Efficient Algorithms for Personalized...
Transcript of Efficient Algorithms for Personalized...
![Page 1: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/1.jpg)
Sublinear Algorithms for Personalized PageRank, with
Applications
Ashish Goel
Joint work with Peter Lofgren; Sid Banerjee; C Seshadhri
1
![Page 2: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/2.jpg)
Personalized PageRank
2
Assume a directed graph with n nodes and m edges
![Page 3: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/3.jpg)
Motivation: Personalized Search
3
![Page 4: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/4.jpg)
Motivation: Personalized Search Re-ranked by PPR
4
![Page 5: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/5.jpg)
Applications
• Personalized Web Search [Haveliwala, 2003]
• Product Recommendation [Baluja, et al, 2008]
• Friend Recommendation (SALSA) [Gupta et al, 2013]
5
![Page 6: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/6.jpg)
![Page 7: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/7.jpg)
A Dark Test for Twitter’s People Recommendation System
Run various algorithms to predict follows, but don’t display the results. Instead, just observe how many of the top predictions get followed organically
(Money = Personalized PageRank on a bipartite graph; Love = HITS)
[Bahmani, Chowdhury, Goel; 2010]
![Page 8: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/8.jpg)
Promoted Tweets and Promoted Accounts
![Page 9: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/9.jpg)
Applications
• Community Detection
– Personalized PageRank [Yang, Lescovec 2015],[Andersen, Chung, Lang 2006]
9
![Page 10: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/10.jpg)
Estimation Goal
10
![Page 11: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/11.jpg)
The Challenge
• Every user has different score vector: Full pre-computation: O(n2)
• Computing from scratch previously took Ω(n) time—several minutes on Twitter-2010
11
![Page 12: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/12.jpg)
Previous Algorithms Summary
• Monte-Carlo: Sample random walks.
• (Local) Power-Iteration: Iteratively improve estimates based on recursive equation
12
![Page 13: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/13.jpg)
13
3s
6 min
10+ min
Monte-Carlo
Power Iteration R
un
nin
g Ti
me
per
Es
tim
ate
(s)
Mean relative error set to ≈10% for all algorithms.
Runtime on Twitter-2010 (1.5 billion edges)
Results Preview (Experiment)
Bidirectional
![Page 14: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/14.jpg)
Results Preview (Theory) • Task: estimate of size within relative
error
• Previous Algorithms: – Monte Carlo:
– Power Iteration/ Local Update:
• Bidirectional Estimator for average target:
On Twitter-2010, n=40M, m=1.5B, =40K 14
# Nodes
# Edges
![Page 15: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/15.jpg)
Generalizations • Arbitrary starting distributions.
Uniform ⇒ Global PageRank in average time
• Other Walk Length Distributions like Heat Kernel (used in community detection [Kloster, Gleich 2014],[Chung 2007]): Our estimator is 100x faster on 4 graphs
• Arbitrary Discrete Markov Chain
15
![Page 16: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/16.jpg)
Previous Algorithm: Monte-Carlo
16
[Avrachenkov, et al 2007]
![Page 17: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/17.jpg)
Previous Algorithm: Local Update
17
[Andersen, et al 2007]
• Computes from all s to a single t
• Works from t backwards along edges, updating Personalized PageRank estimates locally.
• Running time for average t: Average Degree Additive Error
![Page 18: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/18.jpg)
Local Update Background
18
0.2 0.8
![Page 19: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/19.jpg)
Local Update Example
19
![Page 20: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/20.jpg)
Local Update Example
20
![Page 21: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/21.jpg)
Local Update Example
21
![Page 22: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/22.jpg)
Local Update Example
22
![Page 23: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/23.jpg)
Local Update Example
23
![Page 24: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/24.jpg)
Local Update Example
24
![Page 25: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/25.jpg)
Analogy: Bidirectional Shortest Path
25
![Page 26: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/26.jpg)
Bidirectional Estimation
The estimates p and residuals r satisfy a loop invariant [Anderson, et al 2007]:
Reinterpret the residuals as an expectation!
26
![Page 27: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/27.jpg)
Bidirectional-PPR Algorithm
27
![Page 28: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/28.jpg)
Number of samples
Every walk gives a sample, with
– Maximum value rmax
– Expected value at least ±
Number of walks needed to get a (1 +/- ²)-approximation with high probability =
28
![Page 29: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/29.jpg)
Bidirectional-PPR Example
29
![Page 30: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/30.jpg)
Theoretical Results
30
![Page 31: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/31.jpg)
Forward vs Reverse Work Trade-off
31 31
More Walks Fewer Reverse Pushes
More Reverse Push Fewer Walks
u
Forward Walks
Reverse Pushes
![Page 32: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/32.jpg)
Theoretical Results
Time-Space Trade-off [Lofgren, Banerjee, Goel, Seshadhri 2014; Lofgren, Banerjee, Goel 2015]
32
![Page 33: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/33.jpg)
Problem: Unbalanced Forward and Reverse Runtime
0.5 s Forward
0.1 ms Reverse
1000 s Reverse
Global PageRank of Target
Ru
nti
me
(s)
33
![Page 34: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/34.jpg)
Heuristic: Balancing Forward and Reverse Runtime
0.5 s Forward
0.1 ms Reverse
Median is 25 ms Forward and Reverse
1000 s Reverse
Global PageRank of Target
Ru
nti
me
(s)
34
50s Forward and Reverse
![Page 35: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/35.jpg)
Experiments
35
![Page 36: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/36.jpg)
Experimental Results: 70x Faster
Mean relative error set to ≈10% for all algorithms.
Ru
nn
ing
Tim
e p
er
Esti
mat
e (m
s)
Running Time (Targets PageRank)
36
20s
50ms
5min
3s
![Page 37: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/37.jpg)
Alternative Estimator for Undirected Graphs
Key property:
We push forwards from s, and take random walks from t.
37
![Page 38: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/38.jpg)
Alternative Algorithm for Undirected Graphs
• Loop Invariant of push-forward algorithm [Andersen, Chung, Lang, 2006]
• Use symmetry, and then interpret as expectation
![Page 39: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/39.jpg)
Alternative Algorithm for Undirected Graphs
• Loop Invariant of push-forward algorithm [Andersen, Chung, Lang, 2006]
• Use symmetry, and then interpret as expectation
rmax
![Page 40: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/40.jpg)
Running time for Undirected Graphs
![Page 41: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/41.jpg)
Open Problems
• Get rid of the dependence on degree, to get an amortized bound of O(±1/2)
• Get a worst-case bound of O(m1/2) for directed graphs under the condition that the target has a high global PageRank
• Find sharding and sampling algorithms that preserve Personalized PageRank (eg. a sparsifier for Personalized PageRank?)
• Build an index around Personalized PageRank to enable network based Personalized Search
41
![Page 42: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/42.jpg)
Open Problems
• Get rid of the dependence on degree, to get an amortized bound of O(±1/2)
• Get a worst-case bound of O(m1/2) for directed graphs under the condition that the target has a high global PageRank
• Find sharding and sampling algorithms that preserve Personalized PageRank (eg. a sparsifier for Personalized PageRank?)
• Build an index around Personalized PageRank to enable network based Personalized Search
42
![Page 43: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/43.jpg)
Personalized Search Problem
Given
– A network with nodes (with keywords) and edges (weighted, directed)—Twitter
– A query, filtering nodes to a set T— “People named Adam”
– A user s (or distribution over nodes) —me
Rank the approximate top-k targets by
43
![Page 44: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/44.jpg)
Personalized Search Problem
Baselines:
– Monte Carlo: Needs many walks to find enough samples within T unless T is very large
– Bidirectional-PPR to each t: Slow unless T is small
Challenge: Can we efficiently find top-k for any size of T?
44
Idea: Modify Bidirectional-PPR to sample in proportion to
![Page 45: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/45.jpg)
Personalized Search Example
t1
t2
t3
b a
s c
Expand targets Random walks To sample a target: layer 1: sample (a,b,c) w.p. (0, 10%, 90%) Layer 2: b→ t1 c→ sample (t1, t2, t3) w.p (56%, 22%, 22%)
45
People Named Adam Searching User
![Page 46: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/46.jpg)
Personalized Search Running Time R
un
nin
g Ti
me
per
Sea
rch
(s)
Target Set Size Precision@3 set to 90% for all algorithms.
46
Runtime on Twitter-2010 (1.5 billion edges)
10 100 1000 10,000
10
100
1000
1
0.1
Significant Pre-computation (3-30MB per keyword)
![Page 47: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/47.jpg)
Personalized Search Result
47
![Page 48: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/48.jpg)
Demo
48
![Page 49: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/49.jpg)
49
![Page 50: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/50.jpg)
50
![Page 51: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/51.jpg)
Demo Task: Find applications of entropy in computer networking.
personalizedsearchdemo.com
51
![Page 52: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/52.jpg)
Distributed PageRank
• Problem: Computing PageRank on graph too large for one machine.
• Algorithm:
– Shard edges randomly,
– compute on each machine
– average results
• Basic idea: Duplicate edges from low-degree nodes. Gives an unbiased* estimator.
52
![Page 53: Efficient Algorithms for Personalized PageRankdimacs.rutgers.edu/Workshops/ParallelAlgorithms/... · A Dark Test for Twitter’s People Recommendation System Run various algorithms](https://reader033.fdocuments.us/reader033/viewer/2022050202/5f55ad167f60fe24c307613c/html5/thumbnails/53.jpg)
53
L1 E
rro
r
Min Degree Parameter
Sharded Global PageRank Accuracy on Twitter-2010
No Edge Duplication
Average 2.1x Duplication