Although it's a bit hard to follow, Dr. Garcia (aka Orion), who operates the Miislita site and blog, has a great blog post on the Latest SEO Incoherences (a clever acronym pun on LSI). He takes issue with SEO/M firms who advertise that they use LSI or SVD tactics to make their services more competitive and valuable to clients:
Their current stop appears to be LSI and SVD, two scientific acronyms that stand for singular value decomposition and latent semantic indexing. In an attempt to sell whatever they sell, one can see now a crew of search engine marketers reciting these terms in their search marketing conferences, forums, blogs, press releases and any other marketing channel one can think of.
They are all over the place: from US to Russia to India to Mexico, claiming that they sell "LSI-based services" or do "LSI-based website optimization", "LSI-based link building", have patented an "LSI tool", "reversed engineered LSI", yadda, yadda, yadda.
Such "bottles of snake oil" are easy to sell to ignorants, especially when clients still have hard time figuring out what SEO or HTML stand for.
Their pattern -talk about something, explain nothing- is obvious. When not quoting each other hearsays, they paraphrase or copy/paste lines from the LSI literature, speculate about how search engines implement LSI or even come up with their own theories.
Their trail - combine science with marketing lingo- is obvious, too. They will put online any argument that will help them to sell something. Quite convenient, hugh.
Still, you will not find these unscrupulous marketers any time soon providing how-to stepwise calculations explaining how LSI or SVD works.
For those who are interested, Dr. Garcia also has a tutorial that fully explains the myths and examines the IR uses for LSI & SVD. I'm struggling to comprehend it (and will probably seek some help from Si when I have the chance), but Dr. Garcia's point about the limited application of this science towards the marketing or rankings of a website from a search-engine external perspective makes a very clear case.
I see at least a few dozen companies attempting to market in this fashion.
Thank, Rand for blogging about the tutorials.
Mi Islita target audience consists of IR students, university researchers and search marketers interested in the scientific aspect of IR and search technologies. The goals of the tutorial series are to show readers how things are done.
One thing is talking about LSI in forums, blogs, podcasts, etc and something different is providing readers with stepwise how-to instructions. In this way readers can replicate the LSI calculations, test things, and in the process, stay away from SEO myths and "LSI based" Snake Oil Marketers. The later is a sector of SEMs that is giving the search industry a black eye before the IR community. Some times I understand why some IRs perceive (incorrectly) all SEOs and SEMS as a bunch of spammers and mere sales people.
About the SVD and LSI series, in Part 1 readers can quickly remove from their head what is not LSI. Part 2 and 3 discuss SVD calculations and require of a linear algebra background. These are necessary if one wants to replicate SVD. That's why a warning was placed in big bold red text in Part 1. Some readers have tried to cut corner and skip the math only to find out that they need to go back.
Part 4 will provide stepwise instructions using a specific LSI algorithm and show how documents and queries are actually scored. In this way, some interested can replicate the calculations. Part 5 discusses several software for crunching LSI data and code lines. I might add additional material and compare LSI with other techniques (nonnegative matrix factorization -NMF-, PCA, latent dirichlet allocation -LDA-, etc.)
I disagree that LSI is useless. LSI is very useful if one knows how to do an LSI run and interpret results. Part 4 of the tutorial covers this.
As with any IR algorithm, LSI is not The Answer for marketers and has some limitations, drawbacks and workarounds. In fact, LSI implementations based on mere word counts can be diluted (gamed). This occurs when spurious induced similarity is injected into collections. This is discussed in issue 1 of IR Watch - The Newsletter.
When some thought that LSI was The Answer, here it comes NMF and LDA. Would these be the next stop for the "snake oil" marketers? Time will tell.
Regards
Dr. E. Garcia
Just a heads up but that link is broken...pity! :)
Dr. Garcia moved his blog here: https://irthoughts.wordpress.com/
What makes LSI unbelievable is that it is utterly useless - it actually makes natural writing LESS valuable, and generated writing MORE valuable.
A handwritten piece looks less legitimate than content formulated through scraping the body of information used to do the analysis (ie: what blackhats do every day).
Think about it. If you use a stack of papers to create a theory - what is going to fit that theory better, something newly written, or something taken directly from several of the papers in that original stack?
I just read through those tutorials and my head hurts very very bad now. I have not dealt with math like that in years - although it was kind of refreshing in a "see, I can still run a marathon holy crap im going to die" kind of way.
Nice post Rand. We all need to do our part to make it as difficult as possible for the snake oil salesmen.
On a side note you aren't alone in your struggle to follow Dr. Garcia's tutorial. I had the privilege of spending a few hours talking with him in San Jose last year and I only caught about a third of our conversation on fractal theory. His tutorial reminds me of that discussion; wonderfully insightful information with dots that require your full concentration to connect.
lol just found it funny reading Rand's post
Then clicking through to Dr. Garcia's tutorial and the first thing I read:
p.s. found graywolf's icon so clever had to make one myself :D
Thanks for posting this.
Snake oil SEO drives me nuts as well.
I have heard the buzz word before and did a little reading about it but eithor didn't get it or it didn't seem like it would be something that I would apply in my daily life. I'll check out the podcast and article.
Greg Niland aka GoodROI did a podcast earlier this year where he talked with a Professor Michael W. Berry from the University of Tennessee Dept. of Computer Science and they discussed LSI. Thank fully it was simple enough even I could understand it without using a translator
https://media.webmasterradio.fm/episodes/audio...
Tangent Ahead: so Rand when can we expect to hear you podcasting ...
I think it's not very far away... Be afraid, be very afraid!