Link Prediction: Leonid E. Zhukov

Link prediction
Leonid E. Zhukov
School of Data Analysis and Artificial Intelligence

Department of Computer Science
National Research University Higher School of Economics
Structural Analysis and Visualization of Networks
Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 1 / 24

Lecture outline
1 Link prediction problem
2 Proximity measures
3 Prediction by supervised learning
4 Other methods

Link prediction
Link prediction. A network is changing over time. Given a snapshot

of a network at time t, predict edges added in the interval (t, t 0 )
Link completion (missing links identification). Given a network, infer
links that are consistent with the structure, but missing (find
unobserved edges)
Link reliability. Estimate the reliability of given links in the graph.
Predictions: link existence, link weight, link type

Link prediction
Graph G(V,E)
Number of ”missing edges”: |V |(|V | − 1)/2 − |E |
In sparse graphs |E | |V |2 , Prob. of correct random guess O( |V1|2 )

Scoring algorithm
Link prediction by proximity scoring

1 For each pair of nodes compute proximity (similarity) score c(v1 , v2 )
2 Sort all pairs by the decreasing score
3 Select top n pairs (or above some threshold) as new links

Scoring functions
Local neighborhood of vi and vj

Number of common neighbors:
|N (vi ) ∩ N (vj )|
Jaccard’s coefficient:
|N (vi ) ∩ N (vj )|
|N (vi ) ∪ N (vj )|
Adamic/Adar:
X 1
log |N (v )|
v ∈N (vi )∩N (vj )
Liben-Nowell and Kleinberg, 2003

Scoring functions
Paths and ensembles of paths between vi and vj

Shortest path:
− min{pathijs > 0}
s
Katz score:
∞
X ∞
X
l (l)
β |paths (vi , vj )| = (βA)lij = (I − βA)−1 − I
l=1 l=1
Personalized (rooted) PageRank:
PR = α(D −1 A)T PR + (1 − α)|

Scoring functions
Expected number of random walk steps:

– hitting time: −Hij
– commute time −(Hij + Hji )
– normalized hitting/commute time (Hij πj + Hji πi )
SimRank:
C X X
SimRank(vi , vj ) = SimRank(m, n)
|N (vi )| · |N (vj )|
m∈N (vi ) n∈N (vj )

Vertex feature aggregations
Preferential attachment:
ki · kj = |N (vi )| · |N (vj )|
or
ki + kj = |N (vi )| + |N (vj )|
Clustering coefficient:
CC (vi ) · CC (vj )
or
CC (vi ) + CC (vj )

Low-rank approximations
Low-rank approximation (truncated SVD)

X
A≈ Uk Sk VkT
k

Link prediction

Link prediction

Link prediction

Binary classificaton
Challenging classification problem:

Computational cost of evaluating of very large number of possible
edges (quadratic in number of nodes)
Highly imbalanced class distribution: number of positive examples
(existing edges) grows linearly and negative quadratically with
number on nodes

Prediction difficulty
Actual and possible collaborations between DBLP authors
from Rattigan and Jensen, 2005

Link prediction with supervised learning
Supervised learning:
1 Features generation
2 Model training
3 Testing (model application)
Features:
Topological proximity features
Aggregated features
Content based node proximity features

Features
Discriminative abilities of features
from Rattigan and Jensen, 2005

Evaluation
Simple ”hold out set” evaluation

Evaluation
Evaluation for evloving networks
image from Y. Yang et.al, 2014

Evaluation metrics
Precision and Recall, F-measure

TP TP
Precision = , Recall =
TP + FP TP + FN
2 · Precision · Recall
F =
Precision + Recall
True positive rate (TPR), False positive rate (FPR), ROC curve, AUC
TP FP
TPR = , FPR =
TP + FN FP + TN

Performance of classification algorithms
BIOBASE database (research publications)
DBLP dataset (research publications)
from M. Al Hasan, 2010

ROC curves
from Lichtenwalter, 2010

Probabilistic models
Local model, Markov random fields [Wang, 2007]

Hierarchical probabilistic model [Clauset, 2008]
Probabilistic relations models:
– Bayesian networks [Getoor, 2002]
– relational Markov networks [Tasker, 2003, 2007]

References
D. Liben-Nowell and J. Kleinberg. The link prediction problem for

social networks. Journal of the American Society for Information
Science and Technology, 58(7):1019?1031, 2007
R. Lichtenwalter, J.Lussier, and N. Chawla. New perspectives and
methods in link prediction. KDD 10: Proceedings of the 16th ACM
SIGKDD, 2010
M. Al Hasan, V. Chaoji, S. Salem, M. Zaki, Link prediction using
supervised learning. Proceedings of SDM workshop on link analysis,
2006
M. Rattigan, D. Jensen. The case for anomalous link discovery. ACM
SIGKDD Explorations Newsletter. v 7, n 2, pp 41-47, 2005
M. Al. Hasan, M. Zaki. A survey of link prediction in social networks.
In Social Networks Data Analytics, Eds C. Aggarwal, 2011.

Link Prediction: Leonid E. Zhukov

Uploaded by

Copyright:

Available Formats

Link Prediction: Leonid E. Zhukov

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Link Prediction: Leonid E. Zhukov

Uploaded by

Copyright:

Available Formats

Link prediction

School of Data Analysis and Artificial Intelligence

Structural Analysis and Visualization of Networks

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 1 / 24

1 Link prediction problem

3 Prediction by supervised learning

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 2 / 24

Link prediction. A network is changing over time. Given a snapshot

Predictions: link existence, link weight, link type

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 3 / 24

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 4 / 24

Link prediction by proximity scoring

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 5 / 24

Local neighborhood of vi and vj

Liben-Nowell and Kleinberg, 2003

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 6 / 24

Paths and ensembles of paths between vi and vj

Personalized (rooted) PageRank:

PR = α(D −1 A)T PR + (1 − α)|

Liben-Nowell and Kleinberg, 2003

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 7 / 24

Expected number of random walk steps:

Liben-Nowell and Kleinberg, 2003

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 8 / 24

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 9 / 24

Low-rank approximation (truncated SVD)

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 10 / 24

Liben-Nowell and Kleinberg, 2007

Liben-Nowell and Kleinberg, 2007

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 12 / 24

Liben-Nowell and Kleinberg, 2007

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 13 / 24

Challenging classification problem:

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 14 / 24

Actual and possible collaborations between DBLP authors

from Rattigan and Jensen, 2005

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 15 / 24

3 Testing (model application)

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 16 / 24

Discriminative abilities of features

from Rattigan and Jensen, 2005

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 17 / 24

Simple ”hold out set” evaluation

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 18 / 24

Evaluation for evloving networks

image from Y. Yang et.al, 2014

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 19 / 24

Precision and Recall, F-measure

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 20 / 24

BIOBASE database (research publications)

DBLP dataset (research publications)

from M. Al Hasan, 2010

from Lichtenwalter, 2010

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 22 / 24

Local model, Markov random fields [Wang, 2007]

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 23 / 24

D. Liben-Nowell and J. Kleinberg. The link prediction problem for

Leonid E. Zhukov (HSE) Lecture 18 26.05.2015 24 / 24

You might also like