| 11. IRCS Improves Sorting of Search Engine Results
Those who frequent Internet search engines are often frustrated by the
results produced by the search terms they give. Some know techniques to limit
extraneous and unrelated entries, but even those cannot totally remove entries
that may contain an important reference in one paragraph. The United States Government
was concerned by this cause of inefficiency as well and ran a competition in 1995
to create a solution to one aspect of the problem: a program that would find all
paragraphs in a text that referred to a particular search term or other words
such as pronouns that referred to the search term. Penn graduate students working
at the Institute for Research in Cognitive
Science (IRCS) created a summarizer tool that would produce the paragraph
or paragraphs that contained the search term. While their work did not win the
competition, they made one modification in their code and produced better results
than the winning entry.
After the competition,
the team refined the program and adapted it so that it could search Internet sites
and gave an apparently impressive demonstration before the U.S. Congress. Unfortunately,
their system took about twelve minutes to use. Using a grant from the U.S. Department
of Defense, the IRCS researchers did further research on reducing the operating
time of their tool. Recently, this research
has been used to manufacture new systems for data mining that now are sold by
Alias I Inc. of Philadelphia, founded by IRCS researcher Breck Baldwin, continuing
the University of Pennsylvania's contributions to Philadelphia's economy through
computer research. |