Blog Spam Panel AIRWeb 2006

1
Blog Spam PanelAIRWeb 2006

• Dennis Fetterly
• Microsoft Research, Silicon Valley

2
Scope of the problem

• Random sample of 2694 blog pages judged by MSN
• Low cost of creation
• 39 of spam blog pages were from 4 popular blog
  hosting sites

3
Blog Spam

• Content-based techniques are effective at finding
  some portion of splog pages
• Identification of re-purposed content is an open
  problem
• Whole document
• Just sentences or phrases
• Several software packages will automate this

4
Detectable example
5
Harder example

• 234 results for phrase query
• "There's one problem with adopting a dog from an
  animal shelter The selection of available canine
  companions can overwhelm you"
• Some are legitimate
• Others arent
• Other posts in these blogs create optimal link
  structure for PageRank as identified by Bianchini
  et al.
• Also significant overlap with other pages
• 12 minor differences in first 776 words
• Missing final 617 words

6
Things to Consider

• Once you have identified re-purposed content,
  what do you do

7

• Thank You!
• http//research.microsoft.com/research/sv/web-grou
  p/