ciml - 5:36 pm on Aug 31, 2002 (gmt 0)

OK, here's some more guessing.

"Keyword in links to page" really means "keyword near links to page", so how about "keyword density and HTML title of the pages that link."?

That might be quite easy for Google to implement compared with contextual PageRank?

Now to get futuristic...

* Keyword density and HTML title of the pages that are linked from the pages that link to the page in question. (In other words, do the 'Similar pages' match the phrase?)

* Keyword density and HTML title of the pages that link to the pages that link to the page in question. (A very low-tech attempt at contextual PageRank).

* Title of the ODP category the page is in (and maybe the parent categories too).

* Title of the ODP category the pages that link to the page (and maybe their parent categories too).

* Full Contextual PageRank (i.e. calculate PageRank across the Web for each phrase. Not likely any time soon.)

* 'Topic Sensitive PageRank' (i.e. calculate PageRank across the Web for a few topics (eg. ODP), then match one of those topical PageRanks to the search phrase instead of the general PR - see Haveliwala's paper of that name.)

* Build the list of phrase hits using PageRank and on-page factors (like Google does) then use those pages to find the most authoritative (similar to Bharat's and Mihaila's Hilltop).

There are so many things for Google to try, but presumably the key is not just how well they work but how practical they are.