Forum Moderators: open

Message Too Old, No Replies

New "clustering" SE

         

NFFC

8:49 pm on Sep 21, 2000 (gmt 0)

WebmasterWorld Senior Member 10+ Year Member



Groups the results into categories [themes?] on the fly.

[cluster.cs.cmu.edu]

Brett_Tabke

9:32 am on Sep 22, 2000 (gmt 0)

WebmasterWorld Administrator 10+ Year Member Top Contributors Of The Month



I tried, but the Jscript wouldn't run (Opera 4) - and for the time being, I am IE'less and Netscape'less (I'm going to see how long I can last).

NFFC

9:37 am on Sep 22, 2000 (gmt 0)

WebmasterWorld Senior Member 10+ Year Member



>I tried, but the Jscript wouldn't run

Don't think it works on a Mac either, to be fair it is in beta, having said that it is well worth seeing.

Added

Wow, try a host: yourdomain.com search with the altavista option.

Brett_Tabke

11:24 am on Sep 22, 2000 (gmt 0)

WebmasterWorld Administrator 10+ Year Member Top Contributors Of The Month



Ok, switching to a machine that has ie on it, I think it is an interesting engine. I can see just how they are doing everything but the javascript.

I can't imagine, the search engines won't have a problem with their setup. Unless they have some back channel agreements.

It is a unique approach.

vivisimo

1:26 pm on Sep 22, 2000 (gmt 0)



Vivisimo is a CMU project that gave birth to a company: [vivisimo.com...] (just a mirror right now). As NFFC asked me I'll give a short description.

Vivisimo clusters on the fly the snippets/titles returned by any kind of search engine and extracts cluster-annotations automatically from the text (in this respect it is completely different from NorthernLight).

It can be hooked up in ten minutes to any intranet/extranet/web search engine and runs in about 60ms for 100 documents (all the code is written in C).

Our main focus has been on generating short, crisp and understandable annotations and that's a key part of the clustering algorithm (the usual way is too create groups based on mathematical properties and then try to annotate them, our approach has been to create groups based on how well they can be described)

Yes it's still in Beta, mostly because of the Javascript used to represent the tree. It's hard to do some complex Javascript that runs on every browser/platform combination. Right now it's working on IE/Netscape on Unix and Windows... we dont have an easy access to Mac machines but it should work on Netscape/Mac soon (IE/Mac is a real problem...)... we didnt try on Opera. We'll also release a Javascript free as well as a frame free version very soon.

If you have any question, I'll be happy to give more details.

grnidone

2:21 pm on Sep 22, 2000 (gmt 0)



I thought it was a little odd that they highlighted what I would call promotional ad copy:

"world's largest manufacturers of mobile phones "

was all highlighted when I searched on Nokia. I can understand the "mobile phones" part of it, but not the "world's largest manufacturers" part.

-G

Brett_Tabke

3:45 pm on Sep 25, 2000 (gmt 0)

WebmasterWorld Administrator 10+ Year Member Top Contributors Of The Month



It probably matched several times within various pulls from Alta and/or whatever other engines they are searching.

The theory is pretty much like Topic at the U of Toronto or the Web Page Reputation Calculator at a beloved site near you.
One possible way it could work:
You enter a search word:
They do the search on Alta under that word.
Alta returns 100-1000 or so results.
They "index" those results by looking at the keyword density of the the results and throwing out the fluff (stop) words.
They then turn around and search on those top words and sort the whole mess into something presentable for the user.

What is really new here is the presentation method. Although I am no fan of proprietary javascript, theirs is an excellent usage of a fresh technique.

engine

4:00 pm on Sep 25, 2000 (gmt 0)

WebmasterWorld Administrator 10+ Year Member Top Contributors Of The Month



This is pretty quick and useful.
Found out some new info i must have missed using the "standard" methods of searching.

Thanks Nottingham.

2_much

6:55 pm on Sep 25, 2000 (gmt 0)

WebmasterWorld Senior Member 10+ Year Member



How are the results ranked? I performed a few searches and the rankings were a bit weird...the sites that ALWAYS appear at the top were at the number 6 or 7 positions. Very unusual.

vivisimo

7:05 pm on Sep 25, 2000 (gmt 0)



Vivisimo is reranking the snippets based on their content... that's an option we are experimenting with.. but the default behavior will certainly be not to rerank the documents at the first level, as this seems to puzzle a lot of people...

NFFC

7:20 pm on Sep 25, 2000 (gmt 0)

WebmasterWorld Senior Member 10+ Year Member



Hi vivisimo, thanks for dropping by.

This is my favourite search, soap [vivisimo.com].
BTW, don't forget us little people when you are rich and famous. ;)

vivisimo, check your local email, link at the top of the page.

NFFC

8:45 pm on Nov 27, 2000 (gmt 0)

WebmasterWorld Senior Member 10+ Year Member



Our old friend vivisimo [vivisimo.com] has got a mention at ResearchBuzz [researchbuzz.com] today.

NFFC

10:20 pm on Jun 15, 2001 (gmt 0)

WebmasterWorld Senior Member 10+ Year Member



"Yahoo! Internet Life Names Vivisimo Search Engine Developed at Carnegie Mellon The `Best New Search Service on the Web'"

Press Release [biz.yahoo.com]