2006 Horizon Awards Winner: University of Maryland's Oasys — Opinion Thermometer

Oasys, the Opinion Analysis System, owes its origins to a conversation two years ago between V.S. Subrahmanian of the University of Maryland and analysts at the U.S. Department of Defense. The analysts were looking for a way to measure and track worldwide opinion on matters related to national security.

"After the Abu Ghraib scandal broke, there was interest in how devastating the reports were in different countries: Is it worse in Saudi Arabia than in Bahrain? How does it change over time? and so on," Subrahmanian recalls.

He says he realized that little research had been done in that area, and he saw both commercial and defense uses for a tool that could dynamically measure the tide of global opinion on some topic specified by the user.

So Subrahmanian, a computer science professor and director of the Institute for Advanced Computer Studies at the university, teamed with like-minded colleagues at the University of Naples in Italy to develop algorithms and build a prototype that could find and measure the strength of opinions in news feeds. He said the technology, which has cost about $200,000 so far, will soon be handed over to the university's technology licensing office for commercialization.

Oasys includes a background crawler that in the prototype version watches 18 news sites in four countries and three languages. It monitors news feeds and extracts information about topics on which opinions are expressed in the news story.

Oasys scans the feeds for words in its dictionary of adjectives, which are coded as positive or negative. The adjectives are also weighted so that, for example, "fabulous" is seen as more positive than "good." Adverbs were recently added, and they're also weighted so that "bad" can be distinguished from, say, "very bad."

A user can enter a topic of interest, such as "Abu Ghraib," and see color-coded graphs illustrating the intensity of opinion over some specified time period, by language or country.

A key difficulty has been coming up with a reliable way to score the various words in opinions, since opinions are expressed in such idiosyncratic ways, Subrahmanian says. The Oasys development team has calibrated the algorithms against human panels where the goal is for Oasys to score opinions at about the median of what human subjects do.

Subrahmanian predicts that companies will want to use the technology to track the opinions of customers and critics on Web sites and blogs. That could help them more effectively target advertising, he says.

"More and more, companies realize that a lot of the information they need to fully understand the dynamics of decision-making resides in unstructured data," says John Hagerty, an analyst at AMR Research Inc. in Boston. "Because of the vast array of data outside the firm, technologies to assess attitudes, resistance, willingness to buy and so on will further enrich the decision support process, giving policy-makers and marketers the tools to help shape debate and demand."

Oasys produces a graph that shows a timeline and the intensity of opinion about a particular topic (in this case, Pakistani leader Pervez "Musharraf").

See the complete Computerworld Horizon Awards special report.