And my feeling is that in the context of machine-generated pages, Page Rank doesn't help that much.
It's better than random. It correctly identified overview-summary as the best "home page" for the collection in both cases. It also identified some core classes (IndexReader in Lucene, Object & String in Java).
Also, it's not clear how to use it e.g. make it the Document boost or put it into a separate field for use by a custom scoring function?
I think using the Document boost makes good sense.
And...I'm pretty sure it can't easily be used w/ incremental index additions as it wants an entire link graph.
A standard way to deal with this is to make a guess for new pages. A new page should probably have a score somewhat less than the page which linked to it, and also probably a bit less than other pages at the same "site" that were previously known.
Doug
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
