Collaborative Filtering In, Link Pop Out?

4 comments
Source Title:
Interview with Dr. Garcia (Orion)
Story Text:

Rand has posted his interview with Dr. Garcia aka Orion at SEOmoz. Link-obsessed folks beware...

Rand: Are there any standout shifts you see towards greater or lesser use of things like PageRank, Trustrank, VIPS, Local Link Popularity, etc.?

Dr. Garcia: Link-based models have failed miserably. Some few “verticals” are using feedbacks and reviews to improve results using Slope One methods and item-to-item collaborative filtering (personalization based on recommendation). I’m not sure if this technology is ready for general searches. A system based on item-to-item collaborative filtering and that considers users’ actions as “review votes” can be gamed, without not even considering the standard issues (scalability, security, spam etc).

If indeed link-based algo's get a downgrade and collaborative filtering is heavily integrated, SEO's could have a field day with "review vote" spamming... but I'm actually not buying it that link pop has failed as the SE algo's work horse.

Comments

Well Noted

Thankfully, I think Dr. Garcia was overstating the failure of link popularity in SE algos. Like on-page, it is very unlikely that these will ever dissappear entirely, but models like PageRank are going the way of the Dodo bird in favor of more accurate and inclusive measurements of link quality.

Collaborative filtering gaming would require access to the filtering, and if Google, Yahoo! and the others are going to use their own labor for this, SEOs have little hope of gaming it. It would be more of an top-down system than a true social tagging.

> models like PageRank are

> models like PageRank are going the way of the Dodo bird in favor of more accurate and inclusive measurements of link quality.

I agree that PageRank is no longer king, nor has it been for a long time.

But LocalRank, TrustRank, Topic-Sensitive PageRank are all basically link pop--elegant link pop, yes--but it's still all about the links, it just requires a different, and smarter, link building strategy now.

>Collaborative filtering gaming would require access to the filtering, and if Google, Yahoo! and the others are going to use their own labor for this, SEOs have little hope of gaming it. It would be more of an top-down system than a true social tagging.

Ah yes I suppose I was thinking of collaborative filtering done by average searchers, but the top-down approach would make a lot more sense.

Links will always be important

I agree with you notredamekid. I respectfully don't agree in its entirety with Dr. Garcia's views on link-based models.

>But LocalRank, TrustRank, Topic-Sensitive PageRank are all basically link pop--elegant link pop, yes--but it's still all about the links, it just requires a different, and smarter, link building strategy now.

Whatever name you want to call the link-popularity algo it doesn't matter. The web is called the web because of how pages are connected amongst one another. That will never change.

IMO, what you will see happening in the future is SE getting more precise in determining which links are:

1) authentic (not used for gaming the SEs)
2) relevant (with a real working formula, not just per-sae)
3) time dependent (old vs. new)
4) dynamic vs. static
5) location of the destination page that was linked to, within the website and in the entire web graph
6) location of the link within the page that is linking out to the web (top of the page, bottom, hidden, in the paragraph, in a title, etc.)

And many other factors as well...

There is a lot of value in links, as they will always be important in the world wide web.

Links alway important

I cannot see the web ever forgoing the value of links. How they are factored will change, but as Nacho notes we are talking about the web - the intricate interrelated world of sites on the internet.

Maybe there comes a time when there is a communal search database more refined as the members rank the results and the scores over time factor placement.... but ultimately linking is a measurable that any large scale computer-based analytic view will rely on.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.