Multiple Links on a Page - Only a fraction of the link love? Options

[randfish]

Bwelford's post in the reciprocal links thread got me thinking about the functionality of how links pass 'relevance' and power according to the major search engines.

As we all know, PageRank's initial formula discriminated against pages with more links on them and lowered the link power of each link based on how many internal/external links a page had. As time has passed and PageRank is now just "pixie dust", is it still the general consensus that more links on a page means less power per link?

I haven't been able to think of a good universal reason why this should be so. In my opinion, a link on a page with 3, 6 or 10 other links, should be equally valuable - the number of external links shouldn't dampen the weight of each link, until you get into higher numbers - 20, 50, 100 external links (at which point you're probably dealing with a list of references, rather than a content subject based link).

I'd love to hear the community's thoughts.

[rustybrick]

I agree with your reasoning.

Grouping of link counts should be ideal:

1 - 10 links on a page = X
10 - 25
25 - 50, etc.

I do not know if all search engines lower the value of a link based on the same method Google has. I do not believe they do. MSN has not, imo - Yahoo, not either.

[Grumpus]

I answered this a bit in the other thread, but there are some new errors here...

lowered the link power of each link based on how many external links a page had

Internal. External. Doesn't matter. It's divided amongst all the links on a page. I don't see a reason to change this, either. PR is a "clean" starting value for a page and it's also a good tie breaker when ranking a page. Beyond that, there are better things being employed nowadays to do the other things that PR used to do.

Now, I definitely agree with your logic here - for ranking pages, the number of links might not be a good measure of the value of any given link on a page. But I still don't think I'd alter the PR value of it.

One thing that you need to keep in mind is that links aren't one-way anymore. A link out tells the engine as much about the linking page as the page being linked to.

With Topic Sensitive Pagerank, it would make sense to divide the power amongst the llinks to and from other pages in that same topic rather than by every link on the page.

With various other things, there is plenty of evidence that more data and information is attached to the link itself (and how it is represented on each end of the link), which is a big change from the methodology that you're considering here where a "page" has the value attached to it from the links it has. Now, in many circumstances, the links have the value attached to them from the pages they have.

G.

[randfish]

Grumpus - Good point on PageRank formula, edit made.

Maybe you could expand on your point about the links getting value from the pages "attached" to them vs. page value from links. I'd like to hear more.

[Grumpus]

There's not a lot more that I can say for the simple fact that I don't know. True, that new patent gives clues. And some of the patents and concepts we were looking at last year gave us some clues, too.

Think of it like a map. Pages are the towns. The roads are the links. In the past, Google was assigning value to towns based upon the roads that came into them. Roads from bigger towns are bigger roads and thus the value of the town that bigger road goes to is given more value. That's all fine and dandy, but we can learn as much (or more) from a map by studying the roads themselves, and not the towns that are on it.

So, how do I know this is happening?

The patent dealing with the age of links speaks specifically about, well, about keeping track of the age of links. Sure, it talks about keeping histories of pages, but it talks about keeping histories of the links themselves, as well.

And, when we look at the technologies in the realm of the Semantic Web, we know that what's important in learning about these things isn't the pages themselves, but how the pages relate to each other. It's those relationships that tell us everything, so we need to track those relationships and not just the pages themselves. And, of course, those relationships are the links themselves.

G.

[invision]

So is it possible for a page from site A linking to site B to have an effect on an outbound link to site C from site B? (Make sense?)

(arrows represents a link)

Site A page -> Site B page -> Site C page

(Is Site C effected by Site A based upon not only their content of all 3 sites, but the link themselves?)

So not only are the relationships between the pages are being considered, but the relationships between the history and data of the links between these pages are considered. So does this ultimately lead to a trickle down effect?

[AbleReach]

One thing that you need to keep in mind is that links aren't one-way anymore. A link out tells the engine as much about the linking page as the page being linked to.

It's about time.

Link exchanges are a simplistic, manipulative PIA. RIP.


Elizabeth

[Grumpus]

Invision - maybe. Can't prove it yet, but there is some positive anecdotal evidence bouncing around that things are calculated across at least two levels as you show, and maybe even further.

G.

[randfish]

Grumpus - I love the road analogy - literally, I love it. I don't think I've ever heard the relationship described so well and the meanings are infinitely useful. Bravo!

Was this your creation, BTW?

Link exchanges are a simplistic, manipulative PIA. RIP.

SEOmoz.org exchanged links with Cre8asite when I became a moderator, Google & DMOZ exchanged links, my mom and my sister exchanged blog links - there are plenty of link exchanges that don't fit into this narrow category - that's the whole discussion in the reciprocal links thread (from Tues.).

I agree that the general - link to me and I'll link to you - spam emails that I get 30 of each day are useless links (hopefully), but just because a link is exchanged, doesn't mean it has no value - these things started for a good reason.

[bwelford]

I believe my view is in line with that set out by Grumpus and by invision.

I'm not sure that all the major search engines will handle this question in the same way. As has been pointed out, we're now talking about something different than the mere passage of PageRank through to referenced (linked) web pages. It's the amount of extra authority that the linked page gets as a result of the link. It's been noted elsewhere that Google looks not only at the link text but also at the text near to the link.

I'll illustrate my view on Google by discussing a simple example. Suppose you had three directories listing different types of manufacturing companies. Thinking about the web page for Engineering manufacturing companies, we'll look at typical entries in each directory. All three directories have say 100 links on this particular page.

Directory A gives only the company name and the URL. A typical entry looks like this

WilsonMachine    URL  http://www.wilsonmachine.com

Directory B gives a few words describing the type of manufacturing.

WilsonMachine, an industrial gear manufacturer URL  http://www.wilsonmachine.com

Directory C gives a short paragraph describing the company.

WilsonMachine is one of the leading precision gear and rotating machinery manufacturers in North America.  It has been successfully meeting the needs of its mechanical power transmission customers for over 90 years. It can handle projects that few other companies in the world can handle with the same accuracy and quality. It is ISO 9001 certified and a long-standing member of the American Gear Manufacturers Association.  URL  http://www.wilsonmachine.com

My view is that C gives much more relevance or authority to the link than B, which in turn gives much more authority than A. I also believe that this amount of authority will largely be the same for both B and C if the page contains only 20 links or, as suggested above, 100 links. On the other hand, the authority in the case of A would depend to some extent on the number of links on the page.

[invision]

Then you can also compare to see if Directory C content is on-topic with the site is links to (obviously in this case it is) but say it wasn't... would that link be just as powerful?

[sanity]

Excellent analogy Stock - thanks for sharing.

[travis]

The PR of a page depends on the incoming inbound links to begin with. The algorithm calculation needs something to begin with. And the algorithm is a mathematical one, so we should not lose sight of that fact. I know people talk about keywords and natural links, but in fact all of those effects (I think) are applied later to compensate for the hard line nature of mathematics to begin with.

The link map structure of the site affects the distribution of PR more than the number of internal and external links.

It changes according to a massive number of variables as the calculations take place.

If the internal pages linked back to the home page only
If the internal pages linked to each other
If the internal pages linked to an external link
If the external links linked back to the home page ( ie reciprocal links )

The question is "What values do links give off as a function of number ?"

Its a very hard question to answer. But here is an attempt.

I like the water pipe analogy.

Links are a series of water pipes and the links flush water through the system and the Google calculations are simply pressure measurements.

If one has no external links, then the system is water-tight and all your incoming PR remains intact and the water circles around your internal links only and never leaves. You can spread the PR around according to link structure patterns to change the relevance of each and every page.

If one has links to external sites who dont link back, then as a result, over a number of calculations, then your system will lose pressure, so to avoid too much loss, people start putting their external links off little pipes, so the leakage is not great, and more then compensated for by the greater inflow to begin with.

The world has, say, 5 billion websites and there is only so much water to share around. PR is a finite resource.

If you link internally to your site, and dont share it around externally, then it does not matter how many internal links you have. The PR of your system is not going anywhere. You merely setup a sharing regime of the PR that you have amongst the pages you think are relevant.

If you have heaps of external links at the top of your home page, then yes it matters. Unless the water you lose from those external links is not returned straight away or replaced by another source, then this will cause the pressure of your system to drop, and your PR or water pressure wil drop.

To answer the question, I think it depends on where you place them in relation to the larger flow of PR within your system, and it also depends on whether you get the flow back if you let it out.

This helps explain why some sites with lots of outbound links rank so well.

Its because they get a lot of it back and also replace it with other sources through the introduction of links pointing to their content.

[Econman]

The water pipe analogy has some merit, but I don't think it fully describes the current state of the art.

My impression is that within Google's system, at least, good outward links can provide an indication of quality for the site or page that has those links. In other words, increasing the number of (on topic, high quality, directly relevant) outward links on a page or site may actually boost the perceived value of that page or site (I'm not sure if the site is affected, or just the page).

At a minimum, Google must realize that ecommerce oriented sites tend to have very few outward links; and the exceptions are likely to be links that were created for reciprocity purposes. At the other end of the spectrum, blogs and forums tend to have lots of outward links, including a mixture of true "votes" and spam.

So, it would seem relatively easy, and useful, from an SE's perspective to pay attention to the number and type of links leaving a page and/or site; if the SE finds a site with very few outward links, and those links are concentrated on just a couple of pages, this fact, in conjunction with other data, may allow the SE to detect whether or not those links represent valid "votes" for the relevance of the site at the other end of the link.

[travis]

Are you saying that theory doesn't hold water ?

[Econman]

The water pipe analogy is useful, but incomplete. To be accurate, you need to add to your theory some sort of a process whereby under certain circumstances, pipes leaving a site effectively add more water pressure than they take away.

[travis]

Mate, excuse my really bad jokes.

Welcome to the forums Econman. I am supposed to throw a waving thing in here, but I dont know where it is.

Definitely a confusing part of the algorithm that I have not quite got my head around yet, but from my own observations, I know you are right.

I like the distinction between e-commerce and blogs. I guess the next question is related to the quality of the person that you link to. There is an opportunity in the algorithm calculation to measure that, so perhaps that is an advancement over the original PR algorithm.

What about linking to W3C XHTML and CSS Validators. Good or Bad ? I nearly did it the other day for a customer, but the logos looked like crap and I was concerned about losing PR off the front page as they would have taken up 66% of the outbound links.

W3C. To link or not to link....

[gemini]

Water pipe analogy is a good example, but why would you limit yourself with 5 billion websites.. ? As we know page rank refers to a page and not to a site and every new page that google knows about already has its own PR which is (1-d) according to the original formula and the amount of web pages is growing every day.

Also, when you think about road maps - they lie in the same plane and links are not - roads usually have two-way direction and links are mostly one way vectors (how often you see links back from the exactly the same page that being linked to). So, in this way water pipes are closer to me. If they were in the same plane it doesn't matter how wide the pipes are - the water level will be even everywhere. In our case the leak has obvious direction, so there are different levels between outgoing and incoming points - in water pipes they have to be in order to pass the water, but in vectors, levels don't affect direction. This sounds like pure math and originally it is (I am not good with math though), and I agree with Grumpus that a value of a page attached to a link, since vector is a "movement of a ..." from one point to another.

You should look at PR as vector value - this is how it referred in the original works and studies. Of course not all the rules of vector apply to links and PR, but I think it worth looking into it just not to get on the wrong track from the start. Also, here is a good review of topic sensitive PR calculation work that has been written before and may be some of the aspects will be implemented by google
http://pr.efactory.de/e-pagerank-themes.shtml

I also found this work http://dbpubs.stanford.edu:8090/pub/1999-31
and this one which is a separate study http://informatics.indiana.edu/fil/Class/b...first-draft.pdf

The stanford's work is pretty interesting from the standpoint of using resources to calculate page rank and rank the page; I'm not sure how it limits their abilities in changing algo and increasing calculations and processing time etc since they use low scale machines, but really interesting..

..kinda got lost in my own talk.. didn't really want to make a point, and I didn't answer any questions, but may be someone will come up with a good idea from all that stuff I just wrote if any of this makes sense.

[gemini]

btw, just found all the papers by Taher Haveliwala http://www.stanford.edu/~taherh/

and here is also interesting link http://labs.google.com/papers.html

Krishna Bharat - Publications (Principal Scientist at Google Inc)
http://searchwell.com/krishna/publications.html

[travis]

Its all good.

Thanks very much for the great post and the links. Its a very useful addition to the discussion and at the very coalface of what every designer /developer is trying to achieve.