When a group of people aims to develop something new – a product, a marketing strategy, a disaster-response plan, an advertising campaign, a corporate logo, an informational website – they will need to draw on data, to share information and ideas, and to discuss scenarios in order to fuel their creativity and innovation.

A small group at a single location working with a small amount of data might be able to handle their task with just pencils, sketch pads, a white board, and their own know-how, but most situations are more complex. Many people at multiple geographically dispersed sites might need to be involved in a collaboration, and the team could have mountains of data to analyze.

Researchers at the National Center for Supercomputing Applications (NCSA) and the Illinois Genetic Algorithms Laboratory (IlliGAL), both at the University of Illinois at Urbana-Champaign, have developed a tool, called DISCUS (Distributed Innovation and Scalable Collaboration in Uncertain Settings), to facilitate creativity, innovation, and collaborative work in such complex situations.

The development of DISCUS has been funded by the Technology Research, Education and Commercialization Center (TRECC), a University program that is funded by the Office of Naval Research and administered by NCSA. The project also has received support from Shalini Dewan, a TRECC technology commercialization and transfer specialist.

“I work with researchers to identify what they are interested in doing and on finding tools to efficiently get them to where they want to go,” Dewan explains. “TRECC can help by identifying funding opportunites, by introducing potential collaborators, by familiarizing scientists will the commercialization process, and by connecting them with the many federal, state and local agencies that are able to provide resources and support.”

In a recent experiment, the DISCUS prototype was successfully used to help a Japanese company gather and analyze information about consumers' cell phone preferences, providing insights that might help in the development of new products.

A Collection of Tools

The story of how the DISCUS project evolved is a little like the old commercials for a certain candy: Chocolate and peanut butter collide and a new taste treat is created.

Similarly, DISCUS brings together the genetic algorithm research of the Illinois Genetic Algorithms Laboratory, led by David Goldberg; the data mining and text mining work of NCSA's Automated Learning Group (ALG), under the direction of Michael Welge; the concept of chance discovery and the KeyGraph technique; and collaboration tools.

DISCUS project leader Xavier F. Llorà, a post-doctoral researcher in the Department of General Engineering, explains that the goal of the project has been to provide an easy-to-use framework that combines those underlying concepts and technologies. DISCUS users don't have to understand all of the subtleties of interactive genetic algorithms or text mining; they just need to familiarize themselves with the DISCUS interface that allows them to tap those powerful tools.   

The features of the DISCUS prototype include:

• Collaboration tools, including message boards, instant messaging capability, and chat rooms. Both parallel and sequential discussions are possible. Information can be shared among group discussions and discussions can be archived for later review and analysis.
• Data and text analysis via the D2K (Data to Knowledge) and ThemeWeaver tools developed at NCSA.
• User-centric genetic algorithm models for innovation support. The shared and archived “solution centers” are provided to users as potential solutions, which the users can evaluate, combine, and elaborate to form new solution options. DISCUS also facilitates web searching as a source of outside input, filters solutions, and matches problems to the most relevant system users.
• Chance discovery computational embodiments. Archived discussion logs can be analyzed, even while the discussion is still active. KeyGraphs of these logs provides a map of the discussion, serving to summarize and focus the discussion as it progresses and helping to identify relevant scenarios and the connections between them. DISCUS provides real-time KeyGraph analysis, allowing any participant to analyze the scenario, highlight key topics, and share this analysis with collaborators.

The DISCUS Advantage

DISCUS has been used in several real-world situations, allowing the developers to test their theories and further refine their prototype. In the fullest DISCUS experiment to date, the team and researchers from the Hakuhodo Institute of Life and Living (the second largest marketing and publicity firm in Japan) conducted a large-scale marketing survey at the University of Illinois campus.

TRECC also assisted during this phase of the project. “I was able to brainstorm with the Hakuhodo scientists and the DISCUS team,” Dewan says. “And I was able to suggest further contacts at the University who could help answer some of their questions and needs.”

Hakuhodo helps its clients discover emerging markets and develop new products. If a client is interested in identifying what new cell phone features will be popular with consumers, for example, Hakuhodo might mine existing data on cell phones, develop a scenario to evaluate (such as, Would consumers like cell phones with built-in cameras?), and test the scenario with focus groups, who provide their feedback on the scenario.  

A traditional focus group gathers participants and moderators in one location and takes several days. Hakuhodo often uses Day 1 to give the focus group information and a scenario, then uses Day 2 for guided discussion, and Day 3 to synthesize the suggestions from the group.

This time, Hakuhodo used DISCUS to streamline their collaboration, data mining, focus group brainstorming, and scenario testing.

Hakuhodo first provided the DISCUS team with a database of nearly one million records, representing information gathered from questionnaires completed by people in New York, Los Angeles, and Chicago. This data was mined using the D2K component of DISCUS in order to determine what characteristics the team should look for in focus group participants. This data mining was guided by a Hakuhodo's model of how ideas and new products spread in Japanese society (young women are typically the first to latch onto new products and to influence others to purchase them). Such diffusion models have been widely studied by Japanese researchers at Hakuhodo, providing a solid background for the group dynamics.  

Then the DISCUS team conducted a similar survey on the Urbana-Champaign campus of the University of Illinois. From the survey respondents they were able to select the individuals with the desired characteristics to participate in the focus groups. The goal was to have a combination of technology innovators, who are power users of gizmos; early adopters; and late adopters. Because college students tend to be heavy users of technology, they found more participants from the first two classifications.

“On this campus, it's difficult to find someone who never uses email,” Xavier laughs.

One the team had the desired mix, the participants were brought to IlliGAL, where they used DISCUS on desktop computers to answer questions and provide feedback. To start the online discussion, the moderator would post a question, such as “What do you like about your current cell phone?” The focus group participants would conduct an online discussion on that topic for about an hour, take a break, and then would return to repeat the procedure with a new question.

During the participants' break, the archived text of the first discussion could be analyzed using DISCUS. Repeated phrases and trends could be detected, correlations could be made, and the team could actually use the information from the first session to formulate a new question or scenario to put to the focus group after the break. When they returned to their computers, the participants might be asked “What features would you like to see on a cell phone in the future?” or asked to evaluate and rate specific scenarios.

Llorà points out that this flexibility also enables DISCUS users to employ the results on one focus group in creating scenarios for a second group. A focus group of innovators, for example, could precede a group of early adopters, which could in turn be followed by a group of late adopters.

“We can create a path of how information flows between groups,” he says.

Group moderators were also able to see when an online discussion was repeating the discussion of a previous group. When that happens, the DISCUS moderators could steer the discussion in a new direction, generating more original input.

Over three days, the team worked with 10 focus groups, spending about four hours on each focus group. That's a dramatically faster turnaround time than the traditional method, which takes several days for each focus group.

Speed isn't the only advantage DISCUS provides. Because all of the online discussions are archived, DISCUS users can easily return to the data for analysis, mining the communication for common and relevant features or interesting crossover of concepts. This information can then be used in a KeyGraph to pinpoint an emerging need or market. A traditional focus group could be recorded, but the data in video recordings can't currently be easily mined or analyzed.

DISCUS also reduces the number of moderators needed to conduct a focus group. A typical focus group that gathers all of the participants in a room for a discussion requires almost as many moderators as participants. In the DISCUS test, however, each group of five or six participants needed only one or two moderators. Using DISCUS, the same number of personnel could conduct more focus groups or larger focus groups, thereby generating more reliable alternatives.

Continuing Development

During the March experiment, the DISCUS developers asked users to assess the interface and the functionality of the tool. They're now using that feedback to make adjustments and enhancements. Continuing research is also aimed at refining the KeyGraph techniques and making sure that the configurations and utilities included in the DISCUS platform are optimal. Advances in interactive and human-based genetic algorithms research are laying the groundwork for innovation and creativity support tools to be incorporated in the next DISCUS release.

Perhaps the biggest current development push, however, is the effort to make DISCUS compatible with Asian character sets so its data mining, text mining, and KeyGraph utilities will function in Japanese. Hakuhodo conducts huge biennial consumer surveys in Japan, and the company is interested in using DISCUS for its 2006 survey.

The team is also looking at other situations in which DISCUS could be applied. The process of gathering requirements for software development – a process of brainstorming, discussion, and compromise — is a potential scenario being discussed with researchers at the Japanese division of Hewlett-Packard.