UNC's homepage
The review might be useful to anyone who wants to use a Web search engine to find papers or other works "similar" to a particular paper or work. For example, you could take an abstract or outline for something you wrote, and see if anything similar is "out there."
The audience for Excite is general: it's intended to be useful to anyone searching for anything on the Internet. However, an article I saw in the June 1996 Wired magazine indicated that Excite actually uses some advanced experimental retrieval methods that I was familiar with -- actually, similar methods to what I wrote the paper about!
Excite has a "power user" page, which lets you specify which terms
are required ("MUST OCCUR") and which are desired
("MAY OCCUR"). Essentially, this works like Boolean logic
for the MUST terms. So, I first tried entering just my
title from the paper, with "information space" as MUST and "building"
as MAY:
building information space
The results were . All I got was stuff about people making new buildings where the space was very nice and lots of information would be stored. In other words, press releases and other publicity information or announcements/descriptions.
So, instead, I tried a "query-by-example" model. By this I mean I
tried to give enough text so that "similar" documents would come up
in Excite. I copied the whole abstract from my article into the
MAY box:
A set of procedures for the creation of metric multidimensional
information space for information retrieval (IR) are presented. The
rationale for spatial approaches to the representation of information
is discussed, and a comparison is made between various representation
methods for IR. The uses of information space for different types of
IR systems are introduced, along with means for assessing information
space.
The first few documents in the list (I had Excite retrieve the top 50 with brief summaries) were disappointing. But further down were two documents that were very good matches. One was a link to an abstract of a Swiss fellow's paper dealing with a very similar topic to mine [1]. I followed this link and discovered I now knew of someone working in this area who I didn't know about before.
I used this for "relevance feedback" with the option to Click here to perform a search for documents like this one, but the results were disappointing: they called up a whole bunch of other stuff by the same author or his research group, and nothing else. It made me wonder if the domain name were considered in doing "similar" searches.
The second useful link was about multidimensional analysis for engineers [2]. This was interesting, because although I mentioned "multidimensional" in my query, I didn't use any other terms I thought were really related to analysis or mathematical techniques. When I tried this for feedback, all I got was more technical mathematical documents, which really didn't help.
In the end, I found a link to one researcher and his group that was relevant and interesting. Also, a number of links to tangentially related mathematical treatment of a topic related to my paper. This was not great, but wasn't terrible. The main thing that this search demonstrated was that the "query-by-example" concept for Excite that I read about in Wired doesn't really work that well; or at least it didn't for entering my whole abstract. Perhaps using an entire document (not just an abstract) would work better.