Keyword Density - A Bogus Metric

<removed>

Let's lower the rep-spreading limits so I can give Ranfish more rep!


I'm glad to hear this, because I'm working on a new site, and I'm having a really hard talking about the thing without saying the thing too many times.

Hard to talk about the benefits of red widgets without saying "red widgets" about a dozen times. Thankfully, the sites I'll be competing with mention red widgets quite a bit more than that.

hey-ho,

I have no idea about the answers to the question of it being bogus.

It definately isn't in Knowledge Discovery in Databases (KDD), digital libraries, bibliographical libraries, and almost every IR system, its valuable knowledge when all you have is some text, its actually a good place to start.

This information allows us to quickly assess regions within the documents that are “hotspots” insofar as they exhibit high query-keyword density or occurrences of a certain word or word combinations (Burstiness). Its quite important for automatic news for example, to know what topic is hot.

There are many different ways of getting those figures:
Hanning windows are one way,and used to weigh more strongly the data values at the center of the finite range, and applies less weight to the values at the extremes of the range. You can use Apply Bayes’ Rule and with non-parametric density estimation. The distribution of the number of keyword vectors in a Web page can tell you that too.

What we're looking for is the frequency distribution patterns.

If you like you can use the ratio between keyword number and total word number in a page and domain for a quick and simple estimation, as is commonly used in SEO.

I know its invaluable in my work. But obviously not used on its own.

Xan -

You're saying it's used to identify relevant documents, but not neccessarily for rankings - is that right? It makes good sense - a document with lots of mentions of a term is probably at least on the subject!

Thanks Randfish... great post. I have never paid attention to KW density - but more by luck, wanting to write naturally and being lazy. Also, I get LOTS of traffic from words that appear just one time on my pages.

That's right. It's not used to rank anything, why should it be, of course many other factors have to be considered and then those all need to be weighted as well, before these measures are even taken into account in the overall calculations in the wider document space! And it does go on, but I'll stop there.

So...it is used for ranking, ... but in the same way that an exhaust pipe contributes to a car working.

That's a beautiful analogy Xan. So appropriate.

Rand, the myth is eroded already my friend. No more KW densities. This is old stuff. When I started 4 years ago, we worked on keyword densities religiously. I had percentages memorized for the various engines, we were comparing the top results, kicking butt and taking names so to say with the rankings. Then I figured that you could accomplish the same thing with just writing regular paragraphs and bam! I never calculated a keyword density or ran one of the many tools we use to do so.



What for? Would be as useful as a screendoor on a submarine. (Jasonnytc said that one day, and loved it) :-)

Great post guys. I've also noticed that KW density isnt playing a very big role in ranking anymore but I think it is still used as a last resort if Google stumbles upon a site with many pages about the same topic/weight and doesnt know which one to display in the SERPs. Google usually picks the page with an average (natural) density over low or high density pages of the same topic but the site ranks the page in the SERP and not the density itself. I've seen it happen to my pages quite a few times.

I was under the impression, kwd is somewhat valuable and can serve as a tool to avoid keyword stuffing and getting sandboxed. Even though it isn't exactly the first order of offense in SEO, I still like to keep my eye on it to avoid keyword stuffing. I read somewhere that kwd for your keywords should stay around 9-10%, anything over that and you may get sandboxed. Does that sound right?

KWD can be measured as part of spam, but it's really relative to document size, so a specific % can't be given. If you normalize document size, however, I would surmise that KWD over 5% is already risky...

So the practical application of this insight is:

1. Don't worry about keywords at all?

2. Only add keywords until they're more frequent than any other words, then stop (possibly because keeping going will trigger some KW stuffing penalty)?

thanks I think that last one dumbed it down to the right level for me.

Er, well, that last one was actually a question. I was trying to dumb it down for myself and make sure I got it right. May not be though!

Ok.... Guess we will both wait for an answer. Sounded right to me in that perspective though.