Subject: Re: Agents: too often misunderstood From: dowjone!rexb (Rex Ballard) Date: Sat, 26 Feb 94 00:18:23 EST

Cc: online-news@marketplace.com
Sender: jvncnet!marketplace.com!owner-online-news
Content-Length: 1215
X-Lines: 23
Status: RO

There is a compromise between the pure "search agent" and general "browsing".
At Dow Jones, we review all of the stories going out over our wires.  An
automated system tells the operator what he thinks the story is about.
These are generated as Dow Jones' Catagory Codes.  A human operator quickly
scans the article to determine when an article references Lotus, publisher
of 1-2-3 or the Lotus eater boutique.

Internet News Groups are a good "first line" sort for this type
of traffic.  In this scenario, cross-postings can actually be a benefit.

Some of the search retrieval techniques used by modern text retrieval engines
provide a combination of browse and search.  Once a search as narrowed
to about 100 articles, the user can view the headlines, or the headlines
and the first page of each story.

The CARL system used in many public libraries is another example.  It is not
uncommon for the first term to produce 3000 to 4000 hits.  Adding terms
can gradually reduce the number of stories to the top 30 or 40.  A well
trained librarian can narrow it down to 5 or 10.

Other tactics such as relevance ranking, searches restricted to leading
paragraphs, and negated terms can also be effective strategies.


From jvncnet!marketplace.com!owner-online-news Mon Feb 28 10:20:42 1994