Download presentation
Presentation is loading. Please wait.
Published byFranklin Green Modified over 8 years ago
1
CMU SCS Talk 3: Graph Mining Tools – Tensors, communities, parallelism Christos Faloutsos CMU
2
CMU SCS (C) 2011, C. Faloutsos 2 Overall Outline Introduction – Motivation Talk#1: Patterns in graphs; generators Talk#2: Tools (Ranking, proximity) Talk#3: Tools (Tensors, scalability) Conclusions KAIST-2011
3
CMU SCS Outline Task 4: time-evolving graphs – tensors Task 5: community detection Task 6: virus propagation Task 7: scalability, parallelism and hadoop Conclusions KAIST-2011(C) 2011, C. Faloutsos 3
4
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 4 Thanks to Tamara Kolda (Sandia) for the foils on tensor definitions, and on TOPHITS
5
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 5 Detailed outline Motivation Definitions: PARAFAC and Tucker Case study: web mining
6
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 6 Examples of Matrices: Authors and terms dataminingclassif.tree... John Peter Mary Nick...
7
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 7 Motivation: Why tensors? Q: what is a tensor?
8
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 8 Motivation: Why tensors? A: N-D generalization of matrix: dataminingclassif.tree... John Peter Mary Nick... KDD’09
9
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 9 Motivation: Why tensors? A: N-D generalization of matrix: dataminingclassif.tree... John Peter Mary Nick... KDD’08 KDD’07 KDD’09
10
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 10 Tensors are useful for 3 or more modes Terminology: ‘mode’ (or ‘aspect’): dataminingclassif.tree... Mode (== aspect) #1 Mode#2 Mode#3
11
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 11 Notice 3 rd mode does not need to be time we can have more than 3 modes... IP destination Dest. port IP source 80 125
12
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 12 Notice 3 rd mode does not need to be time we can have more than 3 modes –Eg, fFMRI: x,y,z, time, person-id, task-id http://denlab.temple.edu/bidms/cgi-bin/browse.cgi From DENLAB, Temple U. (Prof. V. Megalooikonomou +)
13
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 13 Motivating Applications Why tensors are useful? –web mining (TOPHITS) –environmental sensors –Intrusion detection (src, dst, time, dest-port) –Social networks (src, dst, time, type-of-contact) –face recognition –etc …
14
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 14 Detailed outline Motivation Definitions: PARAFAC and Tucker Case study: web mining
15
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 15 Tensor basics Multi-mode extensions of SVD – recall that:
16
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 16 Reminder: SVD –Best rank-k approximation in L2 A m n m n U VTVT
17
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 17 Reminder: SVD –Best rank-k approximation in L2 A m n + 1u1v11u1v1 2u2v22u2v2
18
KAIST-2011(C) 2011, C. Faloutsos 18 Goal: extension to >=3 modes ~ I x R K x R A B J x R C R x R x R I x J x K +…+=
19
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 19 Main points: 2 major types of tensor decompositions: PARAFAC and Tucker both can be solved with ``alternating least squares’’ (ALS)
20
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 20 = U I x R V J x R W K x R R x R x R Specially Structured Tensors Tucker TensorKruskal Tensor I x J x K = U I x R V J x S W K x T R x S x T I x J x K Our Notation +…+= u1u1 uRuR v1v1 w1w1 vRvR wRwR “core”
21
KAIST-2011(C) 2011, C. Faloutsos 21 Tucker Decomposition - intuition I x J x K ~ A I x R B J x S C K x T R x S x T author x keyword x conference A: author x author-group B: keyword x keyword-group C: conf. x conf-group G : how groups relate to each other
22
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 22 Intuition behind core tensor 2-d case: co-clustering [Dhillon et al. Information-Theoretic Co- clustering, KDD’03]
23
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 23 m m n nl k k l eg, terms x documents
24
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 24 term x term-group doc x doc group term group x doc. group med. terms cs terms common terms med. doc cs doc
25
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 25 Tensor tools - summary Two main tools –PARAFAC –Tucker Both find row-, column-, tube-groups –but in PARAFAC the three groups are identical ( To solve: Alternating Least Squares )
26
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 26 Detailed outline Motivation Definitions: PARAFAC and Tucker Case study: web mining
27
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 27 Web graph mining How to order the importance of web pages? –Kleinberg’s algorithm HITS –PageRank –Tensor extension on HITS (TOPHITS)
28
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 28 Kleinberg’s Hubs and Authorities (the HITS method) Sparse adjacency matrix and its SVD: authority scores for 1 st topic hub scores for 1 st topic hub scores for 2 nd topic authority scores for 2 nd topic from to Kleinberg, JACM, 1999
29
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 29 authority scores for 1 st topic hub scores for 1 st topic hub scores for 2 nd topic authority scores for 2 nd topic from to HITS Authorities on Sample Data We started our crawl from http://www-neos.mcs.anl.gov/neos, and crawled 4700 pages, resulting in 560 cross-linked hosts.
30
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 30 Three-Dimensional View of the Web Observe that this tensor is very sparse! Kolda, Bader, Kenny, ICDM05
31
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 31 Three-Dimensional View of the Web Observe that this tensor is very sparse! Kolda, Bader, Kenny, ICDM05
32
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 32 Three-Dimensional View of the Web Observe that this tensor is very sparse! Kolda, Bader, Kenny, ICDM05
33
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 33 Topical HITS (TOPHITS) Main Idea: Extend the idea behind the HITS model to incorporate term (i.e., topical) information. authority scores for 1 st topic hub scores for 1 st topic hub scores for 2 nd topic authority scores for 2 nd topic from to term term scores for 1 st topic term scores for 2 nd topic
34
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 34 Topical HITS (TOPHITS) Main Idea: Extend the idea behind the HITS model to incorporate term (i.e., topical) information. authority scores for 1 st topic hub scores for 1 st topic hub scores for 2 nd topic authority scores for 2 nd topic from to term term scores for 1 st topic term scores for 2 nd topic
35
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 35 TOPHITS Terms & Authorities on Sample Data TOPHITS uses 3D analysis to find the dominant groupings of web pages and terms. authority scores for 1 st topic hub scores for 1 st topic hub scores for 2 nd topic authority scores for 2 nd topic from to term term scores for 1 st topic term scores for 2 nd topic Tensor PARAFAC w k = # unique links using term k
36
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 36 Conclusions Real data are often in high dimensions with multiple aspects (modes) Tensors provide elegant theory and algorithms –PARAFAC and Tucker: discover groups
37
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 37 References T. G. Kolda, B. W. Bader and J. P. Kenny. Higher-Order Web Link Analysis Using Multilinear Algebra. In: ICDM 2005, Pages 242- 249, November 2005. Jimeng Sun, Spiros Papadimitriou, Philip Yu. Window-based Tensor Analysis on High- dimensional and Multi-aspect Streams, Proc. of the Int. Conf. on Data Mining (ICDM), Hong Kong, China, Dec 2006
38
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 38 Resources See tutorial on tensors, KDD’07 (w/ Tamara Kolda and Jimeng Sun): www.cs.cmu.edu/~christos/TALKS/KDD-07-tutorial
39
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 39 Tensor tools - resources Toolbox: from Tamara Kolda: csmr.ca.sandia.gov/~tgkolda/TensorToolbox 2-39 Copyright: Faloutsos, Tong (2009) 2-39 ICDE’09 T. G. Kolda and B. W. Bader. Tensor Decompositions and Applications. SIAM Review, Volume 51, Number 3, September 2009 csmr.ca.sandia.gov/~tgkolda/pubs/bibtgkfiles/TensorReview-preprint.pdf T. Kolda and J. Sun: Scalable Tensor Decomposition for Multi-Aspect Data Mining (ICDM 2008)
40
CMU SCS Outline Task 4: time-evolving graphs – tensors Task 5: community detection Task 6: virus propagation Task 7: scalability, parallelism and hadoop Conclusions KAIST-2011(C) 2011, C. Faloutsos 40
41
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 41 Detailed outline Motivation Hard clustering – k pieces Hard co-clustering – (k,l) pieces Hard clustering – optimal # pieces Observations
42
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 42 Problem Given a graph, and k Break it into k (disjoint) communities
43
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 43 Problem Given a graph, and k Break it into k (disjoint) communities k = 2
44
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 44 Solution #1: METIS Arguably, the best algorithm Open source, at –http://www.cs.umn.edu/~metishttp://www.cs.umn.edu/~metis and *many* related papers, at same url Main idea: –coarsen the graph; –partition; –un-coarsen
45
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 45 Solution #1: METIS G. Karypis and V. Kumar. METIS 4.0: Unstructured graph partitioning and sparse matrix ordering system. TR, Dept. of CS, Univ. of Minnesota, 1998.
46
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 46 Solution #2 (problem: hard clustering, k pieces) Spectral partitioning: Consider the 2 nd smallest eigenvector of the (normalized) Laplacian
47
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 47 Solutions #3, … Many more ideas: Clustering on the A 2 (square of adjacency matrix) [Zhou, Woodruff, PODS’04] Minimum cut / maximum flow [Flake+, KDD’00] …
48
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 48 Detailed outline Motivation Hard clustering – k pieces Hard co-clustering – (k,l) pieces Hard clustering – optimal # pieces Soft clustering – matrix decompositions Observations
49
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 49 Problem definition Given a bi-partite graph, and k, l Divide it into k row groups and l row groups (Also applicable to uni-partite graph)
50
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 50 Co-clustering Given data matrix and the number of row and column groups k and l Simultaneously –Cluster rows into k disjoint groups –Cluster columns into l disjoint groups
51
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 51 Copyright: Faloutsos, Tong (2009) 2-51 Co-clustering Let X and Y be discrete random variables – X and Y take values in {1, 2, …, m} and {1, 2, …, n} – p(X, Y) denotes the joint probability distribution—if not known, it is often estimated based on co-occurrence data –Application areas: text mining, market-basket analysis, analysis of browsing behavior, etc. Key Obstacles in Clustering Contingency Tables –High Dimensionality, Sparsity, Noise –Need for robust and scalable algorithms Reference: 1.Dhillon et al. Information-Theoretic Co-clustering, KDD’03
52
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 52 m m n nl k k l eg, terms x documents
53
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 53 doc x doc group term group x doc. group med. terms cs terms common terms med. doc cs doc term x term-group
54
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 54 Co-clustering Observations uses KL divergence, instead of L2 the middle matrix is not diagonal –we saw that earlier in the Tucker tensor decomposition s/w at: www.cs.utexas.edu/users/dml/Software/cocluster.html
55
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 55 Detailed outline Motivation Hard clustering – k pieces Hard co-clustering – (k,l) pieces Hard clustering – optimal # pieces Soft clustering – matrix decompositions Observations
56
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 56 Problem with Information Theoretic Co-clustering Number of row and column groups must be specified Desiderata: Simultaneously discover row and column groups Fully Automatic: No “magic numbers” Scalable to large graphs
57
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 57 Cross-association Desiderata: Simultaneously discover row and column groups Fully Automatic: No “magic numbers” Scalable to large matrices Reference: 1.Chakrabarti et al. Fully Automatic Cross-Associations, KDD’04
58
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 58 What makes a cross-association “good”? versus Column groups Row groups Why is this better?
59
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 59 What makes a cross-association “good”? versus Column groups Row groups Why is this better? simpler; easier to describe easier to compress!
60
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 60 What makes a cross-association “good”? Problem definition: given an encoding scheme decide on the # of col. and row groups k and l and reorder rows and columns, to achieve best compression
61
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 61 Main Idea size i * H(x i ) + Cost of describing cross-associations Code Cost Description Cost ΣiΣi Total Encoding Cost = Good Compression Better Clustering Minimize the total cost (# bits) for lossless compression
62
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 62 Algorithm k = 5 row groups k=1, l=2 k=2, l=2 k=2, l=3 k=3, l=3 k=3, l=4 k=4, l=4 k=4, l=5 l = 5 col groups
63
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 63 Experiments “CLASSIC” 3,893 documents 4,303 words 176,347 “dots” Combination of 3 sources: MEDLINE (medical) CISI (info. retrieval) CRANFIELD (aerodynamics) Documents Words
64
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 64 Experiments “CLASSIC” graph of documents & words: k=15, l=19 Documents Words
65
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 65 Experiments “CLASSIC” graph of documents & words: k=15, l=19 MEDLINE (medical) insipidus, alveolar, aortic, death, prognosis, intravenous blood, disease, clinical, cell, tissue, patient
66
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 66 Experiments “CLASSIC” graph of documents & words: k=15, l=19 CISI (Information Retrieval) providing, studying, records, development, students, rules abstract, notation, works, construct, bibliographies MEDLINE (medical)
67
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 67 Experiments “CLASSIC” graph of documents & words: k=15, l=19 CRANFIELD (aerodynamics) shape, nasa, leading, assumed, thin CISI (Information Retrieval) MEDLINE (medical)
68
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 68 Experiments “CLASSIC” graph of documents & words: k=15, l=19 paint, examination, fall, raise, leave, based CRANFIELD (aerodynamics) CISI (Information Retrieval) MEDLINE (medical)
69
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 69 Algorithm Code for cross-associations (matlab): www.cs.cmu.edu/~deepay/mywww/software/CrossAssociations -01-27-2005.tgz Variations and extensions: ‘Autopart’ [Chakrabarti, PKDD’04] www.cs.cmu.edu/~deepay
70
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 70 Algorithm Hadoop implementation [ICDM’08] Spiros Papadimitriou, Jimeng Sun: DisCo: Distributed Co-clustering with Map- Reduce: A Case Study towards Petabyte-Scale End-to-End Mining. ICDM 2008: 512-521
71
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 71 Detailed outline Motivation Hard clustering – k pieces Hard co-clustering – (k,l) pieces Hard clustering – optimal # pieces Observations
72
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 72 Observation #1 Skewed degree distributions – there are nodes with huge degree (>O(10^4), in facebook/linkedIn popularity contests!)
73
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 73 Observation #2 Maybe there are no good cuts: ``jellyfish’’ shape [Tauro+’01], [Siganos+,’06], strange behavior of cuts [Chakrabarti+’04], [Leskovec+,’08]
74
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 74 Observation #2 Maybe there are no good cuts: ``jellyfish’’ shape [Tauro+’01], [Siganos+,’06], strange behavior of cuts [Chakrabarti+,’04], [Leskovec+,’08] ? ?
75
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 75 Jellyfish model [Tauro+] … A Simple Conceptual Model for the Internet Topology, L. Tauro, C. Palmer, G. Siganos, M. Faloutsos, Global Internet, November 25-29, 2001 Jellyfish: A Conceptual Model for the AS Internet Topology G. Siganos, Sudhir L Tauro, M. Faloutsos, J. of Communications and Networks, Vol. 8, No. 3, pp 339- 350, Sept. 2006.
76
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 76 Strange behavior of min cuts ‘negative dimensionality’ (!) NetMine: New Mining Tools for Large Graphs, by D. Chakrabarti, Y. Zhan, D. Blandford, C. Faloutsos and G. Blelloch, in the SDM 2004 Workshop on Link Analysis, Counter-terrorism and Privacy Statistical Properties of Community Structure in Large Social and Information Networks, J. Leskovec, K. Lang, A. Dasgupta, M. Mahoney. WWW 2008.
77
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 77 “Min-cut” plot Do min-cuts recursively. log (# edges) log (mincut-size / #edges) N nodes Mincut size = sqrt(N)
78
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 78 “Min-cut” plot Do min-cuts recursively. log (# edges) log (mincut-size / #edges) N nodes New min-cut
79
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 79 “Min-cut” plot Do min-cuts recursively. log (# edges) log (mincut-size / #edges) N nodes New min-cut Slope = -0.5 For a d-dimensional grid, the slope is -1/d
80
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 80 “Min-cut” plot log (# edges) log (mincut-size / #edges) Slope = -1/d For a d-dimensional grid, the slope is -1/d log (# edges) log (mincut-size / #edges) For a random graph, the slope is 0
81
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 81 “Min-cut” plot What does it look like for a real-world graph? log (# edges) log (mincut-size / #edges) ?
82
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 82 Experiments Datasets: –Google Web Graph: 916,428 nodes and 5,105,039 edges –Lucent Router Graph: Undirected graph of network routers from www.isi.edu/scan/mercator/maps.html; 112,969 nodes and 181,639 edges www.isi.edu/scan/mercator/maps.html –User Website Clickstream Graph: 222,704 nodes and 952,580 edges NetMine: New Mining Tools for Large Graphs, by D. Chakrabarti, Y. Zhan, D. Blandford, C. Faloutsos and G. Blelloch, in the SDM 2004 Workshop on Link Analysis, Counter-terrorism and Privacy
83
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 83 Experiments Used the METIS algorithm [ Karypis, Kumar, 1995] log (# edges) log (mincut-size / #edges) Google Web graph Values along the y- axis are averaged We observe a “lip” for large edges Slope of -0.4, corresponds to a 2.5- dimensional grid! Slope~ -0.4
84
CMU SCS Google graph KAIST-2011(C) 2011, C. Faloutsos 84 Log(#edges) log (mincut-size / #edges) Log(#edges) All min-cuts averaged
85
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 85 Experiments Same results for other graphs too… Lucent Router graph Clickstream graph
86
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 86 Conclusions – Practitioner’s guide Hard clustering – k pieces Hard co-clustering – (k,l) pieces Hard clustering – optimal # pieces Observations METIS Co-clustering Cross-associations ‘jellyfish’: Maybe, there are no good cuts
87
CMU SCS Outline Task 4: time-evolving graphs – tensors Task 5: community detection Task 6: virus propagation Task 7: scalability, parallelism and hadoop Conclusions KAIST-2011(C) 2011, C. Faloutsos 87
88
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 88 Detailed outline Problem definition Analysis Experiments
89
CMU SCS Immunization and epidemic thresholds Q1: which nodes to immunize? Q2: will a virus vanish, or will it create an epidemic? KAIST-2011(C) 2011, C. Faloutsos 89
90
CMU SCS Q1: Immunization: ? ? Given a network, k vaccines, and the virus details Which nodes to immunize? KAIST-2011 90 (C) 2011, C. Faloutsos
91
CMU SCS Q1: Immunization: ? ? Given a network, k vaccines, and the virus details Which nodes to immunize? KAIST-2011 91 (C) 2011, C. Faloutsos
92
CMU SCS Q1: Immunization: ? ? Given a network, k vaccines, and the virus details Which nodes to immunize? KAIST-2011 92 (C) 2011, C. Faloutsos
93
CMU SCS Q1: Immunization: ? ? Given a network, k vaccines, and the virus details Which nodes to immunize? A: immunize the ones that maximally raise the `epidemic threshold’ [Tong+, ICDM’10] KAIST-2011 93 (C) 2011, C. Faloutsos
94
CMU SCS Q2: will a virus take over? Flu-like virus (no immunity, ‘SIS’) Mumps (life-time immunity, ‘SIR’) Pertussis (finite-length immunity, ‘SIRS’) KAIST-2011(C) 2011, C. Faloutsos 94 ? ? : attack prob : heal prob
95
CMU SCS Q2: will a virus take over? Flu-like virus (no immunity, ‘SIS’) Mumps (life-time immunity, ‘SIR’) Pertussis (finite-length immunity, ‘SIRS’) KAIST-2011(C) 2011, C. Faloutsos 95 ? ? : attack prob : heal prob depends on connectivity (avg degree? Max degree? variance? Something else?
96
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 96 The model: SIS ‘Flu’ like: Susceptible-Infected-Susceptible Virus ‘strength’ s= / Infected Healthy NN1 N3 N2 Prob. Prob. β Prob.
97
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 97 Epidemic threshold of a graph: the value of , such that if strength s = / < an epidemic can not happen Thus, given a graph compute its epidemic threshold
98
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 98 Detailed outline Problem definition Analysis Experiments
99
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 99 Epidemic threshold What should depend on? avg. degree? and/or highest degree? and/or variance of degree? and/or third moment of degree? and/or diameter?
100
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 100 Epidemic threshold [Theorem] We have no epidemic, if β/δ <τ = 1/ λ 1,A
101
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 101 Epidemic threshold [Theorem] We have no epidemic, if β/δ <τ = 1/ λ 1,A largest eigenvalue of adj. matrix A attack prob. recovery prob. epidemic threshold Proof: [Wang+03] (proof: for SIS=flu only)
102
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 102 Beginning of proof Healthy @ t+1: - ( healthy or healed ) - and not attacked @ t Let: p(i, t) = Prob node i is sick @ t+1 1 - p(i, t+1 ) = (1 – p(i, t) + p(i, t) * ) * j (1 – aji * p(j, t) ) Below threshold, if the above non-linear dynamical system above is ‘stable’ (eigenvalue of Hessian < 1 )
103
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 103 Epidemic threshold for various networks Formula includes older results as special cases: Homogeneous networks [Kephart+White] –λ 1,A = ; τ = 1/ ( : avg degree) Star networks (d = degree of center) –λ 1,A = sqrt(d); τ = 1/ sqrt(d) Infinite power-law networks –λ 1,A = ∞; τ = 0 ; [Barabasi]
104
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 104 Epidemic threshold [Theorem 2] Below the epidemic threshold, the epidemic dies out exponentially
105
CMU SCS Recent generalization [Prakash+, arxiv ‘10]: similar threshold, for almost all virus propagation models (VPM) –SIS -> flu –SIR -> mumps –SIRS -> whooping cough (temporary immunity) –SIIR (-> HIV) –… KAIST-2011(C) 2011, C. Faloutsos 105
106
CMU SCS A2: will a virus take over? For all typical virus propagation models (flu, mumps, pertussis, HIV, etc) The only connectivity measure that matters, is 1 the first eigenvalue of the adj. matrix Proof for all VPM: [Prakash+, ‘10, arxiv] KAIST-2011(C) 2011, C. Faloutsos 106 ? ?
107
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 107 Detailed outline Epidemic threshold –Problem definition –Analysis –Experiments
108
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 108 Experiments (Oregon) / > τ (above threshold) / = τ (at the threshold) / < τ (below threshold)
109
KAIST-2011(C) 2011, C. Faloutsos 109 SIS simulation - # infected nodes vs time Time (linear scale) #inf. (log scale) above at below Log - Lin
110
KAIST-2011(C) 2011, C. Faloutsos 110 SIS simulation - # infected nodes vs time Log - Lin Time (linear scale) #inf. (log scale) above at below Exponential decay
111
KAIST-2011(C) 2011, C. Faloutsos 111 SIS simulation - # infected nodes vs time Log - Log Time (log scale) #inf. (log scale) above at below
112
KAIST-2011(C) 2011, C. Faloutsos 112 SIS simulation - # infected nodes vs time Time (log scale) #inf. (log scale) above at below Log - Log Power-law Decay (!)
113
How about other VPMs? KAIST-2011(C) 2011, C. Faloutsos P6-113
114
CMU SCS A2: will a virus take over? (SIRS case) KAIST-2011(C) 2011, C. Faloutsos 114 Fraction of infected Time ticks Below: exp. extinction Above: take-over Graph: Portland, OR 31M links 1.5M nodes
115
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 115 Conclusions λ 1,A : Eigenvalue of adjacency matrix determines the survival of (almost) any virus measure of connectivity (~ # paths) Can answer ‘what-if’ scenarios –May guide immunization policies Can help us avoid expensive simulations
116
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 116 References D. Chakrabarti, Y. Wang, C. Wang, J. Leskovec, and C. Faloutsos, Epidemic Thresholds in Real Networks, in ACM TISSEC, 10(4), 2008 Ganesh, A., Massoulie, L., and Towsley, D., 2005. The effect of network topology on the spread of epidemics. In INFOCOM.
117
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 117 References (cont’d) Hethcote, H. W. 2000. The mathematics of infectious diseases. SIAM Review 42, 599– 653. Hethcote, H. W. AND Yorke, J. A. 1984. Gonorrhea Transmission Dynamics and Control. Vol. 56. Springer. Lecture Notes in Biomathematics.
118
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 118 References (cont’d) Y. Wang, D. Chakrabarti, C. Wang and C. Faloutsos, Epidemic Spreading in Real Networks: An Eigenvalue Viewpoint, in SRDS 2003 (pages 25-34), Florence, Italy
119
CMU SCS Outline Task 4: time-evolving graphs – tensors Task 5: community detection Task 6: virus propagation Task 7: scalability, parallelism and hadoop Conclusions KAIST-2011(C) 2011, C. Faloutsos 119
120
CMU SCS KAIST-2011(C) 2011, C. Faloutsos P8-120 Scalability How about if graph/tensor does not fit in core? How about handling huge graphs?
121
CMU SCS KAIST-2011(C) 2011, C. Faloutsos P8-121 Scalability How about if graph/tensor does not fit in core? [‘MET’: Kolda, Sun, ICMD’08, best paper award] How about handling huge graphs?
122
CMU SCS KAIST-2011(C) 2011, C. Faloutsos P8-122 Scalability Google: > 450,000 processors in clusters of ~2000 processors each [ Barroso, Dean, Hölzle, “Web Search for a Planet: The Google Cluster Architecture” IEEE Micro 2003 ] Yahoo: 5Pb of data [Fayyad, KDD’07] Problem: machine failures, on a daily basis How to parallelize data mining tasks, then?
123
CMU SCS KAIST-2011(C) 2011, C. Faloutsos P8-123 Scalability Google: > 450,000 processors in clusters of ~2000 processors each [ Barroso, Dean, Hölzle, “Web Search for a Planet: The Google Cluster Architecture” IEEE Micro 2003 ] Yahoo: 5Pb of data [Fayyad, KDD’07] Problem: machine failures, on a daily basis How to parallelize data mining tasks, then? A: map/reduce – hadoop (open-source clone) http://hadoop.apache.org/ http://hadoop.apache.org/
124
CMU SCS KAIST-2011(C) 2011, C. Faloutsos P8-124 2’ intro to hadoop master-slave architecture; n-way replication (default n=3) ‘group by’ of SQL (in parallel, fault-tolerant way) e.g, find histogram of word frequency –compute local histograms –then merge into global histogram select course-id, count(*) from ENROLLMENT group by course-id
125
CMU SCS KAIST-2011(C) 2011, C. Faloutsos P8-125 2’ intro to hadoop master-slave architecture; n-way replication (default n=3) ‘group by’ of SQL (in parallel, fault-tolerant way) e.g, find histogram of word frequency –compute local histograms –then merge into global histogram select course-id, count(*) from ENROLLMENT group by course-id map reduce
126
CMU SCS KAIST-2011(C) 2011, C. Faloutsos P8-126 User Program Reducer Master Mapper fork assign map assign reduce read local write remote read, sort Output File 0 Output File 1 write Split 0 Split 1 Split 2 Input Data (on HDFS) By default: 3-way replication; Late/dead machines: ignored, transparently (!)
127
CMU SCS KAIST-2011(C) 2011, C. Faloutsos P8-127 D.I.S.C. ‘Data Intensive Scientific Computing’ [R. Bryant, CMU] –‘big data’ –www.cs.cmu.edu/~bryant/pubdir/cmu-cs-07- 128.pdf
128
CMU SCS KAIST-2011(C) 2011, C. Faloutsos P8-128 ~200Gb (Yahoo crawl) - Degree Distribution: in 12 minutes with 50 machines Many (link spams ?) at out-degree 1200 Analysis of a large graph
129
CMU SCS (C) 2011, C. Faloutsos 129 CentralizedHadoop/PEG ASUS Degree Distr. old Pagerank old Diameter/ANF old DONE Conn. Comp old DONE TrianglesDONE VisualizationSTARTED Outline – Algorithms & results KAIST-2011
130
CMU SCS HADI for diameter estimation Radius Plots for Mining Tera-byte Scale Graphs U Kang, Charalampos Tsourakakis, Ana Paula Appel, Christos Faloutsos, Jure Leskovec, SDM’10 Naively: diameter needs O(N**2) space and up to O(N**3) time – prohibitive (N~1B) Our HADI: linear on E (~10B) –Near-linear scalability wrt # machines –Several optimizations -> 5x faster (C) 2011, C. Faloutsos 130 KAIST-2011
131
CMU SCS ???? 19+ [Barabasi+] 131 (C) 2011, C. Faloutsos Radius Count KAIST-2011 ~1999, ~1M nodes
132
CMU SCS YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) Largest publicly available graph ever studied. ???? 19+ [Barabasi+] 132 (C) 2011, C. Faloutsos Radius Count KAIST-2011 ?? ~1999, ~1M nodes
133
CMU SCS YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) Largest publicly available graph ever studied. ???? 19+? [Barabasi+] 133 (C) 2011, C. Faloutsos Radius Count KAIST-2011 14 (dir.) ~7 (undir.)
134
CMU SCS YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) 7 degrees of separation (!) Diameter: shrunk ???? 19+? [Barabasi+] 134 (C) 2011, C. Faloutsos Radius Count KAIST-2011 14 (dir.) ~7 (undir.)
135
CMU SCS YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) Q: Shape? ???? 135 (C) 2011, C. Faloutsos Radius Count KAIST-2011 ~7 (undir.)
136
CMU SCS 136 (C) 2011, C. Faloutsos YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) effective diameter: surprisingly small. Multi-modality (?!) KAIST-2011
137
CMU SCS Radius Plot of GCC of YahooWeb. 137 (C) 2011, C. FaloutsosKAIST-2011
138
CMU SCS 138 (C) 2011, C. Faloutsos YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) effective diameter: surprisingly small. Multi-modality: probably mixture of cores. KAIST-2011
139
CMU SCS 139 (C) 2011, C. Faloutsos YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) effective diameter: surprisingly small. Multi-modality: probably mixture of cores. KAIST-2011 EN ~7 Conjecture: DE BR
140
CMU SCS 140 (C) 2011, C. Faloutsos YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) effective diameter: surprisingly small. Multi-modality: probably mixture of cores. KAIST-2011 ~7 Conjecture:
141
CMU SCS Running time - Kronecker and Erdos-Renyi Graphs with billions edges. details
142
CMU SCS (C) 2011, C. Faloutsos 142 CentralizedHadoop/PEG ASUS Degree Distr. old Pagerank old Diameter/ANF old DONE Conn. Comp old DONE TrianglesDONE VisualizationSTARTED Outline – Algorithms & results KAIST-2011
143
CMU SCS Generalized Iterated Matrix Vector Multiplication (GIMV) (C) 2011, C. Faloutsos 143 PEGASUS: A Peta-Scale Graph Mining System - Implementation and ObservationsSystem - Implementation and Observations. U Kang, Charalampos E. Tsourakakis, and Christos Faloutsos. (ICDM) 2009, Miami, Florida, USA. Best Application Paper (runner-up).ICDM KAIST-2011
144
CMU SCS Generalized Iterated Matrix Vector Multiplication (GIMV) (C) 2011, C. Faloutsos 144 PageRank proximity (RWR) Diameter Connected components (eigenvectors, Belief Prop. … ) Matrix – vector Multiplication (iterated) KAIST-2011 details
145
CMU SCS 145 Example: GIM-V At Work Connected Components – 4 observations: Size Count (C) 2011, C. Faloutsos KAIST-2011
146
CMU SCS 146 Example: GIM-V At Work Connected Components Size Count (C) 2011, C. Faloutsos KAIST-2011 1) 10K x larger than next
147
CMU SCS 147 Example: GIM-V At Work Connected Components Size Count (C) 2011, C. Faloutsos KAIST-2011 2) ~0.7B singleton nodes
148
CMU SCS 148 Example: GIM-V At Work Connected Components Size Count (C) 2011, C. Faloutsos KAIST-2011 3) SLOPE!
149
CMU SCS 149 Example: GIM-V At Work Connected Components Size Count 300-size cmpt X 500. Why? 1100-size cmpt X 65. Why? (C) 2011, C. Faloutsos KAIST-2011 4) Spikes!
150
CMU SCS 150 Example: GIM-V At Work Connected Components Size Count suspicious financial-advice sites (not existing now) (C) 2011, C. Faloutsos KAIST-2011
151
CMU SCS 151 GIM-V At Work Connected Components over Time LinkedIn: 7.5M nodes and 58M edges Stable tail slope after the gelling point (C) 2011, C. FaloutsosKAIST-2011
152
CMU SCS KAIST-2011(C) 2011, C. Faloutsos P8-152 Conclusions Hadoop: promising architecture for Tera/Peta scale graph mining Resources: http://hadoop.apache.org/core/ http://hadoop.apache.org/pig/ Higher-level language for data processing
153
CMU SCS KAIST-2011(C) 2011, C. Faloutsos P8-153 References Jeffrey Dean and Sanjay Ghemawat, MapReduce: Simplified Data Processing on Large Clusters, OSDI'04Jeffrey DeanSanjay Ghemawat Christopher Olston, Benjamin Reed, Utkarsh Srivastava, Ravi Kumar, Andrew Tomkins: Pig latin: a not-so-foreign language for data processing. SIGMOD 2008: 1099-1110Benjamin ReedUtkarsh SrivastavaRavi KumarAndrew TomkinsSIGMOD 2008
154
CMU SCS Overall Conclusions Real graphs exhibit surprising patterns (power laws, shrinking diameter, super- linearity on edge weights, triangles etc) SVD: a powerful tool (HITS, PageRank) Several other tools: tensors, METIS, … –But: good communities might not exist… Immunization: first eigenvalue Scalability: hadoop/parallelism KAIST-2011(C) 2011, C. Faloutsos 154
155
CMU SCS (C) 2011, C. Faloutsos 155 Our goal: Open source system for mining huge graphs: PEGASUS project (PEta GrAph mining System) www.cs.cmu.edu/~pegasus code and papers KAIST-2011
156
CMU SCS (C) 2011, C. Faloutsos 156 Project info Akoglu, Leman Chau, Polo Kang, U McGlohon, Mary Tong, Hanghang Prakash, Aditya KAIST-2011 Thanks to: NSF IIS-0705359, IIS-0534205, CTA-INARC ; Yahoo (M45), LLNL, IBM, SPRINT, Google, INTEL, HP, iLab www.cs.cmu.edu/~pegasus Koutra, Danae
157
CMU SCS Extra material E-bay fraud detection Outlier detection KAIST-2011(C) 2011, C. Faloutsos 157
158
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 158 Detailed outline Fraud detection in e-bay Anomaly detection
159
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 159 E-bay Fraud detection w/ Polo Chau & Shashank Pandit, CMU NetProbe: A Fast and Scalable System for Fraud Detection in Online Auction Networks, S. Pandit, D. H. Chau, S. Wang, and C. Faloutsos (WWW'07), pp. 201-210
160
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 160 E-bay Fraud detection lines: positive feedbacks would you buy from him/her?
161
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 161 E-bay Fraud detection lines: positive feedbacks would you buy from him/her? or him/her?
162
CMU SCS KAIST-2011(C) 2011, C. Faloutsos 162 E-bay Fraud detection - NetProbe Belief Propagation gives:
163
CMU SCS Popular press And less desirable attention: E-mail from ‘Belgium police’ (‘copy of your code?’) KAIST-2011(C) 2011, C. Faloutsos 163
164
CMU SCS Extra material E-bay fraud detection Outlier detection KAIST-2011(C) 2011, C. Faloutsos 164
165
CMU SCS OddBall: Spotting A n o m a l i e s in Weighted Graphs Leman Akoglu, Mary McGlohon, Christos Faloutsos Carnegie Mellon University School of Computer Science PAKDD 2010, Hyderabad, India
166
CMU SCS Main idea For each node, extract ‘ego-net’ (=1-step-away neighbors) Extract features (#edges, total weight, etc etc) Compare with the rest of the population (C) 2011, C. Faloutsos 166 KAIST-2011
167
CMU SCS What is an egonet? ego 167 egonet (C) 2011, C. FaloutsosKAIST-2011
168
CMU SCS Selected Features N i : number of neighbors (degree) of ego i E i : number of edges in egonet i W i : total weight of egonet i λ w,i : principal eigenvalue of the weighted adjacency matrix of egonet I 168 (C) 2011, C. FaloutsosKAIST-2011
169
CMU SCS Near-Clique/Star 169 KAIST-2011(C) 2011, C. Faloutsos
170
CMU SCS Near-Clique/Star 170 (C) 2011, C. FaloutsosKAIST-2011
171
CMU SCS END KAIST-2011(C) 2011, C. Faloutsos 171
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.