Welcome to WebmasterWorld Guest from

Forum Moderators: open

Message Too Old, No Replies

Latent Semantic Indexing

So has anyone seen any evidence?



8:31 pm on May 3, 2004 (gmt 0)

WebmasterWorld Senior Member 10+ Year Member

There was a lot of discussion about Hilltop (which I don't understand) and Latent Semantic Indexing earlier this year when the so called "money keyword penalties" were a hot topic. Supposedly LSI and Hilltop were an explanation of how sites could have dropped without a penatly.

Anyhow, LSI works on the premise that the more words you have on your page which appear on other pages containing your keywords, the better you will do. This means adding similies and words associated with your topic to your pages should help your ranking.

I decided to try it out on one of our key pages which was relatively well indexed. I spent an hour writing a paragraph containing all the words I thought would help.

Anyhow, the upshot is that 3 weeks after I did that our position has barely changed. In fact I have never seen things so stable. Has anyone got any strong evidence that optimizing for LSI produces results?


4:44 am on May 4, 2004 (gmt 0)

10+ Year Member

Personally, I have not had luck with LSI, but a friend of mine has. He basically changed his titles to keywordstem1 keywordstem2 and his positions improved dramatically several weeks later. However, it wasn't a perfectly controlled experiment, and his keywords were not extremely competitive. IMHO, using LSI seems to be especially difficult for Travel and Real Estate keywords.

I am seeing more of a Hilltop algo in action. I must admit that I don't know much about it, but to me the practical application seems to be that you need to get quality links from authority sites in your area. For competitive keywords, that is sometimes hard to do. I have just started to achieve very limited success with this (today, in fact).

FWIW, my gut feeling is that three key elements of the new algo include Hilltop, LSI, and an OOP/filter. The OOP/filter seems to be much more forgiving to sites that pass the "Hilltop Test", which is one reason why many people are able to give examples why there does not seem to be an OOP/filter.

Sorry that I do not have more definitive examples, but I wanted to add what I could.


5:12 am on May 4, 2004 (gmt 0)

10+ Year Member

LSI is strongly in play.I've successfully implemented on my site. Research and insert all the related keywords which your page is all about. Don't forget to implement the link building excercise in accordance.

Hiltop is also in play simultaneously. If your site is identified to qualify for a keyword and is linking to an affiliate site(see hiltop algo) on the same topic, both of your sites are sure to suffer.

So keywords present in the text and link building(with whom, using what keywords) is the key.

So Slydog have a look at your site again, and if everything is all right, i think you'd wait a bit to get desired results.


6:33 am on May 4, 2004 (gmt 0)

WebmasterWorld Senior Member caveman is a WebmasterWorld Top Contributor of All Time 10+ Year Member

SOD, did you say 'page' or 'pages'? ;-)

Yes we believe the LSI explanation, with a relatively high degree of certainty. We have adjusted accordingly, and are generally happy with the results.

That said, we feel that the only really good set of SERP's we've seen from G since pre Florida occured in the very early stages of Brandy. I've noticed steveb mention this a number of times too.

Our own observations, including some site based research we have (we have search on our sites) suggest that the pre-Florida results were better than what G is feeding searchers now. Searchers come back more to the same search than they used to, because they are not finding what they want with the CT's. (Less relevant than before IMHO, AFAIK, etc...)


10:01 am on May 14, 2004 (gmt 0)

10+ Year Member

After reading this thread I used tilde~ to get words which Google regards as associated with my keyphrase, widget1widget2, inserted these words into the text once onlyon the index page and a few days later the site sank 4 or 5 slots. A coincidence?
By inserting these words ( and not removing an equivalent number of keywords) is there a chance you might trip the OOP filter?
I was also puzzled that an associated word 'medicine' was popping up in the ~ search but 'medicinal' was not.
Doesn't stemming apply with the ~tilde?
Or, have I got it all wrong?
Help would be appreciated.
Oh, yes! Once last point. I notice Yahoo is using tilde~ too. I thought it was just a Google thing.
And, does tilde ~ work with a two word phrase?


11:05 am on May 14, 2004 (gmt 0)

10+ Year Member

I see it in action and have used it successfully for our keywords. We went from nowhere to number 1 for over 6,000,000 results.


11:42 am on May 14, 2004 (gmt 0)

10+ Year Member

Yes, I'd defo say an LSI-type algo is at work... but it's important to remember that it's in the context of other factors too.

PR for one; I've noticed when doing my own studies that, for instance, where a keyword might apply to many topics, the most relevance is given to the "range" of topics where there are well defined "islands" of high PR auhority sites.

IMO, this is how the ontology of a given topic is being defined.

Using the tilde, as we know, returns a range of associated words that G "expects" to encounter. How does it know? From looking at sites with the highest authority.

It's my belief that if you "test" for a keyword and are not seeing a well-defined result in your topic area, ie... other completely different topics are returned but yours isn't (not your page, the topic - I'm not talkin about pages here), or is very low in prominence, then that keyword topic is ripe for someone to target it, and in the process become an authority too, thus putting the topic on G's "map" so to speak - Very altrusitic, you'll be helping others too!

I think I can get away with using this example...

"Compression", is a word associated with many topics from plumbing, neck injuries to zipped files etc...

However, there is an associated topic in there (which I haven't revealed) that doesn't return - G "knows" about a qualifying secondary KW but misses the topic completely and returns the result coverin the subject where there are high-PR authority sites - Simply put, the "missing" topic doesn't have enough authoritative representation on the web from which G can draw unique related KW's to associate with it.

I'm not sure that makes sense but it's difficult without breaking the TOS.

I've seen this in reverse too where a german language brand name has become the authority and G now expects to find it discussed even within generic topics!

I also have a hunch that linking to the authority on your written subject helps G affirm what your page is actually about and so gets "credit" for that - I've experienced some positive response from doing this small-scale but it's still early days.

[edited by: TheWhippinpost at 12:31 pm (utc) on May 14, 2004]


12:26 pm on May 14, 2004 (gmt 0)

10+ Year Member

How do you define 'authority'?
The site with the greatest PR? Or, the site highest in the Serps?
And, if either of these two is you principal competitor, what then?
Another question: you haven't answered my other point.
Does the tilde work with two-word keyphrases? In another thread someone claimed this wasn't a goer.
What do you think?
And, what about adding tilde words to your text without pruning the existing keywords? Could you be in for an OOP?
Any views on that?


1:15 pm on May 14, 2004 (gmt 0)

10+ Year Member


How do you define 'authority'?
The site with the greatest PR? Or, the site highest in the Serps?


And, if either of these two is you principal competitor, what then?

Yes, good question.

If, linking to an authoritative site on a given topic does lend more "meaning" to your own page, then theoretically, that site will always attract a higher status in G's eyes as everyone will (theoretically) want to link to them too. If the topic is a narrow one you may be knackered - It's the one aspect of this that, if true, irks me.

However, a thematically-written site (or site-section) may be the answer as this adds depth and breadth to the subject-area by defintition. Either that or find a niche by investigatin the related words that the tilde throws up.

And, what about adding tilde words to your text without pruning the existing keywords? Could you be in for an OOP?

Will adding related words without pruning KW's attract an OOP? Not necessarily no.

But what I believe it does allow you to do is to prune the KW's down in the presence of related words - If there is an LSI-type algo at work (and I believe there is as I say), then you don't really need loads of KW repetition - In fact, it could be argued that by overly targetting one KW thru repetition, you're, in turn, overly narrowing the meaning of that page.

It's been posited that G wants to take you to the most relevant site so that the user can hopefuly find the material they're looking for... even if it's not directly on the page served up in the SERPs. An LSI-based algo would/could do that.

Does the tilde work with two-word keyphrases?

Have you tried it? Worked for me though 3-word phrase didn't appear to work.


2:29 pm on May 14, 2004 (gmt 0)

10+ Year Member

Thank you whippinpost for your comments.
I can't make up my mind about the effectiveness of two or three word tilde~ searches.
I searched on REAL ESTATE
And this threw up HOME and HOMES which is what I expected. Yesterday when I did an identical search I also got HOMES FOR SALE. So far so good.
PROPERTY cropped up but way down the list (I only looked at 6 pages of Google)
If you put in REAL-ESTATE you only get REAL ESTATE, no homes or property.

I then did another search on HERBS FOR HORSES, knowing that the common word 'for' would be excluded.
The result, excluding the keywords, was PLANTS and very much down the list HERBAL MEDICINES.This seemed a good result.
But, if you now searched on HORSE HERBS, you got EQUESTRIAN and PONY, a bad result.

Experts on semantics will probably be laughing at my ignorance.

Do you fancy making a comment on the last two searches, and giving a stab at why they are so different?


4:44 pm on May 14, 2004 (gmt 0)

10+ Year Member

Just to add to the confusion.

Doubled barreled words with an independent meaning are odd too.

If you look at ~uptime it produces zilch besides "up time" (notice the space) but ~downtime delivers "outage" & "down time"

What I find interesting is another very very common doubled barrelled word which would give you 15m+ results if you searched for it finds absolutly nothing.


9:00 pm on May 14, 2004 (gmt 0)

WebmasterWorld Senior Member steveb is a WebmasterWorld Top Contributor of All Time 10+ Year Member

~horse herbs

will get related words for "horse" plus the specific word "herbs", not related words for "horse herbs".

I don't think you can get the thing to work for two (or more) word searches.


Featured Threads

Hot Threads This Week

Hot Threads This Month