large graph mining - research.ece.cmu.edu · cmu scs large graph mining - patterns, explanations...
TRANSCRIPT
![Page 1: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/1.jpg)
CMU SCS
Large Graph Mining - Patterns, Explanations and
Cascade Analysis Christos Faloutsos
CMU
![Page 2: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/2.jpg)
CMU SCS
(c) 2014, C. Faloutsos 2
Roadmap
• A case for cross-disciplinarity • Introduction – Motivation
– Why study (big) graphs?
• Part#1: Patterns in graphs • Part#2: Cascade analysis • Conclusions
CMU, Feb 2014
![Page 3: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/3.jpg)
CMU SCS
Speed up of Data-Driven State Estimation Using Low-Complexity Indexing Method
Historical Similar Data Measurements, States
Time Consuming
Data-Driven State Estimation
Observation: Redundancies & correlations
![Page 4: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/4.jpg)
CMU SCS
Problem dfn
CMU, Feb 2014 (c) 2014, C. Faloutsos 4
time
Measurement 1
Measurement M Voltage 1 Voltage N
?
~1000
3yrs, every 5’
![Page 5: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/5.jpg)
CMU SCS
Problem dfn
CMU, Feb 2014 (c) 2014, C. Faloutsos 5
time
Measurement 1
Measurement M Voltage 1 Voltage N
?
Direct solution: Slow (Kirchoff’s eq.)
![Page 6: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/6.jpg)
CMU SCS
Problem dfn
CMU, Feb 2014 (c) 2014, C. Faloutsos 6
time
Measurement 1
Measurement M Voltage 1 Voltage N
?
Look for near-neighbors And use *their* voltages
![Page 7: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/7.jpg)
CMU SCS
Problem dfn
CMU, Feb 2014 (c) 2014, C. Faloutsos 7
time
Measurement 1
Measurement M Voltage 1 Voltage N
?
Look for near-neighbors And use *their* voltages
But sequential scan Is slow, too (MxT) Can we do better?
T
✔ ✔
![Page 8: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/8.jpg)
CMU SCS
Problem dfn
CMU, Feb 2014 (c) 2014, C. Faloutsos 8
time
Measurement 1
Measurement M Voltage 1 Voltage N
?
But sequential scan Is slow, too (MxT) Can we do better?
T
A: yes! We can reduce both • T, and • M
![Page 9: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/9.jpg)
CMU SCS
• Same accuracy, 100x – 100K x faster
Many simulations
Relative Search Time:
Simulation Results
1000 x
CMU, Feb 2014 9 (c) 2014, C. Faloutsos [1] Yang Weng, Christos Faloutsos, Marija D. Ili´c, and Rohit Negi, Speed up of Data-Driven State Estimation Using Low-Complexity Indexing Method, IEEE PES-General Meeting, (accepted), 2014
1 sec vs 15’ vs 1 day
![Page 10: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/10.jpg)
CMU SCS
Step1: Reducing dimensionality M
CMU, Feb 2014 (c) 2014, C. Faloutsos 10
time
Measurement 1
Measurement M
But sequential scan Is slow, too (MxT) Can we do better?
A: yes! We can reduce both • T, and • M SVD
![Page 11: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/11.jpg)
CMU SCS
Step2: Faster than T timeticks
CMU, Feb 2014 (c) 2014, C. Faloutsos 11
time
Measurement 1
Measurement M
But sequential scan Is slow, too (MxT) Can we do better?
A: yes! We can reduce both • T, and • M SVD
K-d trees
![Page 12: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/12.jpg)
CMU SCS
Faster than seq. scan: K-d trees
CMU, Feb 2014 12 (c) 2014, C. Faloutsos
★
![Page 13: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/13.jpg)
CMU SCS
Projection of measurements on to singular vectors of measurement matrix
CMU, Feb 2014 13 (c) 2014, C. Faloutsos [1] Yang Weng, Christos Faloutsos, Marija D. Ili´c, and Rohit Negi, Speed up of Data-Driven State Estimation Using Low-Complexity Indexing Method, IEEE PES-General Meeting, (accepted), 2014
Thanks to SVD: VISUALIZATION!
![Page 14: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/14.jpg)
CMU SCS
Projection of measurements on to singular vectors of measurement matrix
CMU, Feb 2014 14 (c) 2014, C. Faloutsos [1] Yang Weng, Christos Faloutsos, Marija D. Ili´c, and Rohit Negi, Speed up of Data-Driven State Estimation Using Low-Complexity Indexing Method, IEEE PES-General Meeting, (accepted), 2014
Thanks to SVD: VISUALIZATION!
4 (or 5) groups of behavior!
![Page 15: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/15.jpg)
CMU SCS
Projection of measurements on to singular vectors of measurement matrix
CMU, Feb 2014 15 (c) 2014, C. Faloutsos [1] Yang Weng, Christos Faloutsos, Marija D. Ili´c, and Rohit Negi, Speed up of Data-Driven State Estimation Using Low-Complexity Indexing Method, IEEE PES-General Meeting, (accepted), 2014
Thanks to SVD: VISUALIZATION!
![Page 16: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/16.jpg)
CMU SCS
• Same accuracy, 100x – 100K x faster
Crossdisciplinarity: Already started paying off
1000 x faster
CMU, Feb 2014 16 (c) 2014, C. Faloutsos
[1] Yang Weng, Christos Faloutsos, Marija D. Ili´c, and Rohit Negi, Speed up of Data-Driven State Estimation Using Low-Complexity Indexing Method, IEEE PES-General Meeting,
(accepted), 2014
![Page 17: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/17.jpg)
CMU SCS
(c) 2014, C. Faloutsos 17
Roadmap
• A case for cross-disciplinarity • Introduction – Motivation
– Why study (big) graphs?
• Part#1: Patterns in graphs • Part#2: Cascade analysis • Conclusions
CMU, Feb 2014
![Page 18: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/18.jpg)
CMU SCS
(c) 2014, C. Faloutsos 18
Graphs - why should we care?
CMU, Feb 2014
• Power-grid! – Nodes: (plants/
consumers) – Edges: power lines
![Page 19: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/19.jpg)
CMU SCS
(c) 2014, C. Faloutsos 19
Graphs - why should we care?
Internet Map [lumeta.com]
Food Web [Martinez ’91]
>$10B revenue
>0.5B users
CMU, Feb 2014
![Page 20: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/20.jpg)
CMU SCS
(c) 2014, C. Faloutsos 20
Graphs - why should we care? • web-log (‘blog’) news propagation • computer network security: email/IP traffic and
anomaly detection • Recommendation systems • ....
• Many-to-many db relationship -> graph
CMU, Feb 2014
![Page 21: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/21.jpg)
CMU SCS
(c) 2014, C. Faloutsos 21
Roadmap
• A case for cross-disciplinarity • Introduction – Motivation
– Why study (big) graphs?
• Part#1: Patterns in graphs • Part#2: Cascade analysis • Conclusions
CMU, Feb 2014
![Page 22: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/22.jpg)
CMU SCS
CMU, Feb 2014 (c) 2014, C. Faloutsos 22
Part 1: Patterns & Laws
![Page 23: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/23.jpg)
CMU SCS
(c) 2014, C. Faloutsos 23
Laws and patterns • Q1: Are real graphs random?
CMU, Feb 2014
![Page 24: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/24.jpg)
CMU SCS
(c) 2014, C. Faloutsos 24
Laws and patterns • Q1: Are real graphs random? • A1: NO!!
– Diameter – in- and out- degree distributions – other (surprising) patterns
• Q2: why so many power laws? • A2: <self-similarity – stay tuned>
• So, let’s look at the data CMU, Feb 2014
![Page 25: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/25.jpg)
CMU SCS
(c) 2014, C. Faloutsos 25
Solution# S.1 • Power law in the degree distribution
[SIGCOMM99]
log(rank)
log(degree)
internet domains
att.com
ibm.com
CMU, Feb 2014
![Page 26: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/26.jpg)
CMU SCS
(c) 2014, C. Faloutsos 26
Solution# S.1 • Power law in the degree distribution
[SIGCOMM99]
log(rank)
log(degree)
-0.82
internet domains
att.com
ibm.com
CMU, Feb 2014
![Page 27: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/27.jpg)
CMU SCS
(c) 2014, C. Faloutsos 27
Solution# S.1 • Q: So what?
log(rank)
log(degree)
-0.82
internet domains
att.com
ibm.com
CMU, Feb 2014
![Page 28: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/28.jpg)
CMU SCS
(c) 2014, C. Faloutsos 28
Solution# S.1 • Q: So what? • A1: # of two-step-away pairs:
log(rank)
log(degree)
-0.82
internet domains
att.com
ibm.com
CMU, Feb 2014
= friends of friends (F.O.F.)
![Page 29: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/29.jpg)
CMU SCS
(c) 2014, C. Faloutsos 29
Solution# S.1 • Q: So what? • A1: # of two-step-away pairs: O(d_max ^2) ~ 10M^2
log(rank)
log(degree)
-0.82
internet domains
att.com
ibm.com
CMU, Feb 2014
~0.8PB -> a data center(!)
DCO @ CMU
Gaussian trap
= friends of friends (F.O.F.)
![Page 30: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/30.jpg)
CMU SCS
(c) 2014, C. Faloutsos 30
Solution# S.1 • Q: So what? • A1: # of two-step-away pairs: O(d_max ^2) ~ 10M^2
log(rank)
log(degree)
-0.82
internet domains
att.com
ibm.com
CMU, Feb 2014
~0.8PB -> a data center(!)
Gaussian trap
![Page 31: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/31.jpg)
CMU SCS
(c) 2014, C. Faloutsos 31
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
CMU, Feb 2014
A x = λ x
![Page 32: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/32.jpg)
CMU SCS
(c) 2014, C. Faloutsos 32
Roadmap
• Introduction – Motivation • Problem#1: Patterns in graphs
– Static graphs • degree, diameter, eigen, • Triangles
– Time evolving graphs • Problem#2: Tools
CMU, Feb 2014
![Page 33: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/33.jpg)
CMU SCS
(c) 2014, C. Faloutsos 33
Solution# S.3: Triangle ‘Laws’
• Real social networks have a lot of triangles
CMU, Feb 2014
![Page 34: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/34.jpg)
CMU SCS
(c) 2014, C. Faloutsos 34
Solution# S.3: Triangle ‘Laws’
• Real social networks have a lot of triangles – Friends of friends are friends
• Any patterns? – 2x the friends, 2x the triangles ?
CMU, Feb 2014
![Page 35: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/35.jpg)
CMU SCS
(c) 2014, C. Faloutsos 35
Triangle Law: #S.3 [Tsourakakis ICDM 2008]
SN Reuters
Epinions X-axis: degree Y-axis: mean # triangles n friends -> ~n1.6 triangles
CMU, Feb 2014
![Page 36: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/36.jpg)
CMU SCS
(c) 2014, C. Faloutsos 36
Triangle Law: Computations [Tsourakakis ICDM 2008]
But: triangles are expensive to compute (3-way join; several approx. algos) – O(dmax
2) Q: Can we do that quickly? A:
details
CMU, Feb 2014
![Page 37: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/37.jpg)
CMU SCS
(c) 2014, C. Faloutsos 37
Triangle Law: Computations [Tsourakakis ICDM 2008]
But: triangles are expensive to compute (3-way join; several approx. algos) – O(dmax
2) Q: Can we do that quickly? A: Yes!
#triangles = 1/6 Sum ( λi3 )
(and, because of skewness (S2) , we only need the top few eigenvalues! - O(E)
details
CMU, Feb 2014
A x = λ x
![Page 38: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/38.jpg)
CMU SCS
Triangle counting for large graphs?
Anomalous nodes in Twitter(~ 3 billion edges) [U Kang, Brendan Meeder, +, PAKDD’11]
38 CMU, Feb 2014 38 (c) 2014, C. Faloutsos
? ?
?
![Page 39: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/39.jpg)
CMU SCS
Triangle counting for large graphs?
Anomalous nodes in Twitter(~ 3 billion edges) [U Kang, Brendan Meeder, +, PAKDD’11]
39 CMU, Feb 2014 39 (c) 2014, C. Faloutsos
![Page 40: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/40.jpg)
CMU SCS
Triangle counting for large graphs?
Anomalous nodes in Twitter(~ 3 billion edges) [U Kang, Brendan Meeder, +, PAKDD’11]
40 CMU, Feb 2014 40 (c) 2014, C. Faloutsos
![Page 41: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/41.jpg)
CMU SCS
Triangle counting for large graphs?
Anomalous nodes in Twitter(~ 3 billion edges) [U Kang, Brendan Meeder, +, PAKDD’11]
41 CMU, Feb 2014 41 (c) 2014, C. Faloutsos
![Page 42: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/42.jpg)
CMU SCS
(c) 2014, C. Faloutsos 42
Roadmap
• A case for cross-disciplinarity • Introduction – Motivation • Part#1: Patterns in graphs
– Static graphs – Time evolving graphs
• Part#2: Cascade analysis • Conclusions
CMU, Feb 2014
![Page 43: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/43.jpg)
CMU SCS
(c) 2014, C. Faloutsos 43
Problem: Time evolution • with Jure Leskovec (CMU ->
Stanford)
• and Jon Kleinberg (Cornell – sabb. @ CMU)
CMU, Feb 2014
Jure Leskovec, Jon Kleinberg and Christos Faloutsos: Graphs over Time: Densification Laws, Shrinking Diameters and Possible Explanations, KDD 2005
![Page 44: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/44.jpg)
CMU SCS
(c) 2014, C. Faloutsos 44
T.1 Evolution of the Diameter • Prior work on Power Law graphs hints
at slowly growing diameter: – [diameter ~ O( N1/3)] – diameter ~ O(log N) – diameter ~ O(log log N)
• What is happening in real data?
CMU, Feb 2014
diameter
![Page 45: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/45.jpg)
CMU SCS
(c) 2014, C. Faloutsos 45
T.1 Evolution of the Diameter • Prior work on Power Law graphs hints
at slowly growing diameter: – [diameter ~ O( N1/3)] – diameter ~ O(log N) – diameter ~ O(log log N)
• What is happening in real data? • Diameter shrinks over time
CMU, Feb 2014
![Page 46: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/46.jpg)
CMU SCS
(c) 2014, C. Faloutsos 46
T.1 Diameter – “Patents”
• Patent citation network
• 25 years of data • @1999
– 2.9 M nodes – 16.5 M edges
time [years]
diameter
CMU, Feb 2014
![Page 47: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/47.jpg)
CMU SCS
(c) 2014, C. Faloutsos 47
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)
CMU, Feb 2014
Say, k friends on average
k
![Page 48: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/48.jpg)
CMU SCS
(c) 2014, C. Faloutsos 48
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! ~ 3x
– But obeying the ``Densification Power Law’’ CMU, Feb 2014
Say, k friends on average
Gaussian trap
![Page 49: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/49.jpg)
CMU SCS
(c) 2014, C. Faloutsos 49
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! ~ 3x
– But obeying the ``Densification Power Law’’ CMU, Feb 2014
Say, k friends on average
Gaussian trap
✗ ✔ log
log
lin
lin
![Page 50: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/50.jpg)
CMU SCS
(c) 2014, C. Faloutsos 50
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
CMU, Feb 2014
![Page 51: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/51.jpg)
CMU SCS
MORE Graph Patterns
CMU, Feb 2014 (c) 2014, C. Faloutsos 51 RTG: A Recursive Realistic Graph Generator using Random Typing Leman Akoglu and Christos Faloutsos. PKDD’09.
![Page 52: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/52.jpg)
CMU SCS
MORE Graph Patterns
CMU, Feb 2014 (c) 2014, C. Faloutsos 52
✔
✔ ✔ ✔
✔
RTG: A Recursive Realistic Graph Generator using Random Typing Leman Akoglu and Christos Faloutsos. PKDD’09.
![Page 53: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/53.jpg)
CMU SCS
MORE Graph Patterns
CMU, Feb 2014 (c) 2014, C. Faloutsos 53
• Mary McGlohon, Leman Akoglu, Christos Faloutsos. Statistical Properties of Social Networks. in "Social Network Data Analytics” (Ed.: Charu Aggarwal)
• Deepayan Chakrabarti and Christos Faloutsos, Graph Mining: Laws, Tools, and Case Studies Oct. 2012, Morgan Claypool.
![Page 54: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/54.jpg)
CMU SCS
(c) 2014, C. Faloutsos 54
Roadmap
• A case for cross-disciplinarity • Introduction – Motivation • Part#1: Patterns in graphs
– … – Why so many power-laws?
• Part#2: Cascade analysis • Conclusions
CMU, Feb 2014
SKIP
![Page 55: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/55.jpg)
CMU SCS
Why so many P.L.? • Possible answer: self-similarity / fractals
CMU, Feb 2014 (c) 2014, C. Faloutsos 55
SKIP
![Page 56: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/56.jpg)
CMU SCS
20’’ intro to fractals • Remove the middle triangle; repeat • -> Sierpinski triangle • (Bonus question - dimensionality?
– >1 (inf. perimeter – (4/3)∞ ) – <2 (zero area – (3/4) ∞ )
CMU, Feb 2014 (c) 2014, C. Faloutsos 56
…
SKIP
![Page 57: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/57.jpg)
CMU SCS
20’’ intro to fractals
CMU, Feb 2014 (c) 2014, C. Faloutsos 57
Self-similarity -> no char. scale -> power laws, eg: 2x the radius, 3x the #neighbors nn(r) nn(r) = C r log3/log2
SKIP
![Page 58: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/58.jpg)
CMU SCS
20’’ intro to fractals
CMU, Feb 2014 (c) 2014, C. Faloutsos 58
Self-similarity -> no char. scale -> power laws, eg: 2x the radius, 3x the #neighbors nn(r) nn(r) = C r log3/log2
SKIP
![Page 59: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/59.jpg)
CMU SCS
20’’ intro to fractals
CMU, Feb 2014 (c) 2014, C. Faloutsos 59
Self-similarity -> no char. scale -> power laws, eg: 2x the radius, 3x the #neighbors nn = C r log3/log2
N(t)
E(t)
1.66
Reminder: Densification P.L. (2x nodes, ~3x edges)
SKIP
![Page 60: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/60.jpg)
CMU SCS
20’’ intro to fractals
CMU, Feb 2014 (c) 2014, C. Faloutsos 60
Self-similarity -> no char. scale -> power laws, eg: 2x the radius, 3x the #neighbors nn = C r log3/log2
2x the radius, 4x neighbors
nn = C r log4/log2 = C r 2
SKIP
![Page 61: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/61.jpg)
CMU SCS
20’’ intro to fractals
CMU, Feb 2014 (c) 2014, C. Faloutsos 61
Self-similarity -> no char. scale -> power laws, eg: 2x the radius, 3x the #neighbors nn = C r log3/log2
2x the radius, 4x neighbors
nn = C r log4/log2 = C r 2
Fractal dim.
=1.58
SKIP
![Page 62: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/62.jpg)
CMU SCS
20’’ intro to fractals
CMU, Feb 2014 (c) 2014, C. Faloutsos 62
Self-similarity -> no char. scale -> power laws, eg: 2x the radius, 3x the #neighbors nn = C r log3/log2
2x the radius, 4x neighbors
nn = C r log4/log2 = C r 2
Fractal dim.
SKIP
![Page 63: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/63.jpg)
CMU SCS How does self-similarity help in
graphs? • A: RMAT/Kronecker generators
– With self-similarity, we get all power-laws, automatically,
– And small/shrinking diameter – And `no good cuts’
CMU, Feb 2014 (c) 2014, C. Faloutsos 63
R-MAT: A Recursive Model for Graph Mining, by D. Chakrabarti, Y. Zhan and C. Faloutsos, SDM 2004, Orlando, Florida, USA Realistic, Mathematically Tractable Graph Generation and Evolution, Using Kronecker Multiplication, by J. Leskovec, D. Chakrabarti, J. Kleinberg, and C. Faloutsos, in PKDD 2005, Porto, Portugal
SKIP
![Page 64: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/64.jpg)
CMU SCS
CMU, Feb 2014 (c) 2014, C. Faloutsos 64
Graph gen.: Problem dfn • Given a growing graph with count of nodes N1,
N2, … • Generate a realistic sequence of graphs that will
obey all the patterns – Static Patterns
S1 Power Law Degree Distribution S2 Power Law eigenvalue and eigenvector distribution Small Diameter
– Dynamic Patterns T2 Growth Power Law (2x nodes; 3x edges) T1 Shrinking/Stabilizing Diameters
SKIP
![Page 65: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/65.jpg)
CMU SCS
CMU, Feb 2014 (c) 2014, C. Faloutsos 65 Adjacency matrix
Kronecker Graphs
Intermediate stage
Adjacency matrix
SKIP
![Page 66: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/66.jpg)
CMU SCS
CMU, Feb 2014 (c) 2014, C. Faloutsos 66 Adjacency matrix
Kronecker Graphs
Intermediate stage
Adjacency matrix
SKIP
![Page 67: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/67.jpg)
CMU SCS
CMU, Feb 2014 (c) 2014, C. Faloutsos 67 Adjacency matrix
Kronecker Graphs
Intermediate stage
Adjacency matrix
SKIP
![Page 68: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/68.jpg)
CMU SCS
CMU, Feb 2014 (c) 2014, C. Faloutsos 68
Kronecker Graphs • Continuing multiplying with G1 we obtain G4 and
so on …
G4 adjacency matrix
SKIP
![Page 69: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/69.jpg)
CMU SCS
CMU, Feb 2014 (c) 2014, C. Faloutsos 69
Kronecker Graphs • Continuing multiplying with G1 we obtain G4 and
so on …
G4 adjacency matrix
SKIP
![Page 70: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/70.jpg)
CMU SCS
CMU, Feb 2014 (c) 2014, C. Faloutsos 70
Kronecker Graphs • Continuing multiplying with G1 we obtain G4 and
so on …
G4 adjacency matrix
SKIP
![Page 71: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/71.jpg)
CMU SCS
CMU, Feb 2014 (c) 2014, C. Faloutsos 71
Kronecker Graphs • Continuing multiplying with G1 we obtain G4 and
so on …
G4 adjacency matrix
Holes within holes; Communities
within communities
SKIP
![Page 72: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/72.jpg)
CMU SCS
CMU, Feb 2014 (c) 2014, C. Faloutsos 72
Properties:
• We can PROVE that – Degree distribution is multinomial ~ power law – Diameter: constant – Eigenvalue distribution: multinomial – First eigenvector: multinomial
new
Self-similarity -> power laws SKIP
![Page 73: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/73.jpg)
CMU SCS
CMU, Feb 2014 (c) 2014, C. Faloutsos 73
Problem Definition • Given a growing graph with nodes N1, N2, … • Generate a realistic sequence of graphs that will obey all
the patterns – Static Patterns
Power Law Degree Distribution Power Law eigenvalue and eigenvector distribution Small Diameter
– Dynamic Patterns Growth Power Law Shrinking/Stabilizing Diameters
• First generator for which we can prove all these properties
SKIP
![Page 74: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/74.jpg)
CMU SCS
Impact: Graph500 • Based on RMAT (= 2x2 Kronecker) • Standard for graph benchmarks • http://www.graph500.org/ • Competitions 2x year, with all major
entities: LLNL, Argonne, ITC-U. Tokyo, Riken, ORNL, Sandia, PSC, …
CMU, Feb 2014 (c) 2014, C. Faloutsos 74
R-MAT: A Recursive Model for Graph Mining, by D. Chakrabarti, Y. Zhan and C. Faloutsos, SDM 2004, Orlando, Florida, USA
To iterate is human, to recurse is devine
SKIP
![Page 75: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/75.jpg)
CMU SCS
Summary of Part#1 • *many* patterns in real graphs
– Small & shrinking diameters – Power-laws everywhere – Gaussian trap
• Self-similarity (RMAT/Kronecker): good model
CMU, Feb 2014 (c) 2014, C. Faloutsos 75
![Page 76: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/76.jpg)
CMU SCS
(c) 2014, C. Faloutsos 76
Roadmap
• A case for cross-disciplinarity • Introduction – Motivation • Part#1: Patterns in graphs • Part#2: Cascade analysis • Conclusions
CMU, Feb 2014
![Page 77: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/77.jpg)
CMU SCS
Comic relief: • What would a barefooted man get if he
steps on an electric wire?
CMU, Feb 2014 (c) 2014, C. Faloutsos 77
http://energyquest.ca.gov/games/jokes/george.html
![Page 78: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/78.jpg)
CMU SCS
Comic relief: • What would a barefooted man get if he
steps on an electric wire? (Answer) A pair of shocks
CMU, Feb 2014 (c) 2014, C. Faloutsos 78
http://energyquest.ca.gov/games/jokes/george.html
![Page 79: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/79.jpg)
CMU SCS
CMU, Feb 2014 (c) 2014, C. Faloutsos 79
Part 2: Cascades &
Immunization
![Page 80: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/80.jpg)
CMU SCS
Why do we care? • Information Diffusion • Viral Marketing • Epidemiology and Public Health • Cyber Security • Human mobility • Games and Virtual Worlds • Ecology • ........
(c) 2014, C. Faloutsos 80 CMU, Feb 2014
![Page 81: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/81.jpg)
CMU SCS
(c) 2014, C. Faloutsos 81
Roadmap
• A case for cross-disciplinarity • Introduction – Motivation • Part#1: Patterns in graphs • Part#2: Cascade analysis
– (Fractional) Immunization – Epidemic thresholds
• Conclusions
CMU, Feb 2014
![Page 82: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/82.jpg)
CMU SCS
Fractional Immunization of Networks B. Aditya Prakash, Lada Adamic, Theodore Iwashyna (M.D.), Hanghang Tong,
Christos Faloutsos
SDM 2013, Austin, TX
(c) 2014, C. Faloutsos 82 CMU, Feb 2014
![Page 83: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/83.jpg)
CMU SCS
Whom to immunize? • Dynamical Processes over networks
• Each circle is a hospital • ~3,000 hospitals • More than 30,000 pa4ents transferred
[US-‐MEDICARE NETWORK 2005]
Problem: Given k units of disinfectant, whom to immunize?
(c) 2014, C. Faloutsos 83 CMU, Feb 2014
![Page 84: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/84.jpg)
CMU SCS
Whom to immunize?
CURRENT PRACTICE OUR METHOD
[US-‐MEDICARE NETWORK 2005]
~6x fewer!
(c) 2014, C. Faloutsos 84 CMU, Feb 2014
Hospital-acquired inf. : 99K+ lives, $5B+ per year
![Page 85: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/85.jpg)
CMU SCS
Fractional Asymmetric Immunization
Hospital Another Hospital
Drug-‐resistant Bacteria (like XDR-‐TB)
(c) 2014, C. Faloutsos 85 CMU, Feb 2014
![Page 86: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/86.jpg)
CMU SCS
Fractional Asymmetric Immunization
Hospital Another Hospital
(c) 2014, C. Faloutsos 86 CMU, Feb 2014
![Page 87: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/87.jpg)
CMU SCS
Fractional Asymmetric Immunization
Hospital Another Hospital
(c) 2014, C. Faloutsos 87 CMU, Feb 2014
![Page 88: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/88.jpg)
CMU SCS
Fractional Asymmetric Immunization
Hospital Another Hospital
Problem: Given k units of disinfectant, distribute them to maximize hospitals saved
(c) 2014, C. Faloutsos 88 CMU, Feb 2014
![Page 89: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/89.jpg)
CMU SCS
Fractional Asymmetric Immunization
Hospital Another Hospital
Problem: Given k units of disinfectant, distribute them to maximize hospitals saved @ 365 days
(c) 2014, C. Faloutsos 89 CMU, Feb 2014
![Page 90: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/90.jpg)
CMU SCS
Straightforward solution: Simulation: 1. Distribute resources 2. ‘infect’ a few nodes 3. Simulate evolution of spreading
– (10x, take avg) 4. Tweak, and repeat step 1
CMU, Feb 2014 (c) 2014, C. Faloutsos 90
![Page 91: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/91.jpg)
CMU SCS
Straightforward solution: Simulation: 1. Distribute resources 2. ‘infect’ a few nodes 3. Simulate evolution of spreading
– (10x, take avg) 4. Tweak, and repeat step 1
CMU, Feb 2014 (c) 2014, C. Faloutsos 91
![Page 92: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/92.jpg)
CMU SCS
Straightforward solution: Simulation: 1. Distribute resources 2. ‘infect’ a few nodes 3. Simulate evolution of spreading
– (10x, take avg) 4. Tweak, and repeat step 1
CMU, Feb 2014 (c) 2014, C. Faloutsos 92
![Page 93: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/93.jpg)
CMU SCS
Straightforward solution: Simulation: 1. Distribute resources 2. ‘infect’ a few nodes 3. Simulate evolution of spreading
– (10x, take avg) 4. Tweak, and repeat step 1
CMU, Feb 2014 (c) 2014, C. Faloutsos 93
![Page 94: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/94.jpg)
CMU SCS
Running Time
Simula4ons SMART-‐ALLOC
> 1 week Wall-‐Clock
Time ≈
14 secs
> 30,000x speed-‐up!
be?er
(c) 2014, C. Faloutsos 94 CMU, Feb 2014
![Page 95: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/95.jpg)
CMU SCS
Experiments
K = 120
be?er
(c) 2014, C. Faloutsos 95 CMU, Feb 2014
# epochs
# infected uniform
SMART-ALLOC
![Page 96: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/96.jpg)
CMU SCS
What is the ‘silver bullet’? A: Try to decrease connectivity of graph
Q: how to measure connectivity? – Avg degree? Max degree? – Std degree / avg degree ? – Diameter? – Modularity? – ‘Conductance’ (~min cut size)? – Some combination of above?
CMU, Feb 2014 (c) 2014, C. Faloutsos 96
14 secs
> 30,000x speed-‐up!
![Page 97: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/97.jpg)
CMU SCS
What is the ‘silver bullet’? A: Try to decrease connectivity of graph
Q: how to measure connectivity? A: first eigenvalue of adjacency matrix
Q1: why?? (Q2: dfn & intuition of eigenvalue ? )
CMU, Feb 2014 (c) 2014, C. Faloutsos 97
Avg degree Max degree Diameter Modularity ‘Conductance’
![Page 98: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/98.jpg)
CMU SCS
Why eigenvalue? A1: ‘G2’ theorem and ‘eigen-drop’:
• For (almost) any type of virus • For any network • -> no epidemic, if small-enough first
eigenvalue (λ1 ) of adjacency matrix
• Heuristic: for immunization, try to min λ1
• The smaller λ1, the closer to extinction. CMU, Feb 2014 (c) 2014, C. Faloutsos 98
Threshold Conditions for Arbitrary Cascade Models on Arbitrary Networks, B. Aditya Prakash, Deepayan Chakrabarti, Michalis Faloutsos, Nicholas Valler, Christos Faloutsos, ICDM 2011, Vancouver, Canada
![Page 99: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/99.jpg)
CMU SCS
Why eigenvalue? A1: ‘G2’ theorem and ‘eigen-drop’:
• For (almost) any type of virus • For any network • -> no epidemic, if small-enough first
eigenvalue (λ1 ) of adjacency matrix
• Heuristic: for immunization, try to min λ1
• The smaller λ1, the closer to extinction. CMU, Feb 2014 (c) 2014, C. Faloutsos 99
![Page 100: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/100.jpg)
CMU SCS
Threshold Conditions for Arbitrary Cascade Models on Arbitrary Networks B. Aditya Prakash, Deepayan Chakrabarti, Michalis Faloutsos, Nicholas Valler, Christos Faloutsos IEEE ICDM 2011, Vancouver
extended version, in arxiv http://arxiv.org/abs/1004.0060
G2 theorem
~10 pages proof
![Page 101: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/101.jpg)
CMU SCS
Our thresholds for some models • s = effective strength • s < 1 : below threshold
(c) 2014, C. Faloutsos 101 CMU, Feb 2014
Models Effective Strength (s)
Threshold (tipping point)
SIS, SIR, SIRS, SEIR s = λ .
s = 1 SIV, SEIV s = λ .
(H.I.V.) s = λ .
![Page 102: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/102.jpg)
CMU SCS
Our thresholds for some models • s = effective strength • s < 1 : below threshold
(c) 2014, C. Faloutsos 102 CMU, Feb 2014
Models Effective Strength (s)
Threshold (tipping point)
SIS, SIR, SIRS, SEIR s = λ .
s = 1 SIV, SEIV s = λ .
(H.I.V.) s = λ .
No immunity
Temp. immunity
w/ incubation
![Page 103: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/103.jpg)
CMU SCS
(c) 2014, C. Faloutsos 103
Roadmap
• Introduction – Motivation • Part#1: Patterns in graphs • Part#2: Cascade analysis
– (Fractional) Immunization – intuition behind λ1
• Conclusions
CMU, Feb 2014
![Page 104: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/104.jpg)
CMU SCS
Intuition for λ
“Official” definitions: • Let A be the adjacency
matrix. Then λ is the root with the largest magnitude of the characteristic polynomial of A [det(A – λI)].
• Also: A x = λ x
Neither gives much intuition!
“Un-official” Intuition • For ‘homogeneous’
graphs, λ == degree
• λ ~ avg degree – done right, for skewed
degree distributions
(c) 2014, C. Faloutsos 104 CMU, Feb 2014
![Page 105: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/105.jpg)
CMU SCS
Largest Eigenvalue (λ)
λ ≈ 2 λ = N λ = N-‐1
N = 1000 nodes λ ≈ 2 λ= 31.67 λ= 999
beaer connec4vity higher λ �
(c) 2014, C. Faloutsos 105 CMU, Feb 2014
![Page 106: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/106.jpg)
CMU SCS
Largest Eigenvalue (λ)
λ ≈ 2 λ = N λ = N-‐1
N = 1000 nodes λ ≈ 2 λ= 31.67 λ= 999
beaer connec4vity higher λ �
(c) 2014, C. Faloutsos 106 CMU, Feb 2014
![Page 107: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/107.jpg)
CMU SCS
Examples: Simulations – SIR (mumps) Fr
actio
n of
Infe
ctio
ns
Foot
prin
t Effective Strength Time ticks
(c) 2014, C. Faloutsos 107 CMU, Feb 2014
(a) Infection profile (b) “Take-off” plot PORTLAND graph: syntheGc populaGon,
31 million links, 6 million nodes
![Page 108: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/108.jpg)
CMU SCS Examples: Simulations – SIRS
(pertusis)
Frac
tion
of In
fect
ions
Foot
prin
t Effective Strength Time ticks
(c) 2014, C. Faloutsos 108 CMU, Feb 2014
(a) Infection profile (b) “Take-off” plot PORTLAND graph: syntheGc populaGon,
31 million links, 6 million nodes
![Page 109: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/109.jpg)
CMU SCS
Immunization - conclusion In (almost any) immunization setting, • Allocate resources, such that to • Minimize λ1
• (regardless of virus specifics)
• Conversely, in a market penetration setting – Allocate resources to – Maximize λ1
CMU, Feb 2014 (c) 2014, C. Faloutsos 109
![Page 110: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/110.jpg)
CMU SCS
(c) 2014, C. Faloutsos 110
Roadmap
• Introduction – Motivation • Part#1: Patterns in graphs • Part#2: Cascade analysis
– (Fractional) Immunization – Epidemic thresholds
• Acks & Conclusions
CMU, Feb 2014
![Page 111: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/111.jpg)
CMU SCS
(c) 2014, C. Faloutsos 111
Thanks
CMU, Feb 2014
Thanks to: NSF IIS-0705359, IIS-0534205, CTA-INARC; Yahoo (M45), LLNL, IBM, SPRINT, Google, INTEL, HP, iLab
Disclaimer: All opinions are mine; not necessarily reflecting the opinions of the funding agencies
![Page 112: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/112.jpg)
CMU SCS
(c) 2014, C. Faloutsos 112
Project info: PEGASUS
CMU, Feb 2014
www.cs.cmu.edu/~pegasus Results on large graphs: with Pegasus +
hadoop + M45 Apache license Code, papers, manual, video
Prof. U Kang Prof. Polo Chau
![Page 113: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/113.jpg)
CMU SCS
(c) 2014, C. Faloutsos 113
Cast
Akoglu, Leman
Chau, Polo
Kang, U
McGlohon, Mary
Tong, Hanghang
Prakash, Aditya
CMU, Feb 2014
Koutra, Danai
Beutel, Alex
Papalexakis, Vagelis
![Page 114: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/114.jpg)
CMU SCS
(c) 2014, C. Faloutsos 114
CONCLUSION#1 – Big data
• Large datasets reveal patterns/outliers that are invisible otherwise
CMU, Feb 2014
![Page 115: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/115.jpg)
CMU SCS
(c) 2014, C. Faloutsos 115
CONCLUSION#2 – self-similarity
• powerful tool / viewpoint – Power laws; shrinking diameters
– Gaussian trap (eg., F.O.F.)
– RMAT – graph500 generator
CMU, Feb 2014
![Page 116: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/116.jpg)
CMU SCS
(c) 2014, C. Faloutsos 116
CONCLUSION#3 – eigen-drop
• Cascades & immunization: G2 theorem & eigenvalue
CMU, Feb 2014
CURRENT PRACTICE OUR METHOD
[US-‐MEDICARE NETWORK 2005]
~6x fewer!
14 secs
> 30,000x speed-‐up!
![Page 117: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/117.jpg)
CMU SCS
(c) 2014, C. Faloutsos 117
References • D. Chakrabarti, C. Faloutsos: Graph Mining – Laws,
Tools and Case Studies, Morgan Claypool 2012 • http://www.morganclaypool.com/doi/abs/10.2200/
S00449ED1V01Y201209DMK006
CMU, Feb 2014
![Page 118: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/118.jpg)
CMU SCS
TAKE HOME MESSAGE: Cross-disciplinarity
CMU, Feb 2014 (c) 2014, C. Faloutsos 118
![Page 119: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/119.jpg)
CMU SCS
• Same accuracy, 100x – 100K x faster
Already started paying off for power grids
1000 x
CMU, Feb 2014 119 (c) 2014, C. Faloutsos
[1] Yang Weng, Christos Faloutsos, Marija D. Ili´c, and Rohit Negi, Speed up of Data-Driven State Estimation Using Low-Complexity Indexing Method, IEEE PES-General Meeting,
(accepted), 2014
Kd-tree SVD
![Page 120: Large Graph Mining - research.ece.cmu.edu · CMU SCS Large Graph Mining - Patterns, Explanations and Cascade Analysis Christos Faloutsos CMU](https://reader035.vdocuments.mx/reader035/viewer/2022071401/60ebd06a36b357059f0ff90e/html5/thumbnails/120.jpg)
CMU SCS
• Same accuracy, 100x – 100K x faster
1000 x
CMU, Feb 2014 120 (c) 2014, C. Faloutsos
[1] Yang Weng, Christos Faloutsos, Marija D. Ili´c, and Rohit Negi, Speed up of Data-Driven State Estimation Using Low-Complexity Indexing Method, IEEE PES-General Meeting,
(accepted), 2014
THANK YOU!
Kd-tree SVD