Download presentation
Presentation is loading. Please wait.
Published byRalph Byrd Modified over 8 years ago
1
FAST-PPR: Personalized PageRank Estimation for Large Graphs Peter Lofgren (Stanford) Joint work with Siddhartha Banerjee (Stanford), Ashish Goel (Stanford), and C. Seshadhri (Sandia)
2
Motivation: Personalized Search 2
3
3 Re-ranked by PPR
4
Result Preview 4 2 sec 6 min 1.2 hour Fast-PPRMonte- Carlo Local- Update
5
Personalized PageRank 5
6
Goal 6
7
Previous Algorithm: Monte-Carlo 7 [Avrachenkov, et al 2007]
8
Previous Algorithm: Local Update 8 [Anderson, et al 2007]
9
Main Result 9
10
Analogy: Bidirectional Search 10
11
Bidirectional PageRank Algorithm 11 Reverse Work (Frontier Discovery) Forward Work (Random Walks) u
12
Main Idea 12
13
Experimental Setup 13
14
Empirical Running Time 14 Log Scale
15
Summary 15
16
Thank You Paper available on Arxiv Code available at cs.stanford.edu/~plofgren cs.stanford.edu/~plofgren 16
17
17
18
Frontier is Important 18 Frontier Aided Significance Thresholding
19
Algorithm (Simple Version) 19
20
Algorithm (Simple Version) 20
21
Average Running Time 21 Reverse Work (Local Update) Forward Work (Monte-Carlo)
22
Correctness 22
23
Algorithm (Theoretical Version) 23
24
24 Algorithm (Theoretical Version)
25
v1v1 Local Update Algorithm 25 U u U v2v2 U v3v3 U t
26
Local Update Algorithm 26
27
Local Update Algorithm 27
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.