SEOmoz LDA tool experience?

[RyanKent]

http://www.seomoz.org/labs/lda

Does anyone presently use this tool as a regular part of their SEO analysis?

Does anyone have any updates regarding this tool? It is still in the lab despite no updates for over 6 months. Is it being retired? Or promoted to production?

If you do use the tool, what % of relevance is your goal when optimizing pages?

[randfish]

I still use it on occasion, but we recognize that it was far more of a research project than anything else. I will say the Virante implementation might be worth considering, as it's both good for thinking through topic modeling stuff and potentially helpful on the pure keyword research/targeting front, too: http://www.virante.com/seo-tools/lda-content-optimizer

Our goal is to bring aboard some more folks with serious topic modeling experience and, hopefully, find even better methodologies and implementations for this project so we can eventually start grading pages and making recommendations closer to how Google/Bing do.

The engineer who developed the tool is now working inside Google, which we'll take as a compliment, though it's also a bit sad to lose him

[RyanKent]

Thanks for the tip on Virante and taking the time to share this update.

[eggdaddy]

Hi Rand. Firstly I only recently discovered your LDA tool. Seen the ups and downs surrounding your release also. Overall, I think your definitely trail blazing with this. At the very least its something for the seo world to experiment with and in my opinion fits exactly with googles intention to deliver relevance. Now with that said, I dont yet know if there is anything in this as I can not run the tool. Or at least i tried to ask a question on the original blog page. It simply doesn't seem to work. Here's the issue I am getting, in the UK. If you run the tool over and over, each time you do so, even if your content remains unchanged, each time the % goes down. So to me this suggests its broken. Did ben screw it up before leaving for google?? I think you could argue you may be getting close to something important in their algo if the big G took your coder away?

Anyway, all I want to do is test this out as its pretty mouth watering. Also I note since the LDA tools release Google has withdrawn its related term finder tool ( the name i forget ) so can you recomend a google work around to understand what google considers consistent and therfore within you target term topic model?

Hope to hear from you

Cheers

[randfish]

Yeah - there could indeed be issues with it. Right now, the team's focus is all on Linkscape - improving quality/size/freshness of the index. Afterward, however, we do plan to return to LDA (or other topic modeling projects). I suspect by end of 2012, we'll have something out.

[eggdaddy]

Hi rand, You guys caused quite a stir with this tool. Kinda would expect you to know if it broken? Real easy test for you to confirm if you get that effect. Just try the tool without changing content between requests. Could you do that old chap? Also word on finding those related words in googles eyes now Google Sets is no longer with us?

Thanks

[KeriMorgret]

Eggdaddy, it's Rand's wedding anniversary and one week a year of vacation right now, which is why he may not have answers to everything off the top of his head at the moment.

[RyanKent]

We need to recognize the LDA tool is not part of the SEOmoz toolset. It exists in the SEOmoz Labs toolset. The page title clearly indicates it is part of the "SEO Research and Testing" area. The tool even clearly states "The tool is in flux. We say that about all of our tools in labs, but this time we mean it. Please continue to send issues to ben@seomoz.org".

Rand even went further to explain how the tool's lead developer moved on to Google, and further development has been delayed until the end of the year.

Based on this information, your last reply seems rather unfair. Perhaps you might reconsider your position on this topic.

With the above noted, I see the LDA tool as having great potential and hope work continues on it, or another similar tool which offers detailed content analysis.

[eggdaddy]

HI Guys, I dont mean anything other than to to advise the tool may be broken? No?

I cant see how this can be unfair Ryan. I think you may have miss understood what I am flagging? I know all the caveats and that its in Dev im cool with that. But if the tool gives a consistently degrading score for a page that has not changed, Q) is this what you get or is it just me? If you were aiming to use it to optimise you would expect to hit the req a number of times and it would seem there is a natural degredaton score simply for asking over and over. Is this just how it works or? I need some brass tacks answers to this without getting into whether its fair or not to ask if your tool is broken? :s

If this is what it is supposed to do, then fine but its kinda a chocolate tea pot if that's the case. Dont get me wrong, have loads of respect for the Moz.

[randfish]

Hey there - sorry for my delay. Just got back to Seattle.

Basic story is - there was fluctuation and that downgrading on successive runs even from the start. Ben explained it to me, but I could never fully wrap my head around it. From memory, I recall that it tries to sample a subset (as calculating across the entire corpus would be too computationally expensive). An artifact of that process is that the scores tended to vary wildly. Ben made some changes, and this produced the less wild, but now declining scores on successive runs.

Unfortunately, there's no one at Moz right now with the time to invest in understanding/fixing/repairing. We think we've got bigger/better plans for something much more useful along this trajectory once the Linkscape work is done, but that won't be for some time.

In the meantime, we maintain the tool "as is" in Labs. If you find it useful, great, but Labs is an area where we don't do maintenance of have guarantees on availability/accuracy.

Hope that helps!

[eggdaddy]

Rand, Congrats on the Anniversary. And Yes very helpful answer. Just was unsure if it was a local quirk. But as you explain, this is exactly what it does. My page started at around %30 and after a time trying to improve that by adding some of the theory my score is now %19 but I have run it perhaps 10 times and I think i was seeing something like a 1.5% degradation for repetition. I think if succeed in adding this to your Linkscape or Site explorer toolset. It could be very potent technology for SEO. Don't forget if your too good at it Google will have to adapt Thanks again for the answer and Keep up the good work, SEOMOZ has really kept innovating and hats off to you guys.

[eggdaddy]

I should say a 1.5% degredation for each repetition cumilatively. So could I assume I have lost 15% as error in your tool? meaning I have a score of %35 now ?? Who knows