CN111008196A - Depth-first search-based frequent pattern mining method - Google Patents

Depth-first search-based frequent pattern mining method Download PDF

Info

Publication number
CN111008196A
CN111008196A CN201911061496.2A CN201911061496A CN111008196A CN 111008196 A CN111008196 A CN 111008196A CN 201911061496 A CN201911061496 A CN 201911061496A CN 111008196 A CN111008196 A CN 111008196A
Authority
CN
China
Prior art keywords
node
frequent
sup
network
graph
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.)
Pending
Application number
CN201911061496.2A
Other languages
Chinese (zh)
Inventor
周月双
戴维迪
刘雪莉
王文俊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tianjin University
Original Assignee
Tianjin University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tianjin University filed Critical Tianjin University
Priority to CN201911061496.2A priority Critical patent/CN111008196A/en
Publication of CN111008196A publication Critical patent/CN111008196A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/22—Indexing; Data structures therefor; Storage structures
    • G06F16/2228—Indexing structures
    • G06F16/2237—Vectors, bitmaps or matrices
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2465—Query processing support for facilitating data mining operations in structured databases

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computational Linguistics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention belongs to the field of complex networks and data mining, and particularly relates to a frequent pattern mining method based on depth-first search in a complex network, which mainly comprises the following steps: and carrying out related concepts and definitions, coding a standardized graph and carrying out frequent pattern mining on the basis of the related concepts and definitions. On the basis of a complex network theory framework, a linear sequence consisting of standardized graph codes is obtained after a graph is traversed based on depth-first search, frequent pattern mining is carried out on the constructed labeled relational network, and while a frequent pattern set in the network is mined, the problem of subgraph isomorphism is effectively reduced on the method, and the generation of redundant candidate patterns is avoided; the linear sequence is mapped, so that the use of a memory is greatly saved.

Description

Depth-first search-based frequent pattern mining method
Technical Field
The invention belongs to the field of complex networks and data mining, relates to a depth-first search-based frequent pattern mining method in a tag-oriented heterogeneous network, and particularly relates to a depth-first search-based frequent pattern mining method in a complex network.
Background
Nowadays, with the advent of the big data era and the rapid development of network information technology, human society has moved forward into the complex network era. The complex network is not only a representation form of data, but also a means for scientific research. Scientific problems built based on complex networks are more and more diversified, and good possibility is provided for interdisciplinary discipline. Complex networks are often powerful representations of complex systems, such as social networks, biological networks, literature networks, and the like. Tagged heterogeneous networks are commonly used to study and analyze these data.
Due to the advantages of the network and the topological structure of the graph, the graph mining method has wide applicability, and related research aiming at graph mining also draws more and more attention of researchers. Moreover, graph mining has great potential in many practical applications, including semantic networking, behavioral modeling, biological network analysis, chemical compound classification, and the like. At present, many efficient algorithms have been proposed to mine subgraphs or patterns in graph data, and the effect of these algorithms is very different due to the difference of the applied objects and the problem background. The purpose of frequent pattern mining is to find a set of patterns that appear frequently in an atlas, providing important help for the next research and analysis work. The current frequent subgraph mining algorithm can be roughly divided into two categories: one is breadth first algorithm, which includes AGM and FSG, etc.; another class is depth-first algorithms, which include gSpan and FFSM, among others. Most conventional algorithms have one of the most significant problems in time complexity, which is the sub-graph isomorphism problem.
Therefore, considering the fact that real complex networks have diversified structures and how to save time and space costs, and the fact that labeled nodes in heterogeneous networks can not only help to improve the result of pattern mining but also assist in analyzing the underlying regularity of mined patterns. Mining and analysis of frequent patterns in complex networks is a very worthy problem to study. The method is characterized in that a graph is traversed based on a depth-first search method, a researched graph data object is constructed into a linear sequence formed by standardized graph codes, the support degree and the linear sequence of the graph are defined, and all frequent patterns are mined according to the rightmost expansion and the optimal matching principle, so that the mining result has wide applicability and significance.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a depth-first search-based frequent pattern mining method for a labeled heterogeneous network.
Aiming at the difficulties of exploring a frequent pattern structure in a complex network and how to save time and space cost in the process of matching the pattern and the problem of modeling by simultaneously linking the attribute of a node and the topological structure of the network, the invention provides a frequent pattern mining method based on depth-first search, which is used for mining frequent patterns in a real network and finding out potential rules of the frequent patterns. The method has wide application value in event detection and prediction and event clustering, and can help solve the actual problem in a real scene.
In order to solve the technical problems in the background technology, the invention adopts the technical scheme that: a frequent pattern mining method based on depth-first search comprises the following steps:
s1: related concepts and definitions involved in the method:
(1) a tagged network: a tagged network is considered to be a five-tuple, G ═ V, E, Σ V, Σ E, L. Wherein V represents a set of nodes in the network;
Figure BDA0002258081020000021
the sets of edges in the network, Σ V and Σ E, represent sets of labels for nodes and edges, respectively, and L is a label function, V ∪ E → L, which functions to perform the mapping of labels to nodes and edges, and thus, there are V → Σ V, E → Σ E.
(2) And (3) isomorphism of subgraphs: the isomorphism of the graph is a bijection f:
Figure BDA0002258081020000022
if the graph G ═ { V, E, Σ V, Σ E, L } and the graph G '} { V', E ', Σ V', Σ E ', L' } are isomorphic, the following condition is satisfied:
①
Figure BDA0002258081020000023
LG(u)＝LG'(f(u))，
②
Figure BDA0002258081020000024
③
Figure BDA0002258081020000025
L(u,v)＝LG'(f(u),f(v))。
subgraph isomorphism is the NP-complete problem. How to reduce or reduce the computational overhead of sub-graph isomorphism is one of the main research contents of graph mining work.
(3) Frequent mode: given a set of networks GD, the set of patterns mined from GD is PD, { P }iI-0, 1, …, n, and given a minimum support threshold of min _ sup, we call the pattern set PD frequent if and only if each element P in the set is PiIs not less than a minimum support threshold, i.e. SUPPi≥min_sup。
(4) Support of the mode: given a set of networks GD, the set of patterns mined from GD is PD, { P }iI | ═ 0,1, …, n }, pattern Pi(i is more than or equal to 0 and less than or equal to n) is expressed as SUPPiThe calculation method is based on P obtained in the first stepiWith P, the designated pivot (i.e., the starting point for mining) in (1)iThe obtained mapping of pivot in all the matches, namely the pattern PiThe number of occurrences of the middle node pivot is recorded as the support SUP of the modePi。
S2: encoding a normalization graph:
(1) as can be seen from (1) in S1, there are 5 basic elements in each edge in the tagged network, and we directly use this form of quintuple to encode the edge, and the edge e is (v ═ v)i,vj) Is represented by (v)i,vj,li,le,lj) Wherein v isi，vjIs a unique identification of a node,/i，ljAre node numbers, leAre edge-marked. For example, in the example of fig. 1, the edge e ═ v1,v2) Can be represented as (1,2, A, a, B). Each edge in the network can be represented by a code form, so that the whole network can be equivalent to a linear sequence formed by a series of graph codes, namely a normalized graph code sequence G _ code sequence { e } of the networkk|k＝1,2,…,n}。
(2) Rule of coding order: certain rules exist for determining the sequence of the node identifiers. If during the DFS traversal of the graph, the node traversed first is viThe node traversed after is vjThen viAnd vjHas a coding order of vi＜vj. Similarly, the edge labels are determined to have a certain rule according to the node identifiers, and then the linear relation between the edges is defined. For example, can be represented by v0＜v1To determine (v)0,v1)＜(v1,v2) The edge numbering also complies with such a rule, e.g. by e1＝(v0,v1)＜e2＝(v1,v2) To determine le1＝a<le2＝b。
(3) Linear order of encoding: given the code of any two edges in a tagged network as e1＝(a1,a2,a3,a4,a5)，e2＝(b1,b2,b3,b4,b5) The linear order is determined by the following conditions:
①e1＝e2if and only if ai＝bi，i＝1,2,…,5；
②e1＜e2And if and only if
Figure BDA0002258081020000031
K is not less than 1 and not more than 5, so that aj＝bj(1≤j<k) And a is ak＜bk；
③e1＜e2And others.
S3: the process of the frequent pattern mining algorithm:
and performing frequent pattern mining on the basis of related concepts and definitions of S1 and S2. The starting of the algorithm Gfpm is to traverse the whole network set GD by adopting depth-first search, randomly select a starting vertex and mark the visited vertex, the visited vertex corresponds to a code G _ code, the code sequence is gradually expanded from few to many until the traversal is finished, and a complete linear sequence G _ code sequence { e } formed by standardized graph codes is established for the whole networkk|k＝1,2,…,n}。
And (4) taking the graph coding linear sequence G _ code sequence obtained by traversal as an input, and mining the frequent mode. Starting from the first edge of the coded sequence, gradually expanding downwards to perform pattern matching calculation, following the rightmost path expansion principle each time expansion is performed, and expanding once by an edge ekAnd a new node vi+1All need to recalculate and update the support SUP of the schemaPi. The rightmost path expansion principle is as follows: the node found last in a pattern is called the rightmost node, and the straight-line path from the first node to the rightmost node is called the rightmost path. Given graph G, a new edge e may be added between the rightmost node and another node on the rightmost path, in an extension referred to as backward extension; a new node may also be introduced and connected to the node on the rightmost path, in an extension referred to as forward extension. Since both extensions occur on the rightmost path, they are called rightmost extensions. Fig. 2 shows several possible scenarios for the rightmost path expansion.
Support degree SUP of each update modePiAnd comparing the support degree with a given support degree threshold value min _ SUP of the frequent pattern if SUPPiIf not less than min _ sup, the new mode is a frequent mode, and the set FPD of the frequent mode is put into the set FPD ═ PiI | ═ 0,1, …, m }, if SUP iPiAnd if not more than min _ sup, judging that the new mode is not the frequent mode, and pruning the updated node and edge.
When pruning is carried out on the infrequent mode, if the updated node vi+1If there is no neighbor node cluster in the follow-up, i.e. the node is the rightmost node, then the node v is selectedi+1And edge ekTo carry outPruning is only needed, if the node v is updatedi+1If the neighbor node cluster exists subsequently, the neighbor node cluster needs to be pruned integrally. Integral pruning of the neighboring node cluster according to the node vi+1The mapping information of the neighbor node cluster of the node in the graph coding sequence is determined by the position information of the node, and then all the information of the neighbor node cluster is pruned.
After pruning is finished, returning to the current node to continue rightmost expansion, and following the rightmost path expansion principle when expansion is carried out each time, namely returning to the fifth step, and circulating the processes of expansion-calculation matching-pruning until the whole network expansion is finished, so that a frequent pattern set FPD (P) can be obtainediI ═ 0,1, …, m }; outputting the final frequency mode set FPD ═ { P ═ PiAnd i is 0,1, …, m, and the mining process is finished.
Advantageous effects
The method provided by the invention is based on a complex network theory framework, a linear sequence formed by standardized graph codes is obtained after a graph is traversed based on depth-first search, the constructed labeled relational network is subjected to frequent pattern mining, and the problem of sub-graph isomorphism is effectively reduced on the method while a frequent pattern set in the network is mined out, so that the generation of redundant candidate patterns is avoided; the linear sequence is mapped, so that the use of a memory is greatly saved. The method can also be applied to various scenes and the field of mining, and is beneficial to the analysis and further research of practical problems.
Drawings
FIG. 1 is an exemplary illustration of normalized graph coding;
fig. 2 is an illustration of several cases of rightmost path expansion.
Detailed Description
The technical solutions of the present invention are further described in detail with reference to the accompanying drawings and specific embodiments, which are only illustrative of the present invention and are not intended to limit the present invention.
Aiming at researching the frequent pattern structure in the complex network, the invention provides a frequent pattern mining method based on depth-first search, which is used for mining the frequent pattern in the labeled relational network and finding out the potential law of the frequent pattern. The method can be mainly applied to the research of the mining result of the frequent pattern in some fields, and can be carried out according to the following steps when a frequent pattern mining algorithm is implemented:
the first step is as follows: firstly, carrying out data processing on a labeled relational network to be analyzed, and confirming that each node and each edge have a corresponding node label, a corresponding node label and a corresponding edge label;
the second step is that: traversing the whole network by adopting depth-first search, randomly selecting a starting vertex, marking the visited vertex, enabling the visited vertex to correspond to a code G _ code, and gradually expanding the code sequence from few to many until the traversal is finished;
the third step: and after the traversal is finished, establishing a complete linear sequence G _ code sequence { e } consisting of the normalized graph codes for the whole networkk|k＝1,2,…,n}；
The fourth step: taking the graph coding linear sequence G _ code sequence obtained by traversal as input, and mining a frequent mode;
the fifth step: starting from the first edge of the coded sequence, gradually expanding downwards to perform pattern matching calculation, following the rightmost path expansion principle each time the expansion is performed, and recalculating and updating the support SUP of the pattern each time the expansion is performedPi；
And a sixth step: support degree SUP of each update modePiAnd comparing the support degree with a given support degree threshold value min _ SUP of the frequent pattern if SUPPiIf not less than min _ sup, the new mode is a frequent mode, and the set FPD of the frequent mode is put into the set FPD ═ PiI | ═ 0,1, …, m }, if SUP iPiIf the node is not more than min _ sup, judging that the new mode is not the frequent mode, and pruning the updated node and edge;
the seventh step: pruning the infrequent mode, if the updated node has no neighbor node cluster in the follow-up process, namely the node is the rightmost node, pruning the node and the edge, and if the updated node has a neighbor node cluster in the follow-up process, performing integral pruning on the neighbor node cluster;
eighth step: integrally pruning a neighbor node cluster of the updated node, determining mapping information of the neighbor node cluster of the node in the graph coding sequence according to the position information of the node, and pruning all information of the neighbor node cluster;
the ninth step: after pruning is finished, returning to the current node to continue rightmost expansion, and following the rightmost path expansion principle when expansion is carried out each time, namely returning to the fifth step, and circulating the processes of expansion-calculation matching-pruning until the whole network expansion is finished, so that a frequent pattern set FPD (P) can be obtainedi|i＝0,1,…,m}；
The tenth step: outputting the final frequency mode set FPD ═ { P ═ PiI |, 0,1, …, m }, and the algorithm ends.

Claims (7)

1. The frequent pattern mining method based on depth-first search is characterized by comprising the following steps of:
s1: related concepts and definitions:
(1) a tagged network:
considering a labeled network as a five-tuple, G ═ V, E, Σ V, Σ E, L, where V represents the set of nodes in the network;
Figure FDA0002258081010000011
representing a set of edges in a network;
Σ V and Σ E denote sets of labels of nodes and edges, respectively;
l is the label function, V ∪ E → L, which performs the mapping of labels to nodes and edges,
therefore, there are: v → Σ V, E → Σ E;
(2) and (3) isomorphism of subgraphs: the isomorphism of the figure is a bijection
Figure FDA0002258081010000012
If the graph G ═ { V, E, Σ V, Σ E, L } and the graph G '} { V', E ', Σ V', Σ E ', L' } are isomorphic, the following condition is satisfied:
①
Figure FDA0002258081010000013
LG(u)＝LG'(f(u))，
②
Figure FDA0002258081010000014
③
Figure FDA0002258081010000015
L(u,v)＝LG'(f(u),f(v))；
(3) frequent mode: given a network-set GD, the set of patterns mined from the GD are PDs,
PD＝{Pii-0, 1, …, n, and given a minimum support threshold of min _ sup, we call the pattern set PD frequent if and only if each element P in the set is PiIs not less than a minimum support threshold, i.e. SUPPi≥min_sup；
(4) Support of the mode: given a network-set GD, the set of patterns mined from the GD are PDs,
PD＝{Pii | ═ 0,1, …, n }, pattern Pi(i is more than or equal to 0 and less than or equal to n) is expressed as SUPPiThe calculation method is based on P obtained in the first stepiWith P, the designated pivot (i.e., the starting point for mining) in (1)iThe obtained mapping of pivot in all the matches, namely the pattern PiThe number of occurrences of the middle node pivot is recorded as the support SUP of the modePi；
S2: encoding a normalization graph:
(1) as can be seen from (1) in S1, there are 5 basic elements in each edge in the tagged network, and the edge e ═ v is encoded directly by using the quintuple in this form to encode the edgei,vj) Is represented by (v)i,vj,li,le,lj) Wherein v isi，vjIs a unique identification of a node,/i，ljAre node numbers, leAre edge labels;
(2) rule of coding order: if during the DFS traversal of the graph, the node traversed first is viThe node traversed after is vjThen viAnd vjHas a coding order of vi＜vj；
Similarly, determining that the edge labels have a certain rule according to the node identifiers, and further defining the linear relationship between the edges;
(3) linear order of encoding: given the code of any two edges in a tagged network as e1＝(a1,a2,a3,a4,a5)，e2＝(b1,b2,b3,b4,b5) The linear order is determined by the following conditions:
①e1＝e2if and only if ai＝bi，i＝1,2,…,5；
②e1＜e2And if and only if
Figure FDA0002258081010000021
A is caused to bej＝bj(1≤j<k) And a is ak＜bk；
③e1＜e2And other cases;
s3: the process of the frequent pattern mining algorithm:
and performing frequent pattern mining on the basis of related concepts and definitions of S1 and S2.
2. The frequent pattern mining method based on depth-first search according to claim 1, wherein the step S3 is specifically: the starting of the algorithm Gfpm is to traverse the whole network set GD by adopting depth-first search, randomly select a starting vertex and mark the visited vertex, the visited vertex corresponds to a code G _ code, the code sequence is gradually expanded from few to many until the traversal is finished, and a complete linear sequence G _ code sequence { e } formed by standardized graph codes is established for the whole networkk|k＝1,2,…,n}。
3. The depth-first search-based frequent pattern mining method according to claim 2, wherein the frequent pattern mining is performed by using a graph coding linear sequence G _ code sequence obtained by traversal as an input; starting from the first edge of the coded sequence, gradually expanding downwards to perform pattern matching calculation, following the rightmost path expansion principle each time expansion is performed, and expanding once by an edge ekAnd a new node vi+1All need to recalculate and update the support SUP of the schemaPi。
4. The frequent pattern mining method based on depth-first search according to claim 2, wherein the rightmost path expansion principle is as follows: the node found finally in one mode is called the rightmost node, and the straight line path from the first node to the rightmost node is called the rightmost path; given graph G, a new edge e may be added between the rightmost node and another node on the rightmost path, in an extension referred to as backward extension; a new node may also be introduced and connected to the node on the rightmost path, in an extension referred to as forward extension.
5. The frequent pattern mining method based on depth-first search of claim 2, wherein SUP is the support degree of each time pattern is updatedPiAnd comparing the support degree with a given support degree threshold value min _ SUP of the frequent pattern if SUPPiIf not less than min _ sup, the new mode is a frequent mode, and the set FPD of the frequent mode is put into the set FPD ═ PiI | ═ 0,1, …, m }, if SUP iPiAnd if not more than min _ sup, judging that the new mode is not the frequent mode, and pruning the updated node and edge.
6. The frequent pattern mining method based on depth-first search of claim 2, wherein when pruning the infrequent pattern, if the updated node v is a nodei+1If there is no neighbor node cluster in the follow-up, i.e. the node is the rightmost node, then the node v is selectedi+1And edge ekPruning is carried out, if the node v is updatedi+1If a neighbor node cluster exists subsequently, the neighbor node cluster needs to be pruned integrally;
integral pruning of the neighboring node cluster according to the node vi+1The mapping information of the neighbor node cluster of the node in the graph coding sequence is determined by the position information of the node, and then all the information of the neighbor node cluster is pruned.
7. The frequent pattern mining method based on depth-first search as claimed in claim 6, wherein after pruning is finished, returning to the current node to continue rightmost expansion, and when expansion is performed each time, following the rightmost path expansion principle, returning to the fifth step, and looping the process of expansion-calculation matching-pruning until the whole network expansion is finished to obtain the frequent pattern set FPD ═ { P ═ PiI ═ 0,1, …, m }; outputting the final frequency mode set FPD ═ { P ═ PiAnd i is 0,1, …, m, and the mining process is finished.
CN201911061496.2A 2019-11-01 2019-11-01 Depth-first search-based frequent pattern mining method Pending CN111008196A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911061496.2A CN111008196A (en) 2019-11-01 2019-11-01 Depth-first search-based frequent pattern mining method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911061496.2A CN111008196A (en) 2019-11-01 2019-11-01 Depth-first search-based frequent pattern mining method

Publications (1)

Publication Number Publication Date
CN111008196A true CN111008196A (en) 2020-04-14

Family

ID=70111385

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911061496.2A Pending CN111008196A (en) 2019-11-01 2019-11-01 Depth-first search-based frequent pattern mining method

Country Status (1)

Country Link
CN (1) CN111008196A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113225199A (en) * 2020-11-17 2021-08-06 中国人民解放军国防科技大学 Interactive behavior prediction method and device based on time sequence network mining and electronic equipment
US20220327514A1 (en) * 2018-11-30 2022-10-13 Visa International Service Association System, method, and computer program product for generating embeddings for objects
CN112287118B (en) * 2020-10-30 2023-06-02 西南电子技术研究所（中国电子科技集团公司第十研究所） Event mode frequent subgraph mining and prediction method

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220327514A1 (en) * 2018-11-30 2022-10-13 Visa International Service Association System, method, and computer program product for generating embeddings for objects
CN112287118B (en) * 2020-10-30 2023-06-02 西南电子技术研究所（中国电子科技集团公司第十研究所） Event mode frequent subgraph mining and prediction method
CN113225199A (en) * 2020-11-17 2021-08-06 中国人民解放军国防科技大学 Interactive behavior prediction method and device based on time sequence network mining and electronic equipment

Similar Documents

Publication Publication Date Title
CN106909643B (en) Knowledge graph-based social media big data topic discovery method
WO2023071845A1 (en) Knowledge graph processing
CN111008196A (en) Depth-first search-based frequent pattern mining method
CN104992078B (en) A kind of protein network complex recognizing method based on semantic density
Przybyła-Kasperek et al. A dispersed decision-making system–The use of negotiations during the dynamic generation of a system’s structure
CN113902034A (en) Vector road data change information identification and extraction method and device
CN112905690A (en) Financial time sequence data mining method and system based on hypergraph
CN108573020B (en) Three-dimensional assembly model retrieval method integrating assembly information
Gouda et al. An improved global lower bound for graph edit similarity search
Fan et al. Cadtransformer: Panoptic symbol spotting transformer for cad drawings
CN103077216A (en) Sub-graph matching device and sub-graph matching method
CN108182230B (en) Bit vector quadtree-based mobile object convergence pattern mining method
CN113779085A (en) Method and device for acquiring isomorphic subgraph, computer equipment and readable storage medium
CN112597399A (en) Graph data processing method and device, computer equipment and storage medium
CN115062619B (en) Chinese entity linking method, device, equipment and storage medium
CN110807061A (en) Method for searching frequent subgraphs of uncertain graphs based on layering
CN117010373A (en) Recommendation method for category and group to which asset management data of power equipment belong
KR101985961B1 (en) Similarity Quantification System of National Research and Development Program and Searching Cooperative Program using same
CN117150664A (en) CAD model analysis method, device, equipment and storage medium
Acosta-Mendoza et al. A new algorithm for approximate pattern mining in multi-graph collections
CN115794884A (en) Method and device for pre-computing subgraph query based on graph abstract technology
Wang et al. A novel subgraph querying method on directed weighted graphs
Hellal et al. Nodar: mining globally distributed substructures from a single labeled graph
Fattore et al. Measuring structural dissimilarity between finite partial orders
CN107886107B (en) Big data fusion method, system and device

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
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20200414

WD01 Invention patent application deemed withdrawn after publication