07-26 | 浙江工业大学 | Hierarchical visual abstraction method based on improved force guide diagram layout |
| CN106934422A (en) * | 2017-03-16 | 2017-07-07 | 浙江工业大学 | Hierarchical visual abstraction method based on improved force guide diagram layout |
| CN109426458A (en) * | 2017-09-04 | 2019-03-05 | 阿里巴巴集团控股有限公司 | A kind of Method of printing and device of relational graph |
| CN109426458B (en) * | 2017-09-04 | 2022-05-17 | 阿里巴巴集团控股有限公司 | Method and device for printing relation graph |
| CN108363797A (en) * | 2018-01-04 | 2018-08-03 | 北京工商大学 | A kind of associated diagram visual analysis method and its system based on transformation |
| CN108600022B (en) * | 2018-04-28 | 2022-01-04 | 中国人民解放军国防科技大学 | A Dynamic Network Layout Acceleration Method |
| CN108600022A (en) * | 2018-04-28 | 2018-09-28 | 中国人民解放军国防科技大学 | Dynamic network layout accelerating method |
| CN108595659A (en) * | 2018-04-28 | 2018-09-28 | 中国人民解放军国防科技大学 | A method for network multi-granularity organization |
| CN110032603A (en) * | 2019-01-22 | 2019-07-19 | 阿里巴巴集团控股有限公司 | The method and device that node in a kind of pair of relational network figure is clustered |
| CN110059227A (en) * | 2019-01-22 | 2019-07-26 | 阿里巴巴集团控股有限公司 | A kind of method and device determining the network structure between multiple samples |
| CN110059227B (en) * | 2019-01-22 | 2023-08-04 | 创新先进技术有限公司 | Method and device for determining network structure among multiple samples |
| CN109859204A (en) * | 2019-02-22 | 2019-06-07 | 厦门美图之家科技有限公司 | Convolutional neural networks Model Checking and device |
| CN110225006B (en) * | 2019-05-27 | 2022-01-04 | 国家计算机网络与信息安全管理中心 | Network security data visualization method, controller and medium |
| CN110225006A (en) * | 2019-05-27 | 2019-09-10 | 国家计算机网络与信息安全管理中心 | Network security data method for visualizing, controller and medium |
| CN112183179A (en) * | 2019-07-01 | 2021-01-05 | 是德科技股份有限公司 | Method of analyzing a plurality of EDSs and computer readable medium |
| WO2021134807A1 (en) * | 2020-01-02 | 2021-07-08 | 深圳计算科学研究院 | Method for calculating similarity in heterogeneous network and related component therefor |
| CN111126510A (en) * | 2020-01-02 | 2020-05-08 | 深圳计算科学研究院 | Method for calculating similarity in heterogeneous network and related components thereof |
| CN111309917A (en) * | 2020-03-11 | 2020-06-19 | 上海交通大学 | Method and system for visualization of ultra-large-scale academic network based on galaxy map of conference journals |
| CN112953825A (en) * | 2021-01-26 | 2021-06-11 | 中山大学 | Attribute heterogeneous network embedding method, device, equipment and medium |
| CN115061760A (en) * | 2022-05-30 | 2022-09-16 | 四川大学 | State perception element visualization method oriented to analysis process |
| CN115061760B (en) * | 2022-05-30 | 2024-09-10 | 四川大学 | State perception element visualization method oriented to analysis process |
| CN115241976A (en) * | 2022-07-27 | 2022-10-25 | 国网江苏省电力有限公司电力科学研究院 | A kind of distribution network monitoring data visualization method, computer equipment and storage medium |
| CN115241976B (en) * | 2022-07-27 | 2023-11-17 | 国网江苏省电力有限公司电力科学研究院 | A distribution network monitoring data visualization method, computer equipment and storage medium |
| CN117994007A (en) * | 2024-04-03 | 2024-05-07 | 山东科技大学 | Social recommendation method based on multi-view fusion heterogeneous graph neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104090957A (en) | 2014-10-08 | Heterogeneous network interactive visualization method |
| Beck et al. | 2014 | The State of the Art in Visualizing Dynamic Graphs. |
| Vehlow et al. | 2015 | The State of the Art in Visualizing Group Structures in Graphs. |
| Gou et al. | 2011 | Sfviz: interest-based friends exploration and recommendation in social networks |
| Zhou et al. | 2013 | Edge bundling in information visualization |
| Batty | 2006 | Hierarchy in cities and city systems |
| Bae et al. | 2016 | Combining microsimulation and agent-based model for micro-level population dynamics |
| CN105224656B (en) | 2018-06-15 | A kind of comparison association visual analysis methods and applications for being directed to two class hierarchy data |
| CN108255933A (en) | 2018-07-06 | A kind of social media dynamic event develops visual analysis method and system |
| KR101710606B1 (en) | 2017-03-08 | Apparatus and Method for Interactive Visualization for Analyzing Sets in Large Networks |
| Ullah et al. | 2022 | A novel relevance-based information interaction model for community detection in complex networks |
| Sridevi et al. | 2014 | An intelligent classifier for breast cancer diagnosis based on K-Means clustering and rough set |
| Molina-Solana et al. | 2017 | Improving data exploration in graphs with fuzzy logic and large-scale visualisation |
| Bi et al. | 2014 | A clustering method for evaluating the environmental performance based on slacks-based measure |
| Sathiyanarayanan et al. | 2016 | Spherule diagrams with graph for social network visualization |
| Wang et al. | 2021 | Hierarchical visualization of geographical areal data with spatial attribute association |
| Pop et al. | 2017 | A hybrid diploid genetic based algorithm for solving the generalized traveling salesman problem |
| Keyvanpour | 2013 | A survey on community detection methods based on the nature of social networks |
| Buono et al. | 2005 | Visualizing association rules in a framework for visual data mining |
| Gajawada et al. | 2012 | Projected clustering using particle swarm optimization |
| Maivizhi et al. | 2016 | A survey of tools for community detection and mining in social networks |
| Abdelaal et al. | 2017 | ColTop: Visual topic-based analysis of scientific community structure |
| Gupta et al. | 2011 | Statistical approach of social network in community mining |
| Maia et al. | 2013 | On the analysis of the collaboration network of the Brazilian symposium on computer networks and distributed systems: 30 Editions of history |
| Zolkepli et al. | 2014 | Visualizing fuzzy relationship in bibliographic big data using hybrid approach combining fuzzy c-means and Newman-Girvan algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 2014-10-08 | C06 | Publication | |
| 2014-10-08 | PB01 | Publication | |
| 2014-10-29 | C10 | Entry into substantive examination | |
| 2014-10-29 | SE01 | Entry into force of request for substantive examination | |
| 2018-01-09 | WD01 | Invention patent application deemed withdrawn after publication | |
| 2018-01-09 | WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20141008 |