Election
2000 Web site at UC Berkeley demonstrates power of new Internet
technology for mining the "deep Web"
25
Oct 2000
By
Robert Sanders, Media Relations
Berkeley
- Want to find out which Hollywood stars donated to Vice
President Al Gore's Presidential campaign? How about the
home prices of the donors to Texas Governor George Bush's
campaign? Or the crime rates in the neighborhoods of donors
to either candidate?
As this
year's Presidential campaign climaxes, a University of California,
Berkeley, professor has created a Web site that makes such
searches easy, and demonstrates the power of new Internet
technology he has developed to mine the "deep Web."
"This
is more powerful than search engines on the Web," said Joseph
Hellerstein, associate professor of computer sciences in
the College of Engineering at UC Berkeley, who created the
site with fellow computer sciences associate professor Michael
Franklin and the help of five graduate students and one
undergraduate. "With this you can do real data analysis,
not just find a neat new Web page."
The software
that Hellerstein and Franklin developed is called Telegraph,
after the street near the UC Berkeley campus famous for
its street vendors and street people.
"Like
the Berkeley main street after which it is named, Telegraph
is the natural thoroughfare for a volatile, eclectic mix
coming from all over the world," Hellerstein wrote on his
Web site.
The "deep
Web" refers to information on the Internet that is not available
by simply following hyperlinks, and thus not accessible
through search engines like Google or Inktomi. Some people
estimate the deep Web contains 500 times as much information
as the rest of the Web, most of it in free databases that
require a person to fill out a form in order to submit a
query.
The database
searches and cross-referencing that Hellerstein, Franklin
and their students make available can be done by anyone
willing to delve into publicly available databases compiled
and run by the Federal Election Commission, the APBNews.com
Crime Statistics site, the Yahoo Real Estate database, the
Yahoo Actor and Actress List, the U.S. Census, and others.
But such painstaking digging is laborious and time consuming.
Hellerstein's
Web site makes it easy by automating the form searches,
so that information in one database can be brought up for
comparison and correlation with information in other databases.
The computer does the tedious data searching - "screen scraping"
in computer jargon - while the new UC Berkeley technology
choreographs the search in the most efficient way.
"This
is about the facts and figures on the Web. Web crawlers
can't get to this information," he said. "We call this data
the 'Facts and Figures Federation.'"
For example,
journalists today can trace money spent by political action
committees (PACs) by laboriously pouring over lists of donors
and tracking the money back through numerous other PACs
to its corporate or private source. The UC Berkeley team
has set up a way to connect the money easily by automatically
"crawling" the donations back to the source.
"You
can track the six degrees of separation of PACs - which
PACs give to other PACs," Hellerstein said. Philip Morris,
for example, doesn't give directly to Bush, but through
its PAC gives to other PACs, like the Fund For a Responsible
Future, that in turn give to Bush. Similarly, the AFL-CIO
doesn't give directly to Gore, but does give to other PACs,
like the Evergreen Fund, that give to Gore.
Alternatively,
a comparison of crime rates in the neighborhoods of donors
to the two candidates show that Bush's donors live predominantly
in low crime areas, while Gore's donors are spread out among
low and medium crime areas.
And what
do Gwyneth Paltrow, Jack Nicholson, Candice Bergen and Jerry
Seinfeld have in common? They've each given to the Gore
campaign, along with a slew of other actors and actresses.
It's hard to identify any Hollywood donors to the Bush campaign.
"With
software like this, the publicly available databases are
more powerful than people thought," Hellerstein said.
And,
he added, potentially more scary. With all the databases
on the Web, it also is possible to correlate names with
addresses for the entire U.S. population, and cross-reference
those with individual home prices, neighborhood crime statistics
and even AIDS infection rates. Marketing firms with their
own private data could mine veritable gold by combining
their data with these deep Web databases.
"This
software enables new things, and those things have consequences,"
he said. "Obviously it's good to have some of these databases
publicly available, but they can lead to serious invasions
of privacy. This technology may cause people to rethink
the balance between freedom of information and privacy."
Telegraph
merges two technologies that make the Internet work: the
technology that makes data flow smoothly through the Web,
and database query technology. He calls Telegraph an adaptive
data flow system because it adapts to the fact that data
doesn't flow at the same rate from all sources. One database
may be slower than another, or the Internet may slow down
and then speed up. Correlating data in real time requires
a system that can adapt to such unpredictable behavior.
Using
the metaphor of flowing water, he described the technique
as creating an eddy of data analysis within a river of information
streaming across the Internet. He employs a kind of lottery
to determine which database is queried at each step, and
is able to optimize and speed up the process of data collection.
Telegraph also is designed to harness streams of live data
coming from networks of sensors on the Internet, or even
from smart devices.
The adaptive
dataflow technique is well suited to searching databases
that are updated frequently, such as lists of campaign donors,
and also databases that are impractical to download in their
entirely and cache for later analysis. Telegraph downloads
the information "live" from the source. The software also
makes it easy to handle new databases that appear on the
Internet.
"This
will be useful for anyone who wants to look at trends or
the big picture using lots of data. That means marketers,
pollsters, businesses and researchers of all kinds," Hellerstein
said.
For the
moment, Hellerstein and his students have scripted specific
database searches and made them available on their Web site,
such as a correlation of crime rate with a campaign donor's
Zip code. More general comparisons could be allowed, but
a user would have to learn how to write a proper query.
Many users would not go to that trouble, Hellerstein said,
so making that process easier is an area of future research.
But he and his students can script database queries quickly,
and they plan to provide further examples on his Web site.
"Once
the election is over, we'll draw on new sources of information
on the deep Web and elsewhere," he said. "It will be interesting."
###
The Federated
Facts and Figures Website, complete with access to Election
2000 information, is at http://fff.cs.berkeley.edu/.
For more
information on the adaptive dataflow system, Telegraph,
check out http://telegraph.cs.berkeley.edu/.
.
|