Project seeks to nurture 'the wisdom of [online] crowds'

If you belong to a church, a model railroad club or a volunteer fire brigade, chances are the leaders try to guide your participation. "You're a Harry Potter fan? You should meet Sally." "I didn't know you played guitar. How about doing something with the choir?"

Today more and more social groups are on the Web, often involving thousands or even millions of people. Can computers take the role of human leaders to encourage members to interact positively and productively, especially in groups that produce useful products, like encyclopedia articles or music recommendations?

That's the goal of a new Cornell interdisciplinary project, funded by $3 million from the National Science Foundation Human Centered Computing program and conducted by a large team headed by Jon Kleinberg, the Tisch University Professor of Computer Science.

Making online groups function more effectively is a challenging problem. "The wisdom of crowds can turn into the idiocy of crowds or even the evil of crowds," explained Michael Macy, Goldwin Smith Professor of Sociology, a co-principal investigator.

Online communities often suffer from polarization, deception and outright fraud. The researchers hope to identify such problems through computer processing of natural language. There are, for example, recognizable language patterns in online lying.

They also plan to develop systems that can distinguish acceptance from criticism and agreement from disagreement, and tell the difference between parroting, paraphrasing and accidental mutation of content. This will require advancing the ability of computers to identify sentiment and recognize paraphrasing, said the researchers.

The next step will be to develop systems that encourage positive behavior, such as automatically channeling needed tasks to people best equipped and most likely to take them on. When people are given tasks suited to their abilities they are more likely to feel positive about the group and continue to participate, the researchers noted.

"Intelligent matching" will seek to create one-on-one conversations between members of the group. "People form stronger attachments to groups through social bonds with individual members and learn to become valuable members of a community through observing and working with experienced members," the researchers explained.

The team also will expand on the idea of "games with a purpose," like the ESP game co-developed by co-principal investigator and computer scientists Luis von Ahn at Carnegie Mellon University, in which players try to find the best labels for images, in the process tagging the actual images to make them easier for search engines to find.

The project will use the arXiv database of physical science papers and the Legal Information Institute Web site, both based at Cornell, and gwap.com, which hosts games with a purpose, as test beds. Team members have previously collaborated extensively with Google, Yahoo, Wikipedia and other major online institutions, and expect to try out some of their ideas on those platforms.

Geri Gay, communication, Dan Huttenlocher, computing, information science and business, and Lillian Lee, computer science, also are co-principal investigators. Additional senior personnel on the team are Jeff Hancock, communication, and Dan Cosley and Claire Cardie, computer science.

Cornell is leading in making the greatest paradigm shift in the social sciences since the 1950s: Its social scientists are collaborating in new ways with computer and information scientists to access the digital traces of online social interaction, says Michael Macy, Goldwin Smith Professor of Sociology.

It's not just about using computer tools to do social science, but about having access to vast amounts of data and opportunities to study the new kinds of social behavior emerging in the online world, Macy said.

"We have data on social interactions that I never thought I would see in my lifetime," he said. In addition to mapping connections between individuals and groups, he explained, we can now see the content of their interactions. "It is relatively easy to study friends, much more difficult to observe friendships," he said. The rapid growth in online interaction is changing that.

Meanwhile, the Internet has enabled such cross-cultural research as comparing American and Japanese book reviewers, which reveals cultural differences in an individual's willingness to go against the crowd.

"Although computer and information scientists were not trained in the social sciences, scholars like Jon Kleinberg and Dan Huttenlocher are also brilliant social scientists," Macy said. "I actually think some of the best social science at Cornell is being done by people with highly advanced skills in the analysis of large and complex networks and a fresh perspective on what these tools can tell us about some of the fundamental questions about social life. I'm cautiously optimistic that these cross-disciplinary collaborations are going to produce some major breakthroughs."

 

Media Contact

Claudia Wheatley