Linking Theory Axioms

By Michael Martinez on February 5, 2008 at 4:51 pm

eserrano: “I’ve noticed a dampening effect when repeating query terms on a page whereby there seems to be a decreasing benefit after so many repeats. Any thoughts on this?”

Michael: I’ve never noticed natural language repetition to have any ill effects but people can certainly make their repetition look too contrived. It may be that the search engines try to adjust for that.

However, the only way to be sure is to test the idea thoroughly: do it, undo it, do it, undo it, etc. Evaluate the search results after the pages have been re-indexed.

eserrano: “Also, I’m increasingly seeing traffic from searches on terms that have topical relationships to target keywords on stories. For example, a story using the term “parents” will rank for “family” even though the later is not mentioned in the article. While I’ve seen hints of this in the past, there seems to be more of a sudden prevalence in Google, so relevance may be given to terms outside of one’s target within a page.”

Michael: I have seen similar crossover between queries recently but cannot rule out either a) link anchor text or b) query analysis (where a search engine determines that two queries may be looking for the same result).

Most people would be quick to guess that the search engine has applied latent semantic analysis, but that’s not so easy to show.

That said, I’ve seen no indication that the old relevance factors have been scaled back. As long as the search engines allow links to pass anchor text (and in my opinion they should NOT) then the model still works.

But if we assume for the sake of discussion that search engines are beginning to implement semantic comparison at the query level, there is no reason for why that model should have to change much. You’re just looking at a more semantic-like model.

By Michael Martinez on February 6, 2008 at 12:30 am

Derek Edmond: “I would be curious to know if you had seen or heard indication of search engines attempting to infer ‘opinion’ or possibly intent.”

Michael: I think some people have suggested that “rel=nofollow” reflects an opinion (I don’t trust what I am linking to) but the implicit lack of trust is relevant only to search engine indexing. Nofollow doesn’t in itself mean “the destination is not important or unworthy of being linked”.

SEOs who have jumped on the idea of using NoFollow to “sculpt PageRank” have thus clouded the interpretation of NoFollow as an expression of opinion since, clearly, they implicitly trust their own pages. They’re just trying to tell the search engine not to pass PageRank (or anchor text) to specific pages.

The traditional Google view that links constitute votes for other sites or endorsements is another attempt to impose opinion on links. But there are millions, possibly tens of millions of links that have been shared across the years and archived in numerous places where people linked to Web sites with warnings, derision, disagreement, objection, outrage, etc. The links in no way convey any sense of confidence in the destinations but Google has relied upon many of them for years to influence its search indexing.

Ask’s ExpertRank (based at least in part on IBM’s HITS and CLEVER algorithms) treats linking relationships as implicit opinions (”this document is relevant to my topic”, “this document organizes content relevant to my topic”, “this document is important to my topic”) — again, without regard for whether the linking document author objects to or disagrees with the destination document’s content.

Opinions can be expressions of approval or disapproval (or neutrality) with respect to any assertion, but they can also be expressions of conclusions or deductions. I have been fairly simplistic in my criticism of the “links are endorsements or votes” point of view because you really only need 3 states (approval, disapproval, or neutrality) to show that link architecture is incapable of expressing opinion.

But even if Google comes up with something that is incorporated into HTML 6 (HTML 5 is only being put together in the proposal stage) as a mechanism for “expressing opinion through links”, it would probably be woefully inadequate.

Opinions can express weight, authority, humor, disbelief, extreme bias, cold objectivity, and other very subtle points of view. And who would realistically want to spend their typing energy flagging each link with “rel=approve”, “rel=disagree”, “rel=weight”, etc.?

I would NOT want to do that. However, as semantic markup becomes more refined and sophisticated, we may indeed find ourselves setting standards for expressing opinions through markup language. We have smilies, geekcode, and possibly half a dozen other forms of cryptic Net Expression that search engines could — if they put their resources to work — index and interpret.