1. CMU SCS Large Graph Mining – Patterns, Tools and Cascade analysis Christos Faloutsos CMU

2. CMU SCS Thank you! Brian Gallagher Jan Winfield LLNL, Jan 2013C. Faloutsos (CMU) 2

3. CMU SCS C. Faloutsos (CMU) 3 Roadmap Introduction – Motivation –Why ‘big data’ –Why (big) graphs? Problem#1: Patterns in graphs Problem#2: Tools Problem#3: Scalability Conclusions LLNL, Jan 2013

4. CMU SCS Why ‘big data’ Why? What is the problem definition? What are the major research challenges? LLNL, Jan 2013C. Faloutsos (CMU) 4

5. CMU SCS Main message: Big data: often > experts ‘Super Crunchers’ Why Thinking-By-Numbers is the New Way To Be Smart by Ian Ayres, 2008 Google won the machine translation competition 2005 http://www.itl.nist.gov/iad/mig//tests/mt/2005/doc/mt05eval_official_r esults_release_20050801_v3.htmlhttp://www.itl.nist.gov/iad/mig//tests/mt/2005/doc/mt05eval_official_r esults_release_20050801_v3.html LLNL, Jan 2013C. Faloutsos (CMU) 5

6. CMU SCS Problem definition – big picture LLNL, Jan 2013C. Faloutsos (CMU) 6 Tera/Peta-byte data AnalyticsInsights, outliers

7. CMU SCS Problem definition – big picture LLNL, Jan 2013C. Faloutsos (CMU) 7 Tera/Peta-byte data AnalyticsInsights, outliers Main emphasis in this talk

8. CMU SCS Problem definition – big picture LLNL, Jan 2013C. Faloutsos (CMU) 8 Tera/Peta-byte data (my personal) rules of thumb: if data fits in memory -> R, matlab, scipy single disk -> RDBMS (sqlite3, mysql, postgres) multiple (<100-1000) disks: parallel RDBMS (Vertica, TeraData) multiple (>1000) disks: hadoop, pig

9. CMU SCS C. Faloutsos (CMU) 9 (Free) Resource for graphs Open source system for mining huge graphs: PEGASUS project (PEta GrAph mining System) www.cs.cmu.edu/~pegasus Apache license for s/w code and papers LLNL, Jan 2013

10. CMU SCS Research challenges The usual ones from data mining –Data cleansing –Feature engineering –… PLUS –Scalability ( < O(N**2)) –Real data *disobey* textbook assumptions (uniformity, independence, Gaussian, Poisson) with huge performance implications LLNL, Jan 2013C. Faloutsos (CMU) 10

11. CMU SCS C. Faloutsos (CMU) 11 Roadmap Introduction – Motivation –Why ‘big data’ –Why (big) graphs? Problem#1: Patterns in graphs Problem#2: Tools Problem#3: Scalability Conclusions LLNL, Jan 2013

12. CMU SCS C. Faloutsos (CMU) 12 Graphs - why should we care? Internet Map [lumeta.com] Food Web [Martinez ’91] >$10B revenue >0.5B users LLNL, Jan 2013

13. CMU SCS C. Faloutsos (CMU) 13 Graphs - why should we care? IR: bi-partite graphs (doc-terms) web: hyper-text graph... and more: D1D1 DNDN T1T1 TMTM... LLNL, Jan 2013

14. CMU SCS C. Faloutsos (CMU) 14 Graphs - why should we care? ‘viral’ marketing web-log (‘blog’) news propagation computer network security: email/IP traffic and anomaly detection.... Subject-verb-object -> graph Many-to-many db relationship -> graph LLNL, Jan 2013

15. CMU SCS C. Faloutsos (CMU) 15 Outline Introduction – Motivation Problem#1: Patterns in graphs –Static graphs –Weighted graphs –Time evolving graphs Problem#2: Tools Problem#3: Scalability Conclusions LLNL, Jan 2013

16. CMU SCS C. Faloutsos (CMU) 16 Problem #1 - network and graph mining What does the Internet look like? What does FaceBook look like? What is ‘normal’/‘abnormal’? which patterns/laws hold? LLNL, Jan 2013

17. CMU SCS C. Faloutsos (CMU) 17 Problem #1 - network and graph mining What does the Internet look like? What does FaceBook look like? What is ‘normal’/‘abnormal’? which patterns/laws hold? –To spot anomalies (rarities), we have to discover patterns LLNL, Jan 2013

18. CMU SCS C. Faloutsos (CMU) 18 Problem #1 - network and graph mining What does the Internet look like? What does FaceBook look like? What is ‘normal’/‘abnormal’? which patterns/laws hold? –To spot anomalies (rarities), we have to discover patterns –Large datasets reveal patterns/anomalies that may be invisible otherwise… LLNL, Jan 2013

19. CMU SCS C. Faloutsos (CMU) 19 Graph mining Are real graphs random? LLNL, Jan 2013

20. CMU SCS C. Faloutsos (CMU) 20 Laws and patterns Are real graphs random? A: NO!! –Diameter –in- and out- degree distributions –other (surprising) patterns So, let’s look at the data LLNL, Jan 2013

21. CMU SCS C. Faloutsos (CMU) 21 Solution# S.1 Power law in the degree distribution [SIGCOMM99] log(rank) log(degree) internet domains att.com ibm.com LLNL, Jan 2013

22. CMU SCS C. Faloutsos (CMU) 22 Solution# S.1 Power law in the degree distribution [SIGCOMM99] log(rank) log(degree) -0.82 internet domains att.com ibm.com LLNL, Jan 2013

23. CMU SCS C. Faloutsos (CMU) 23 Solution# S.2: Eigen Exponent E A2: power law in the eigenvalues of the adjacency matrix E = -0.48 Exponent = slope Eigenvalue Rank of decreasing eigenvalue May 2001 LLNL, Jan 2013

24. CMU SCS C. Faloutsos (CMU) 24 Solution# S.2: Eigen Exponent E [Mihail, Papadimitriou ’02]: slope is ½ of rank exponent E = -0.48 Exponent = slope Eigenvalue Rank of decreasing eigenvalue May 2001 LLNL, Jan 2013

25. CMU SCS C. Faloutsos (CMU) 25 But: How about graphs from other domains? LLNL, Jan 2013

26. CMU SCS C. Faloutsos (CMU) 26 More power laws: web hit counts [w/ A. Montgomery] Web Site Traffic in-degree (log scale) Count (log scale) Zipf users sites ``ebay’’ LLNL, Jan 2013

27. CMU SCS C. Faloutsos (CMU) 27 epinions.com who-trusts-whom [Richardson + Domingos, KDD 2001] (out) degree count trusts-2000-people user LLNL, Jan 2013

28. CMU SCS And numerous more # of sexual contacts Income [Pareto] –’80-20 distribution’ Duration of downloads [Bestavros+] Duration of UNIX jobs (‘mice and elephants’) Size of files of a user … ‘Black swans’ LLNL, Jan 2013C. Faloutsos (CMU) 28

29. CMU SCS C. Faloutsos (CMU) 29 Roadmap Introduction – Motivation Problem#1: Patterns in graphs –Static graphs degree, diameter, eigen, triangles cliques –Weighted graphs –Time evolving graphs Problem#2: Tools LLNL, Jan 2013

30. CMU SCS C. Faloutsos (CMU) 30 Solution# S.3: Triangle ‘Laws’ Real social networks have a lot of triangles LLNL, Jan 2013

31. CMU SCS C. Faloutsos (CMU) 31 Solution# S.3: Triangle ‘Laws’ Real social networks have a lot of triangles –Friends of friends are friends Any patterns? LLNL, Jan 2013

32. CMU SCS C. Faloutsos (CMU) 32 Triangle Law: #S.3 [Tsourakakis ICDM 2008] ASNHEP-TH Epinions X-axis: # of participating triangles Y: count (~ pdf) LLNL, Jan 2013

33. CMU SCS C. Faloutsos (CMU) 33 Triangle Law: #S.3 [Tsourakakis ICDM 2008] ASNHEP-TH Epinions LLNL, Jan 2013 X-axis: # of participating triangles Y: count (~ pdf)

34. CMU SCS C. Faloutsos (CMU) 34 Triangle Law: #S.4 [Tsourakakis ICDM 2008] SNReuters Epinions X-axis: degree Y-axis: mean # triangles n friends -> ~n 1.6 triangles LLNL, Jan 2013

35. CMU SCS C. Faloutsos (CMU) 35 Triangle Law: Computations [Tsourakakis ICDM 2008] But: triangles are expensive to compute (3-way join; several approx. algos) Q: Can we do that quickly? details LLNL, Jan 2013

36. CMU SCS C. Faloutsos (CMU) 36 Triangle Law: Computations [Tsourakakis ICDM 2008] But: triangles are expensive to compute (3-way join; several approx. algos) Q: Can we do that quickly? A: Yes! #triangles = 1/6 Sum ( i 3 ) (and, because of skewness (S2), we only need the top few eigenvalues! details LLNL, Jan 2013

37. CMU SCS C. Faloutsos (CMU) 37 Triangle Law: Computations [Tsourakakis ICDM 2008] 1000x+ speed-up, >90% accuracy details LLNL, Jan 2013

38. CMU SCS Triangle counting for large graphs? Anomalous nodes in Twitter(~ 3 billion edges) [U Kang, Brendan Meeder, +, PAKDD’11] 38 LLNL, Jan 2013 38 C. Faloutsos (CMU) ?? ?

39. CMU SCS Triangle counting for large graphs? Anomalous nodes in Twitter(~ 3 billion edges) [U Kang, Brendan Meeder, +, PAKDD’11] 39 LLNL, Jan 2013 39 C. Faloutsos (CMU)

40. CMU SCS Triangle counting for large graphs? Anomalous nodes in Twitter(~ 3 billion edges) [U Kang, Brendan Meeder, +, PAKDD’11] 40 LLNL, Jan 2013 40 C. Faloutsos (CMU)

41. CMU SCS Triangle counting for large graphs? Anomalous nodes in Twitter(~ 3 billion edges) [U Kang, Brendan Meeder, +, PAKDD’11] 41 LLNL, Jan 2013 41 C. Faloutsos (CMU)

42. CMU SCS C. Faloutsos (CMU) 42 Roadmap Introduction – Motivation Problem#1: Patterns in graphs –Static graphs degree, diameter, eigen, triangles cliques –Weighted graphs –Time evolving graphs Problem#2: Tools LLNL, Jan 2013

43. CMU SCS C. Faloutsos (CMU) 43 Observations on weighted graphs? A: yes - even more ‘laws’! M. McGlohon, L. Akoglu, and C. Faloutsos Weighted Graphs and Disconnected Components: Patterns and a Generator. SIG-KDD 2008 LLNL, Jan 2013

44. CMU SCS C. Faloutsos (CMU) 44 Observation W.1: Fortification Q: How do the weights of nodes relate to degree? LLNL, Jan 2013

45. CMU SCS C. Faloutsos (CMU) 45 Observation W.1: Fortification More donors, more $ ? $10 $5 LLNL, Jan 2013 ‘Reagan’ ‘Clinton’ $7

46. CMU SCS Edges (# donors) In-weights ($) C. Faloutsos (CMU) 46 Observation W.1: fortification: Snapshot Power Law Weight: super-linear on in-degree exponent ‘iw’: 1.01 < iw < 1.26 Orgs-Candidates e.g. John Kerry, $10M received, from 1K donors More donors, even more $ $10 $5 LLNL, Jan 2013

47. CMU SCS C. Faloutsos (CMU) 47 Roadmap Introduction – Motivation Problem#1: Patterns in graphs –Static graphs –Weighted graphs –Time evolving graphs Problem#2: Tools … LLNL, Jan 2013

48. CMU SCS C. Faloutsos (CMU) 48 Problem: Time evolution with Jure Leskovec (CMU -> Stanford) and Jon Kleinberg (Cornell – sabb. @ CMU) LLNL, Jan 2013

49. CMU SCS C. Faloutsos (CMU) 49 T.1 Evolution of the Diameter Prior work on Power Law graphs hints at slowly growing diameter: –diameter ~ O(log N) –diameter ~ O(log log N) What is happening in real data? LLNL, Jan 2013

50. CMU SCS C. Faloutsos (CMU) 50 T.1 Evolution of the Diameter Prior work on Power Law graphs hints at slowly growing diameter: –diameter ~ O(log N) –diameter ~ O(log log N) What is happening in real data? Diameter shrinks over time LLNL, Jan 2013

51. CMU SCS C. Faloutsos (CMU) 51 T.1 Diameter – “Patents” Patent citation network 25 years of data @1999 –2.9 M nodes –16.5 M edges time [years] diameter LLNL, Jan 2013

52. CMU SCS C. Faloutsos (CMU) 52 T.2 Temporal Evolution of the Graphs N(t) … nodes at time t E(t) … edges at time t Suppose that N(t+1) = 2 * N(t) Q: what is your guess for E(t+1) =? 2 * E(t) LLNL, Jan 2013

53. CMU SCS C. Faloutsos (CMU) 53 T.2 Temporal Evolution of the Graphs N(t) … nodes at time t E(t) … edges at time t Suppose that N(t+1) = 2 * N(t) Q: what is your guess for E(t+1) =? 2 * E(t) A: over-doubled! –But obeying the ``Densification Power Law’’ LLNL, Jan 2013

54. CMU SCS C. Faloutsos (CMU) 54 T.2 Densification – Patent Citations Citations among patents granted @1999 –2.9 M nodes –16.5 M edges Each year is a datapoint N(t) E(t) 1.66 LLNL, Jan 2013

55. CMU SCS C. Faloutsos (CMU) 55 Roadmap Introduction – Motivation Problem#1: Patterns in graphs –Static graphs –Weighted graphs –Time evolving graphs Problem#2: Tools … LLNL, Jan 2013

56. CMU SCS C. Faloutsos (CMU) 56 T.3 : popularity over time Post popularity drops-off – exponentially? lag: days after post # in links 1 2 3 @t @t + lag LLNL, Jan 2013

57. CMU SCS C. Faloutsos (CMU) 57 T.3 : popularity over time Post popularity drops-off – exponentially? POWER LAW! Exponent? # in links (log) days after post (log) LLNL, Jan 2013

58. CMU SCS C. Faloutsos (CMU) 58 T.3 : popularity over time Post popularity drops-off – exponentially? POWER LAW! Exponent? -1.6 close to -1.5: Barabasi’s stack model and like the zero-crossings of a random walk # in links (log) -1.6 days after post (log) LLNL, Jan 2013

59. CMU SCS LLNL, Jan 2013C. Faloutsos (CMU) 59 -1.5 slope J. G. Oliveira & A.-L. Barabási Human Dynamics: The Correspondence Patterns of Darwin and Einstein. Nature 437, 1251 (2005). [PDF]PDF Response time (log) Prob(RT > x) (log)

60. CMU SCS C. Faloutsos (CMU) 60 Roadmap Introduction – Motivation Problem#1: Patterns in graphs Problem#2: Tools –Belief Propagation –Tensors –Spike analysis Problem#3: Scalability Conclusions LLNL, Jan 2013

61. CMU SCS LLNL, Jan 2013C. Faloutsos (CMU) 61 E-bay Fraud detection w/ Polo Chau & Shashank Pandit, CMU [www’07]

62. CMU SCS LLNL, Jan 2013C. Faloutsos (CMU) 62 E-bay Fraud detection

63. CMU SCS LLNL, Jan 2013C. Faloutsos (CMU) 63 E-bay Fraud detection

64. CMU SCS LLNL, Jan 2013C. Faloutsos (CMU) 64 E-bay Fraud detection - NetProbe

65. CMU SCS LLNL, Jan 2013C. Faloutsos (CMU) 65 E-bay Fraud detection - NetProbe FAH F99% A H49% Compatibility matrix heterophily

66. CMU SCS C. Faloutsos (CMU) 66 Roadmap Introduction – Motivation Problem#1: Patterns in graphs Problem#2: Tools –Belief Propagation –Tensors –Spike analysis Problem#3: Scalability Conclusions LLNL, Jan 2013

67. CMU SCS GigaTensor: Scaling Tensor Analysis Up By 100 Times – Algorithms and Discoveries U Kang Christos Faloutsos KDD’12 Evangelos Papalexakis Abhay Harpale LLNL, Jan 2013 67 C. Faloutsos (CMU)

68. CMU SCS Background: Tensor Tensors (=multi-dimensional arrays) are everywhere –Hyperlinks &anchor text [Kolda+,05] URL 1 URL 2 Anchor Text Java C++ C# 1 1 1 1 1 1 1 LLNL, Jan 2013 68 C. Faloutsos (CMU)

69. CMU SCS Background: Tensor Tensors (=multi-dimensional arrays) are everywhere –Sensor stream (time, location, type) –Predicates (subject, verb, object) in knowledge base “Barrack Obama is the president of U.S.” “Eric Clapton plays guitar” (26M) (48M) NELL (Never Ending Language Learner) data Nonzeros =144M LLNL, Jan 2013 69 C. Faloutsos (CMU)

70. CMU SCS Background: Tensor Tensors (=multi-dimensional arrays) are everywhere –Sensor stream (time, location, type) –Predicates (subject, verb, object) in knowledge base LLNL, Jan 2013 70 C. Faloutsos (CMU) IP-destination IP-source Time-stamp Anomaly Detection in Computer networks

71. CMU SCS all I learned on tensors: from LLNL, Jan 2013C. Faloutsos (CMU) 71 Nikos Sidiropoulos UMN Tamara Kolda, Sandia Labs (tensor toolbox)

72. CMU SCS Problem Definition How to decompose a billion-scale tensor? –Corresponds to SVD in 2D case LLNL, Jan 2013 72 C. Faloutsos (CMU)

73. CMU SCS Problem Definition  Q1: Dominant concepts/topics?  Q2: Find synonyms to a given noun phrase?  (and how to scale up: |data| > RAM) (26M) (48M) NELL (Never Ending Language Learner) data Nonzeros =144M LLNL, Jan 2013 73 C. Faloutsos (CMU)

74. CMU SCS Experiments GigaTensor solves 100x larger problem Number of nonzero = I / 50 (J) (I) (K) GigaTensor Tensor Toolbox Out of Memory 100x LLNL, Jan 2013 74 C. Faloutsos (CMU)

75. CMU SCS A1: Concept Discovery Concept Discovery in Knowledge Base LLNL, Jan 2013 75 C. Faloutsos (CMU)

76. CMU SCS A1: Concept Discovery LLNL, Jan 2013 76 C. Faloutsos (CMU)

77. CMU SCS A2: Synonym Discovery LLNL, Jan 2013 77 C. Faloutsos (CMU)

78. CMU SCS C. Faloutsos (CMU) 78 Roadmap Introduction – Motivation Problem#1: Patterns in graphs Problem#2: Tools –Belief propagation –Tensors –Spike analysis Problem#3: Scalability -PEGASUS Conclusions LLNL, Jan 2013

79. CMU SCS Meme (# of mentions in blogs) –short phrases Sourced from U.S. politics in 2008 79 “you can put lipstick on a pig” “yes we can” Rise and fall patterns in social media C. Faloutsos (CMU)LLNL, Jan 2013

80. CMU SCS Rise and fall patterns in social media 80 Can we find a unifying model, which includes these patterns? four classes on YouTube [Crane et al. ’08] six classes on Meme [Yang et al. ’11] C. Faloutsos (CMU)LLNL, Jan 2013

81. CMU SCS Rise and fall patterns in social media 81 Answer: YES! We can represent all patterns by single model C. Faloutsos (CMU)LLNL, Jan 2013 In Matsubara+ SIGKDD 2012

82. CMU SCS 82 Main idea - SpikeM -1. Un-informed bloggers (uninformed about rumor) -2. External shock at time n b (e.g, breaking news) -3. Infection (word-of-mouth) Time n=0Time n=n b β C. Faloutsos (CMU)LLNL, Jan 2013 Infectiveness of a blog-post at age n: -Strength of infection (quality of news) -Decay function Time n=n b +1

83. CMU SCS 83 -1. Un-informed bloggers (uninformed about rumor) -2. External shock at time n b (e.g, breaking news) -3. Infection (word-of-mouth) Time n=0Time n=n b β C. Faloutsos (CMU)LLNL, Jan 2013 Infectiveness of a blog-post at age n: -Strength of infection (quality of news) -Decay function Time n=n b +1 Main idea - SpikeM

84. CMU SCS SpikeM - with periodicity Full equation of SpikeM 84 Periodicity noon Peak 3am Dip Time n Bloggers change their activity over time (e.g., daily, weekly, yearly) Bloggers change their activity over time (e.g., daily, weekly, yearly) activity C. Faloutsos (CMU)LLNL, Jan 2013

85. CMU SCS Details Analysis – exponential rise and power-raw fall 85 Lin-log Log-log Rise-part SI -> exponential SpikeM -> exponential Rise-part SI -> exponential SpikeM -> exponential C. Faloutsos (CMU)LLNL, Jan 2013

86. CMU SCS Details Analysis – exponential rise and power-raw fall 86 Lin-log Log-log Fall-part SI -> exponential SpikeM -> power law Fall-part SI -> exponential SpikeM -> power law C. Faloutsos (CMU)LLNL, Jan 2013

87. CMU SCS Tail-part forecasts 87 SpikeM can capture tail part C. Faloutsos (CMU)LLNL, Jan 2013

88. CMU SCS “What-if” forecasting 88 e.g., given (1) first spike, (2) release date of two sequel movies (3) access volume before the release date ? ? (1) First spike (2) Release date (3) Two weeks before release C. Faloutsos (CMU)LLNL, Jan 2013 ? ?

89. CMU SCS “What-if” forecasting 89 SpikeM can forecast upcoming spikes (1) First spike (2) Release date (3) Two weeks before release C. Faloutsos (CMU)LLNL, Jan 2013

90. CMU SCS C. Faloutsos (CMU) 90 Roadmap Introduction – Motivation Problem#1: Patterns in graphs Problem#2: Tools Problem#3: Scalability –PEGASUS –Diameter –Connected components Conclusions LLNL, Jan 2013

91. CMU SCS LLNL, Jan 2013C. Faloutsos (CMU) 91 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/

92. CMU SCS C. Faloutsos (CMU) 92 CentralizedHadoop/PEG ASUS Degree Distr. old Pagerank old Diameter/ANF old HERE Conn. Comp old HERE TrianglesdoneHERE Visualizationstarted Roadmap – Algorithms & results LLNL, Jan 2013

93. 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. Faloutsos (CMU) 93 LLNL, Jan 2013

94. CMU SCS ???? 19+ [Barabasi+] 94 C. Faloutsos (CMU) Radius Count LLNL, Jan 2013 ~1999, ~1M nodes

95. CMU SCS YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) Largest publicly available graph ever studied. ???? 19+ [Barabasi+] 95 C. Faloutsos (CMU) Radius Count LLNL, Jan 2013 ?? ~1999, ~1M nodes

96. CMU SCS YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) Largest publicly available graph ever studied. ???? 19+? [Barabasi+] 96 C. Faloutsos (CMU) Radius Count LLNL, Jan 2013 14 (dir.) ~7 (undir.)

97. CMU SCS YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) 7 degrees of separation (!) Diameter: shrunk ???? 19+? [Barabasi+] 97 C. Faloutsos (CMU) Radius Count LLNL, Jan 2013 14 (dir.) ~7 (undir.)

98. CMU SCS YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) Q: Shape? ???? 98 C. Faloutsos (CMU) Radius Count LLNL, Jan 2013 ~7 (undir.)

99. CMU SCS 99 C. Faloutsos (CMU) YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) effective diameter: surprisingly small. Multi-modality (?!) LLNL, Jan 2013

100. CMU SCS Radius Plot of GCC of YahooWeb. 100 C. Faloutsos (CMU)LLNL, Jan 2013

101. CMU SCS 101 C. Faloutsos (CMU) YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) effective diameter: surprisingly small. Multi-modality: probably mixture of cores. LLNL, Jan 2013

102. CMU SCS 102 C. Faloutsos (CMU) YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) effective diameter: surprisingly small. Multi-modality: probably mixture of cores. LLNL, Jan 2013 EN ~7 Conjecture: DE BR

103. CMU SCS 103 C. Faloutsos (CMU) YahooWeb graph (120Gb, 1.4B nodes, 6.6 B edges) effective diameter: surprisingly small. Multi-modality: probably mixture of cores. LLNL, Jan 2013 ~7 Conjecture:

104. CMU SCS Running time - Kronecker and Erdos-Renyi Graphs with billions edges. details

105. CMU SCS C. Faloutsos (CMU) 105 Roadmap Introduction – Motivation Problem#1: Patterns in graphs Problem#2: Tools Problem#3: Scalability –PEGASUS –Diameter –Connected components Conclusions LLNL, Jan 2013

106. CMU SCS Generalized Iterated Matrix Vector Multiplication (GIMV) C. Faloutsos (CMU) 106 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 LLNL, Jan 2013

107. CMU SCS Generalized Iterated Matrix Vector Multiplication (GIMV) C. Faloutsos (CMU) 107 PageRank proximity (RWR) Diameter Connected components (eigenvectors, Belief Prop. … ) Matrix – vector Multiplication (iterated) LLNL, Jan 2013 details

108. CMU SCS 108 Example: GIM-V At Work Connected Components – 4 observations: Size Count C. Faloutsos (CMU) LLNL, Jan 2013

109. CMU SCS 109 Example: GIM-V At Work Connected Components Size Count C. Faloutsos (CMU) LLNL, Jan 2013 1) 10K x larger than next

110. CMU SCS 110 Example: GIM-V At Work Connected Components Size Count C. Faloutsos (CMU) LLNL, Jan 2013 2) ~0.7B singleton nodes

111. CMU SCS 111 Example: GIM-V At Work Connected Components Size Count C. Faloutsos (CMU) LLNL, Jan 2013 3) SLOPE!

112. CMU SCS 112 Example: GIM-V At Work Connected Components Size Count 300-size cmpt X 500. Why? 1100-size cmpt X 65. Why? C. Faloutsos (CMU) LLNL, Jan 2013 4) Spikes!

113. CMU SCS 113 Example: GIM-V At Work Connected Components Size Count suspicious financial-advice sites (not existing now) C. Faloutsos (CMU) LLNL, Jan 2013

114. CMU SCS 114 GIM-V At Work Connected Components over Time LinkedIn: 7.5M nodes and 58M edges Stable tail slope after the gelling point C. Faloutsos (CMU)LLNL, Jan 2013

115. CMU SCS C. Faloutsos (CMU) 115 Roadmap Introduction – Motivation Problem#1: Patterns in graphs Problem#2: Tools Problem#3: Scalability Conclusions LLNL, Jan 2013

116. CMU SCS C. Faloutsos (CMU) 116 OVERALL CONCLUSIONS – low level: Several new patterns (fortification, triangle-laws, conn. components, etc) New tools: –belief propagation, gigaTensor, etc Scalability: PEGASUS / hadoop LLNL, Jan 2013

117. CMU SCS C. Faloutsos (CMU) 117 OVERALL CONCLUSIONS – high level BIG DATA: Large datasets reveal patterns/outliers that are invisible otherwise LLNL, Jan 2013

118. CMU SCS ML & Stats. Comp. Systems Theory & Algo. Biology Econ. Social Science Physics 118 Graph Analytics LLNL, Jan 2013C. Faloutsos (CMU)

119. CMU SCS C. Faloutsos (CMU) 119 References Leman Akoglu, Christos Faloutsos: RTG: A Recursive Realistic Graph Generator Using Random Typing. ECML/PKDD (1) 2009: 13-28 Deepayan Chakrabarti, Christos Faloutsos: Graph mining: Laws, generators, and algorithms. ACM Comput. Surv. 38(1): (2006) LLNL, Jan 2013

120. CMU SCS C. Faloutsos (CMU) 120 References Deepayan Chakrabarti, Yang Wang, Chenxi Wang, Jure Leskovec, Christos Faloutsos: Epidemic thresholds in real networks. ACM Trans. Inf. Syst. Secur. 10(4): (2008) Deepayan Chakrabarti, Jure Leskovec, Christos Faloutsos, Samuel Madden, Carlos Guestrin, Michalis Faloutsos: Information Survival Threshold in Sensor and P2P Networks. INFOCOM 2007: 1316-1324 LLNL, Jan 2013

121. CMU SCS C. Faloutsos (CMU) 121 References Christos Faloutsos, Tamara G. Kolda, Jimeng Sun: Mining large graphs and streams using matrix and tensor tools. Tutorial, SIGMOD Conference 2007: 1174 LLNL, Jan 2013

122. CMU SCS C. Faloutsos (CMU) 122 References T. G. Kolda and J. Sun. Scalable Tensor Decompositions for Multi-aspect Data Mining. In: ICDM 2008, pp. 363-372, December 2008. LLNL, Jan 2013

123. CMU SCS C. Faloutsos (CMU) 123 References Jure Leskovec, Jon Kleinberg and Christos Faloutsos Graphs over Time: Densification Laws, Shrinking Diameters and Possible Explanations, KDD 2005 (Best Research paper award). Jure Leskovec, Deepayan Chakrabarti, Jon M. Kleinberg, Christos Faloutsos: Realistic, Mathematically Tractable Graph Generation and Evolution, Using Kronecker Multiplication. PKDD 2005: 133-145 LLNL, Jan 2013

124. CMU SCS References Yasuko Matsubara, Yasushi Sakurai, B. Aditya Prakash, Lei Li, Christos Faloutsos, "Rise and Fall Patterns of Information Diffusion: Model and Implications", KDD’12, pp. 6-14, Beijing, China, August 2012 LLNL, Jan 2013C. Faloutsos (CMU) 124

125. CMU SCS C. Faloutsos (CMU) 125 References Jimeng Sun, Yinglian Xie, Hui Zhang, Christos Faloutsos. Less is More: Compact Matrix Decomposition for Large Sparse Graphs, SDM, Minneapolis, Minnesota, Apr 2007. Jimeng Sun, Spiros Papadimitriou, Philip S. Yu, and Christos Faloutsos, GraphScope: Parameter- free Mining of Large Time-evolving Graphs ACM SIGKDD Conference, San Jose, CA, August 2007 LLNL, Jan 2013

126. CMU SCS References Jimeng Sun, Dacheng Tao, Christos Faloutsos: Beyond streams and graphs: dynamic tensor analysis. KDD 2006: 374- 383 LLNL, Jan 2013C. Faloutsos (CMU) 126

127. CMU SCS C. Faloutsos (CMU) 127 References Hanghang Tong, Christos Faloutsos, and Jia-Yu Pan, Fast Random Walk with Restart and Its Applications, ICDM 2006, Hong Kong. Hanghang Tong, Christos Faloutsos, Center-Piece Subgraphs: Problem Definition and Fast Solutions, KDD 2006, Philadelphia, PA LLNL, Jan 2013

128. CMU SCS C. Faloutsos (CMU) 128 References Hanghang Tong, Christos Faloutsos, Brian Gallagher, Tina Eliassi-Rad: Fast best-effort pattern matching in large attributed graphs. KDD 2007: 737-746 LLNL, Jan 2013

129. CMU SCS C. Faloutsos (CMU) 129 Project info & ‘thanks’ LLNL, Jan 2013 Thanks to: NSF IIS-0705359, IIS-0534205, CTA-INARC ; Yahoo (M45), LLNL, IBM, SPRINT, Google, INTEL, HP, iLab www.cs.cmu.edu/~pegasus

130. CMU SCS C. Faloutsos (CMU) 130 Cast Akoglu, Leman Chau, Polo Kang, U McGlohon, Mary Tong, Hanghang Prakash, Aditya LLNL, Jan 2013 Koutra, Danae Beutel, Alex Papalexakis, Vagelis

131. CMU SCS C. Faloutsos (CMU) 131 Thanks to LLNL colleagues LLNL, Jan 2013 Brian Gallagher Keith Henderson

132. CMU SCS Take-home message LLNL, Jan 2013C. Faloutsos (CMU) 132 Tera/Peta-byte data AnalyticsInsights, outliers Big data reveal insights that would be invisible otherwise (even to experts)