Optimizing LDA Scores

16 Sep, 2010  |  Written by  |  under SEO

(There’s an updated version of this article as of 12/19/2010.)

If you have a page about “basketball drills”, shouldn’t it include these words?


What if it didn’t?  How would you find that out quickly?

SEOmoz recently announced (and then corrected the stats for) a new tool in their labs called the “LDA Tool“.  The tool basically takes a search phrase and a page and tells you how much they relate topically.

Unfortunately, it’s not very clear from that report what topics it identifies or what you can do to improve your page’s score. In this post I’ll show you how to just that, and and give you a new LDA Optimization Tool I wrote to make it easy.

Step 1: Identify your Keyword

First up, figure out what keyword you want to improve your page’s rank for. For this test, I’m using the phrase “basketball drills”. I just chose this at random and hoped I’d be able to find a good example site in it. See how that worked out below!

Step 2: Identify the Competition

Keyword Difficulty for basketball drills

A quick way (thanks, Rand!) to do this is with the SEOmoz Keyword Difficulty tool, but you can just do a search and make a spreadsheet too.  To the right you can see what I got for the test phrase:

Exporting this to a CSV and using it as your starting point for LDA analysis speeds things up. You can get the raw CSV here.

I uploaded the CSV to Google Apps as a spreadsheet, and you can see a read-only version of it here (note that this spreadsheet contains my final data as well).

Add your page to the spreadsheet if you’re not (yet) in the top rankings.

Step 3: Gather LDA Scores and Competitor Words

Add columns to your spreadsheet for LDA %, and for site keywords (a really wide column!).  Head to the LDA Tool and type in your test phrase and first site URL, then gather the % and keywords for each of the sites.


Repeat this process for each of the sites. Here’s how my spreadsheet ended up.

Step 4: Find Your Missing Keywords

paste-in-wordsAt this point you should have a good feel for how your page stacks up against the competition.  Now let’s see what words they have that you don’t have, and what words they have in greater frequency than you have.

For the example keyword basketball drills, let’s say that our site is www.basketballteacher.com, which has LDA of roughly 61% and ranks #17 on Google for the phrase. Here’s where we’ll use this new LDA Optimizer tool.

This tool takes your words and up to 9 competitors’ words. It then finds words they use that you don’t, and words that they use more on average than you do, and provides a word cloud showing your opportunities.

To exclude words specific to just a handful of sites, you can specify “Show Keywords Seen on at Least” at the bottom of the form. I used a setting of “4 sites” for my test but got similar results with as many as 8 sites. I recommend playing with this until you’re happy with the results, seeing lots of good words to add but not a lot of extraneous noise.

Step 5: Reviewing Missing/Insufficient Keyword Use

For the sample site, we get this word cloud (on the Wordle link, their clouds are prettier than mine!) for missing words:


So this page related to basketball training is missing words like “rebounding”, “dribble”, “jump”, etc. Lots of opportunities to improve LDA score.  (Note that singular vs. plural isn’t handled well here so some of these words (drill) might exist on the page in other forms that Google would find identical.)

Here are the words the page doesn’t have “enough” of relative to competing sites:


A bit more emphasis on these words might help the site show that it is more topically related to the search phrase.

Step 6 Optimize your LDA Score

The LDA Optimizer tool then provides a suggested set of words and densities to improve the page’s LDA score. Just taking these and pasting them in to your page is unlikely to help, but you should now have some guidance in terms of how many instances of the various words to shoot for as you revise copy.

Adding the suggested words to the site’s existing set moves its LDA from 61% to 86%. So we’ve at least helped our LDA score with a change along these lines. Whether that improves search rank is up to you to test!

Step 7 Check for Higher Google Ranking

Hopefully making changes along these lines will improve your page’s Google ranking, but that’s relatively uncharted territory with regard to LDA today. I’d love to hear your results!

One Response so far | Have Your Say!

  1. randfish  |  September 17th, 2010 at 1:10 am #

    Thanks for writing this up, Glenn – I think it’s a really solid methodology for testing whether the LDA tool and the thinking around topic modeling / relevancy can help boost a page’s rankings. We still don’t know how effective/useful the tool might be, but anecdotal evidence thus far suggests some folks have had really good experiences and others less so. Hopefully more folks will see this and give it a try.


    randfish - Gravatar

Leave a Reply