PageRank - 國立臺灣大學

19
PageRank Hung-yi Lee

Transcript of PageRank - 國立臺灣大學

Page 1: PageRank - 國立臺灣大學

PageRankHung-yi Lee

Page 2: PageRank - 國立臺灣大學

Search

http://incomebully.com/does-pr-pagerank-still-matter/

Page 3: PageRank - 國立臺灣大學

http://www.hobo-web.co.uk/google-pr-update/

Page 4: PageRank - 國立臺灣大學

PageRank

• Information of 2008

Source: http://terrylogin.blogspot.com/2008/09/pagerank.html

Rank:8Google台灣首頁:www.google.com.twYoutube台灣首頁:http://tw.youtube.com台灣大學網站首頁:www.ntu.edu.tw

Rank:7痞客邦首頁:www.pixnet.net104人力銀行:www.104.com.tw無名小站首頁:www.wretch.cc

Rank:6博客來網站:www.books.com.tw聯合新聞網首頁:http://udn.com天下雜誌網站首頁:www.cw.com.tw

Rank:5推推王網站:http://funp.com愛情公寓網站:www.i-part.com.twKKman網站首頁:www.kkman.com.tw

Rank:4工頭堅部落格:http://worker.bluecircus.net白木怡言部落格:www.yubou.twRO仙境傳說網站:http://ro.gameflier.com

Page 5: PageRank - 國立臺灣大學

PageRank

Page 6: PageRank - 國立臺灣大學

PageRank

• Webpages with a higher PageRank are more likely to appear at the top of Google search results.

• PageRank relies on the uniquely democratic nature of the web by using its vast link structure as an indicator of an individual page’s value.

• Google interprets a link from page A to page B as a vote, by page A, for page B.

Page 7: PageRank - 國立臺灣大學

Importance

很多網頁連到

很多網頁連到 被重要網頁連到

hyperlink

Page 8: PageRank - 國立臺灣大學

PageRank

Page 9: PageRank - 國立臺灣大學

Importance - Formulas

1

2

3

4

𝑥1

𝑥2

𝑥3

𝑥4

𝑥1 = 𝑥3 +1

2𝑥4

𝑥2 =1

3𝑥1

𝑥3 =1

3𝑥1 +

1

2𝑥2 +

1

2𝑥4

𝑥4 =1

3𝑥1 +

1

2𝑥2

Consider a random surfer

1/2

1/2

Page 10: PageRank - 國立臺灣大學

Importance - Formulas

𝑥1 = 𝑥3 +1

2𝑥4

𝑥2 =1

3𝑥1

𝑥3 =1

3𝑥1 +

1

2𝑥2 +

1

2𝑥4

𝑥4 =1

3𝑥1 +

1

2𝑥2

𝐴𝑥 = 𝑥

The solution x is in the eigenspace of eigenvalue 𝜆 = 1

𝑥 =

𝑥1𝑥2𝑥3𝑥4

Page 11: PageRank - 國立臺灣大學

Importance - Formulas

1

2

3

4

𝑥1

𝑥2

𝑥3

𝑥4

12

4

9

6

𝐴𝑥 = 𝑥

𝑥 =

𝑥1𝑥2𝑥3𝑥4

The solution x is in the eigenspace of eigenvalue 𝜆 = 1

𝑆𝑝𝑎𝑛 12 4 9 6 𝑇

Page 12: PageRank - 國立臺灣大學

Eigenvalue = 1

Column-stochastic Matrix

Column-stochastic matrix always have eigenvalue 𝜆 = 1

How about the Dangling nodes (只入不出)?

Proof

𝐴𝑥 = 𝑥

Page 13: PageRank - 國立臺灣大學

Unique Ranking?

1

2

3

4

𝑥1

𝑥2

𝑥3

𝑥4

12

4

9

6

Having eigenvalue 𝜆 = 1

The dimension of the subspace is must 1

Unique Ranking

Unique Scoreconstraint

Page 14: PageRank - 國立臺灣大學

Unique Ranking?

Dim for 𝜆 = 1 is 2

1

2

3

4

5

1/2

1/2

0

0

0

1

2

3

4

50

0

0

1/2

1/2

Any linear combination is in the eigenspace

Basis:

Not Unique Ranking

How about dimension > 1

Page 15: PageRank - 國立臺灣大學

Unique Ranking?All entries are 1/n

1

2

3

4

5

There are two ways to surf the web

Follow the link random

1

2

3

4

5

Prob 1 – m: Prob m:

Can it be non-uniform?

0.15

Page 16: PageRank - 國立臺灣大學

Unique Ranking?

• Unique ranking

• For M, the dim of the eigenvalue 𝜆 = 1 is 1

M is Column-stochastic matrix and “positive”

Proof

Hint: For M, the eigenvectors for eigenvalue 𝜆 = 1 are all “positive” or “negative”

Dim = 1

Page 17: PageRank - 國立臺灣大學

Power method

Find 𝑥∗, such that 𝑥∗ = 𝑀𝑥∗, 𝑥∗ 1 = 1

Start from 𝑥0, 𝑥0 1= 1

𝑥1 = 𝑀𝑥0

𝑥2 = 𝑀𝑥1

𝑥𝑘 = 𝑀𝑥𝑘−1…

If 𝑘 → ∞

𝑥𝑘 = 𝑥∗

M is very large

Proof

Page 18: PageRank - 國立臺灣大學

Power method - Example

1

2

3

4

Page 19: PageRank - 國立臺灣大學

Actually ……

• The Last Toolbar Pagerank Update was December 2013

• Google declared thereafter: “PageRank is something that we haven’t updated for over a year now, and we’re probably not going to be updating it again going forward, at least the Toolbar version.“