21 replies to this topic

[Jill]

I think over the years, Randfish, we've all seen that you just can't optimize via software or a program. You'd certainly think it would work, but it just never does. If it did, those types of software would be worth their weight in gold and everyone and their brother would use them.

[DanThies]

If I hear you right, you're suggesting that the use of term weight, or any other calculation of a statistical formula for that matter, is not useful unless you know precisely how the SEs are weighting them in the overall algo (and which specific pieces get +/-).

Yes, I'm suggesting that if you don't have a pretty good idea of what the weighting is, you don't know whether your math is any good at all. We all used to do this stuff all the time. It's called reverse engineering. But as "on page" became less important, and "off page" and "website" became more important, reverse engineering became sort of pointless because you can't just compare pages that way.

In my experience, examining actual ranked pages is far more revealing than throwing them into some kind of statistical distillery.

Maybe, Dan, you can tell us why, at your level of knowledge, you don't use these (and possibly even hint at what you do use ).

I build pages to serve their purpose with the people who are going to visit them first. Natural copywriting, using relevant search terms supported with basic structural elements like headings, links, etc. If I need a page to rank better, I support it with internal links. If I still need it to rank better, I support it with external links.

BTW, you can read this if you want a new mythology to play with: Patterns In Unstructured Data

We'll be up in Seattle at the end of March, BTW, and I owe you a beer.

[Scottie]

I thought this point in Randfish's blog was excellent:

Engineers helping SEOs understand why well-written, natural language is optimal for SEO thanks to advanced sentence structure analysis and its use to tag spam pages.

It's an excerpt on why SEO's and search engines should be friends and not enemies.

[randfish]

We'll be up in Seattle at the end of March, BTW, and I owe you a beer.

Dan, you're are dead wrong on this one... It is I who owe you a beer, and more. I would love to see you when you're in town. PM me when you know details.

As far as the thread topic goes, I think I'll continue to measure this (along with all the other items too). The reason I think it's so valuable is because I recently used the term weight to check a site's homepage against the top 10 documents at Yahoo! and simply by changing the term weight of a phrase (mostly by shortening the document), I was able to boost the ranking 3 positions (from 7 to 4). Obviously, I was in a rare situation where everything else was equal (you know how people say, all else being equal - that was me) and this pushed me up a tad.

I also think it's critical for SEOs to understand that keyword density is not part of an IR system's measures (unless it's for filtering spam).

BTW - That link makes for some great reading, but (as I'm sure you're aware) LSI/A is being touted as a very UN-likely part of any major SEs algo right now over at SEW.

Edited by randfish, 08 February 2005 - 08:54 PM.

[DanThies]

BTW - That link makes for some great reading, but (as I'm sure you're aware) LSI/A is being touted as a very UN-likely part of any major SEs algo right now over at SEW.

Exactly, it's entertaining, but it's got as much to do with search results today as Hilltop.

[randfish]

Dan & crew,

Spending some more time on this subject recently and I made a discovery that should have been obvious to anyone who has taken college level mathematics (which I have, but promptly forgot).

The Term Weight formulas all carry the same system of normalization - i.e. Freq. / Max Freq.

This means that the best any page could achieve when targeting a phrase is simply to make sure that the most common term/phrase on the page is the targeted term/phrase.

So, for example, I could not achieve a higher term weight by repeating "cute bunnies" on my "cute bunnies" optimized page more times than the maximum number of unique 2-word phrases.

If I had paid more attention to this in the past, I would have realized this and could have saved myself and others a lot of confusion on this subject. I hope this can help out others who may be thinking about kw density and term weight trying to find an "optimized formula" for on-page targeting.

[Peter Hoggan]

Hi randfish, by only looking at 20 pages your sample set is ineffective and incomplete, the inverse document frequency requires that you look at the entire document set. In a competitive market the top ranking pages will have massive support from external links. Take the miserable failure issue for example, the Whitehouse page would create a tfi measurement of 0, and therefore a term weight of 0, although it ranks #1. I am not saying it is a competitive term, just highlighting the influence of links.

By comparing a document against a small set, running a keyword weighting calculation and then upping the keyword count of your page to achieve a higher tfi, all you succeed in doing is altering the keyword to document ratio which in effect is localised Keyword density.

You mentioned that you achieved a result by shortening a document, you may be confusing cause and effect. The benefit may have been achieved by realising greater proximity due to the shorter document. This is why I do not think your means of measurement will work; the results you are trying to analyse have already been ordered and weighted by other factors such as links, position, proximity, semantics etc. You do not have access to the raw data that you require nor the computational power to analyse it.

I agree that it is an absolute must that SEO’s understand the concept of term weighting, but trying to analyse a set of results that are ordered and weighted by many other factors in the way you are attempting is not possible.

Thanks Dan, I didn’t realise you could use –intitle:keyword, so obvious really. You learn something new every day in this game, that’s what makes it so much fun!