The damping factor: PageRank as probability

In order to understand what the damping factor means, it helps to reframe our PageRank™ metaphor from a political election to a Web-surfing monkey.

PageRank™ provides a kind of probability distribution for the following random variable: If a monkey were to sit down at a browser and start clicking and typing, what Web page would be left on the screen when the zookeeper finally takes the monkey away from the computer? PageRank™ tells us the likelihood that any specific page will be the monkey's final destination.

Now suppose we also know how often the monkey likes to click on a random link vs. how often the monkey likes to type a random URL into the browser. In this scenario, the damping factor d expresses the monkey's preference for clicking instead of typing. So...

  • If d=1 then the resulting PageRank™ function assumes that the monkey starts at a random page and then always clicks and never types in a new URL.
  • If d=0 then the resulting PageRank™ function assumes that the monkey never clicks; he always types in another random URL.

Admittedly the d=0 model corresponds to an unrealistic and/or pathological version of Web surfing -- equivalent to the famous team of monkeys still typing up King Lear. Hence our best intelligence says that Google uses d=0.85, which is closer to 1 than 0.

 
Joomla Templates by Joomlashack