Combating Web Spam with TrustRank

1
Combating Web Spam with TrustRank

• Authors Zoltan Gyongyi, Hector Garcia-Molina,
  Jan Pedersen
• Presented by Jing Huang

2
What is Web Spam?

• Definition in this paper the term web spam
  refers to hyperlinked pages on the WorldWideWeb
  that are created with the intention of misleading
  search engines.
• Something behind
• than they deserve
• unjustifiably favorable ranking wrt the
  pages true value
• unethical web page positioning
• It is a problem, not only for search engines
• Primarily for users
• As well as for content providers
• It is first a social problem, then a technical
  one
• Reference P. Takis Metaxas, Web Spam, Propaganda
  and Trust

3
How to Spam?

• Add keywords so as to confuse page relevance
• i.e
• SEX SEXY MONICA LEWINSKY JENNIFER LOPEZ CLAUDIA
  SCHIFFER CINDY CRAWFORD JENNIFER ANNISTON GILLIAN
  ANDERSON MADONNA NIKI TAYLOR ELLE MACPHERSON KATE
  MOSS CAROL ALT TYRA BANKS FREDERIQUE KATHY
  IRELAND PAM ANDERSON KAREN MULDER VALERIA MAZZA
  SHALOM HARLOW AMBER VALLETTA LAETITA CASTA BETTIE
  PAGE HEIDI KLUM PATRICIA FORD DAISY FUENTES KELLY
  BROOK SEX SEXY MONICA LEWINSKY JENNIFER LOPEZ
  CLAUDIA SCHIFFER CINDY CRAWFORD JENNIFER ANNISTON
  GILLIAN ANDERSON MADONNA NIKI TAYLOR ELLE
  MACPHERSON KATE MOSS CAROL ALT TYRA BANKS
  FREDERIQUE KATHY IRELAND PAM ANDERSON KAREN
  MULDER VALERIA MAZZA SHALOM HARLOW AMBER
• creation of a large number of bogus web pages,
  all pointing to a single target page
• Reference P. Takis Metaxas, Web Spam, Propaganda
  and Trust

4
One of the solving methods Combating Web Spam
with TrustRank

• Content
• Preliminaries
• Accessing Trust
• Computing Trust
• Selecting Seeds
• Experiments
• Results

5
Preliminaries

• Web Model
• transition matrix T

5
Indegree 1 Outdegree2
Unreferenced page
Nonreferenced page
6
Overview of PageRank

• PageRank is based on a mutual reinforcement
  between pages the importance of a certain page
  influences and is being influenced by the
  importance of some other pages.
• The PageRank score r(p) of a page p is defined
  as
• Biased PageRank version

Decay factor
Static score distribution vector of arbitrary
7
Assessing Trust

• Oracle Function
• O(p) 0 if p is bad
• 1 if p is good
• This binary function used as the notion for human
  checking a page
• Example

8
Trust Functions

• Definition a trust function T that yields a
  range of values between 0(bad) and 1(good)
• Ideal Trust Property
• Ordered Trust Property
• Threshold Trust Property

If a page P receives a score above the threshold
then P is good
9
Computing Trust
Web graph

• Ignorant Trust Function

10
Computing Trust Example
Set L 3
11
Trust Propagation

• Assumption good pages point to other good pages
  only, assign a score of 1 to all pages that are
  reachable from a page in S in M or fewer steps.
• M- Step Trust Function
• Example

12
Trust Attenuation

• Trust dampening
• Trust splitting

13
TrustRank Algorithm

• Function Trust Rank
• Input T transition matrix, N number of
  pages, L limit of oracle invocations,
• - decay factor for biased pageRank -
  number of biased PageRank iterations
• Output - TrustRank scores
• Begin
• S SelectSeed()
• Rank(1,,N,s)
• d 0
• for 1 1 to L do
• if
• for i 1 to do
• return
• end

Select good seeds
Normalize static score distribution vector
Compute TrustRank scores
14
S 0.08,0.13,0.08,0.10,0.09,0.06,0.02
Order 2,4,5,1,3,6,7
Assume L 3 selected seed set 2,4,5
d 0,1/2,0,1/2,0,0,0
T0,0.18,0.12,0.15,0.13,0.05,0.05
15
Selecting Seeds

• High PageRank
• Inverse PageRank
• Coverage is important
• Build the seed set from those pages which point
  to many pages that in turn point to many pages
  and so on.
• Only different between PageRank is this method
  based on outlinks
• Example

L 2
S 1,2
S 0.05,0.05,0.04,0.02,0.02,0.02,0.02
16
Evaluation Metrics(1)

• Pairwise orderedness( related to ordered trust
  property)

P is a set of ordered pairs of page(p,q), p!q
from the sample X
17
Evaluation Metrics(2)

• Precision
• the fraction of good among all pages in X that
  have a trust score above
• Recall
• the ratio between the number of good pages with a
  trust score above and the total number of
  good pages in X

18
Experiments

• Data set Grouped 31,003,946 sites using a
  proprietary algorithm from Alta Vista.
• Seed set inversed PageRank
• Manually evaluated the top 1250 seeds
• Get 178 sites as good seeds
• Evaluation Sample
• use 748 of the sample sites to evaluate
  TrustRank
• Reputable - 563 Web organization 37
  Advertisement-13 Spam -135

19
Results

• TrustRank
• 178 good seeds
• PageRank
• Use the PageRank of site as the value of T(a)
• Ignorant Trust
• All sites assigned an ignorant trust score of ½
  except for the 1250 seeds

20
Results

• PageRank VS TrustRank

Good Sites in PageRank buckets
Good Sites in TrustRank buckets
Bad Sites in PageRank buckets
Bad Sites in PageRank buckets
21
Results

• Pairwise Orderedness

22
Results

• Precision and Recall

23
Something after TrustRank

• Big sites monopolize the search results
• i.e. wiki
• Reputable not equal to relevance
• How to stop of other types of spam
• Image
• Based on the content of the page that image
  appears
• Video
• music

24
(No Transcript)
25
(No Transcript)