So, a little over a week ago SEOmoz put out some information on something they call LDA, or Latent Dirichlet Allocation. Some SEO people seem to think this is a game changer…well it’s not. In fact, the concept is nothing new, but too many SEO folks have been distracted of late focusing on link building and social media to the exclusion of excellent on-site SEO work and creating user friendly content.
In the most simple terms, LDA means having topically relevant pages on a topically relevant site. Perhaps a bit oversimplified, but there you have it. SEOmoz gave the following example: LDA.
In the past, SEO specialists worked on things like keyword density and TF*IDF (term frequency x inverse document frequency) to make sure a page was well written for search engine keyword relevancy. Looking at both metrics can still provide some value, but are not magic solutions to creating rank-worthy content.
LDA dictates that it is equally if not more important to make sure that other, topically related keywords are in the content in addition to the keyword you are looking to rank for. Thus, if you want to rank for “laptop sales”, using words like “portable”, “computer”, and “good battery life” could improve your topical relevancy and help to rank you for “laptop sales”.
So why is this not a game changer? Simple. Because Google does now and has always looked for quality content, written primarily for users and not search engines. If you focus only on keyword density, you may well write a page that no longer appears to be naturally written and that excludes other topically relevant words.
So, as always, write for your actual users and readers, and write around one main topic per page. If you’re writing naturally, you’ll likely do just fine.
The SEO industry going up in arms about this was because of the SEOmoz lead scientist making a math error. He pulled a correlation of something like a .32 from LDA analysis (which became the highest correlation of any ranking factor, I believe) and it got everyone thinking that maybe on-site optimizations were actually more important than we once previously suspected. It turns out there was an error in calculations that they cleared up later and the correlation actually lies at .17, which is about typical for any type of onsite optimization. Of course, one must remember here, that correlation does not prove causation, and strong LDA scoring is likely indicative of other types of strong on-site SEO optimization and link attractiveness as well. Agreed, not a game changer.
Even if they did find a correlation of .32, this shouldn’t be a game changer–Matt Cutts and others have been saying for years that the best SEO is to serve your user. I’m not going somewhere, or linking somewhere, if the content isn’t any good.
What will be interesting to see is how well Google will learn to pick up on artificial link profiles. If they get reeeeeally good, then on-site factors will become more important again because they’ll be one of the few things that SEOs can change without getting in trouble.
Pingback: Latent Dirichlet Allocation Optimization