Sic Transit Gloria Telae: Towards an Understanding of the Web's Decay

1
Sic Transit Gloria Telae Towards an
Understanding of the Web's Decay
Ziv Bar-Yossef et al IBM Almaden and T.J Watson
Research Centers
Mark Strohmaier
2
Problem Motivation

• Determining if a link is dead is not trivial
• Using dead links as a decay signal is very noisy

3
Estimating Proportion of Dead Pages

• Begin at a page
• Probability 1-s of randomly walking off,
  probability s of declaring success
• If walk to a dead page, declare failure
• Overall decay score of the page is chance of
  failing for that page

4
Differences from PageRank

• Decay of a page can be processed in isolation
• Very easy to reduce a page's decay score

5
Three types of dead pages

• Malformed URL
• Host does not exist
• Page does not exist on host

6
Detecting 'soft-404s'

• Query a given server for a page that does not
  likely exist
• Record the server's response to the dummy
  request
• Observe the behaviour when the legitimate
  request is sent

7
Measure of Decay

• If D is the set of all dead pages

8
Experimental Procedure
9
First Round of Experiments

• 1000 pages were randomly chosen from a web crawl
  of two billion
• 475 were already dead
• Of 710 dead links, 207 pointed to soft-404s

10
First Round of Experiments
11
Decay Score versus Dead Links
12
Second Experiment WWW Conference links
13
Third Experiment Yahoo Leaf Nodes
14
Final Experiment FAQs.org
15
Conclusions and Remarks

• A number of tools exist for identifying dead
  links, but few exist for identifying decay
• Incorporating decay calculations into search
  results could be used to improve rankings
• Decay computations could also be used to improve
  crawling