Machine Learning - Randomness in Search Engine Ranking Algorithms

[DonnieCooper]

I believe, sometimes you may 'deserve' the first position, but get #3. And of course, sometimes you may deserve #3 and instead be #1. All due to a 'randomness factor' in search engine algorithms.

I've been holding this hypothesis for quite sometime. Really, it started when I learned about SEOmoz using machine learning to better investigate SEO best practices.

I suddenly found myself wanting to learn more about machine learning, and the implications of using it for SEO. I'm by no means able to utilize machine learning myself, but I it appears unsupervised learning would have a real chance of cracking search engine algorithms. Hey, even Stuxnet was cracked!

Surely Google/ Bing would know (and account for) this, right?  We can agree they'd obviously prefer a highly skilled mathematician not be able to crack the code.

Therefore, I'm led to believe that search engines use some sort of randomness in their rankings. Maybe not much. Perhaps not all the time. But if a random percentage of search results, had a random variable of sorts included in their calculations... wouldn't that be enough to prevent the vast majority of cracking attempts?

Thoughts, opinions, criticism? Thanks.

[DonnieCooper]

Hey Ryan, thanks for the ideas. Respectfully, I disagree however.

With something as simple as 'rand(1,4)' in php you can generate random #s. There's no telling what Google could do with their algorithm(s). Okay, so it's (relatively) easy to do, but why would they do it? After-all, they have heavily invested in complicated models that do the ranking for them...

Well, they've heavily invested in complicated algorithms, and if cracked, bump- there goes there flagship product.  I'm sure SEOmoz aren't the only people attempting to glean insight using machine learning. And perhaps, others don't like to publish their findings and tip anyone off.

Of course, I'm not an A.I. scientist, so perhaps I misunderstand how unsupervised machine learning works, but I'm under the impression that it very well can be used to logically explain all of those factors. I'm sure there's tons and tons of work that would go into it, but if you had the resources (and some do), wouldn't you do it?

This is why I wanted to post this question though- to get feedback and other opinions, so thanks Ryan!

[RyanKent]

Hi Donnie.

I enjoy your analysis but disagree with your conclusion. I believe search engines are working with an unimaginably vast amount of data. They evaluate hundreds of factors then assign different weights to each. Also consider the source data is constantly changing, and the best anyone can have is a "view" of the data at a given point in time.

There is absolutely no need for any random factor. In my naivety, I believe every last ranking result can be logically explained. We simply do not have proper access to the data, metrics and weighting involved in those calculations. Our educated guesses seem to do a pretty darn good job.

[DonnieCooper]

Darn, I can't delete this comment