types of information about a table.

Example:

At Mitre, we collect information from the internet using commercial

search engines such as Altavista and Lycos for a variety of topics.

This information is stored as a collection of text documents for any

topic. The collection can be searched via keywords. We use a public

domain search engine Glimpse from the University of Arizona to index

and search the document collection.

Some of the document collections can be fairly large (over 1500

documents). If a common keyword is used, then the list of matching

documents will be large. We decided to display the results from the

search engine using Java and JDBC to avoid scanning long lists of

matching documents. Java was used to build a 3D space and plot circles

at locations in the 3D space to represent the frequency of the

occurrences of keywords in a document. JDBC was used to retrieve the

titles of the documents which were stored in a table. Passing all the

titles of all documents in the collection as parameters to the Java

applet would increase the time to load the applet significantly.

Glimpse returns the frequency of occurrence of a keyword in a

document. We use that number to locate a circle representing the

document in 3D space (Fig.3). Each axis represents a keyword. If fewer

than 3 keywords are used, then documents will be displayed in a plane

or on a line. If more than 3 keywords are used, then 3 or fewer

keywords must be chosen to display matching documents. The frequency

of occurrence of keywords was normalized for each axis. The

frequencies of keywords in documents was passed as parameters to the

applet. The color was the circle was computed based on the position of

the circle in the 3 axes. Red was used for documents on the z-axis,

green was used for documents on the y-axis, and blue was used for

documents in the x-axis. Brighter shades of the 3 primary colors were

used for documents with higher keyword frequencies. A mix of the

primary colors was used for circles which contained more than one

keyword.

Fig.3

JDBC was used to retrieve the titles for documents which contained

non-zero occurrences of the keywords. This number was usually fewer

than the total number of documents when a fairly unique keyword was