Re: Search accuracy

Ted Sullivan (tsullivan@blizzard.snowymtn.com)
Fri, 5 Apr 1996 09:29:00 -0800

Yes, I am. My interest is related to data mining Engineering Design /
Construct / Maintain information related to plant development. Mostly
focused on the Pulp and Paper & Mining Industries. The goal is to present
the user a Web Interface that allows the user to query the whole dataset (
2d, 3d, commercial data ( purchase orders, engineering specs etc), project
schedule, documentation (memo's, faxes, contracts and calculation sheets),
and ISO 9000 procedures ... ) and get back links to the data regardless of
what Engineering application created it or uses it.

Of course this is more then a Web server or a search engine, it is trying to
locate important information in the off hours and store up references ( in a
OODBMS) so that during the working hours my customers can find what they are
looking for.

The system uses a inference engine to try and locate the data and tie it
together. The main idea is that once you find a useful piece of information
the page has links to all other related piece and you can navigate around in
a dynamic document.

Bold, yes; can I actually do it, good question; am I trying, sure
am.

Ted
----------
From: robots
To: robots
Subject: Re: Search accuracy
Date: Thursday, April 04, 1996 11:26AM

>A true semantic network ... does not
>force the user to make the precision/recall tradeoff. Yes, the semantic
>network does boost recall by building in literally millions of word links
(so,
>stock is linked to equity, share, trade, bond, security, etc.). However,
>unlike a thesaurus, or any other tool used in search engines today, the
>semantic network also lets you specify word meaning. Thus, you can specify
a
>search on stock as "shares issued by a company...," telling the system to
>ignore references to soup stock, live stock, retail stock, etc.

If there exists a search technology that is so accurate that it never finds
irrelevant documents and always finds all of the relevant ones, we'd like
to buy it. Any time you ask a more accurate question of a good search
engine, you'll get more accurate results, regardless of whether you're
using a "true" semantic network, knowledgebase, Cliff Notes or anything
else that helps define the concept on which you searching.

In any event, this isn't the place to flog our products, features, etc.

One of the things that I'd find really interesting would be research into
the construction of semantic networks or other knowledgebases from Web
topology. That would be a fascinating byproduct of a spider's
explorations. I only know of one experiment along those lines, being done
by one of our customers. Anyone else looking at this?

Nick