Menu
Home
Log in / Register
 
Home arrow Computer Science arrow Social Computing, Behavioral-Cultural Modeling and Prediction
< Prev   CONTENTS   Next >

5 Example Interface Using the Results

While the results from the automated system were not as consistent as the observerbased results, they are far above chance (66% vs. 33%) which suggests this mechanism could be used to identify ETLs in ad hoc teams. To demonstrate how such information could be used, we combined this analysis with a six-topic analysis [3] of the features extracted that were not a DA or stop word. Each message was assigned a proportion of each topic present, the amount of ETL spoken, and these two values were multiplied and accumulated per participant, so the team member indicating the most leadership on each topic could be identified.

A prototype was created which read the transcripts as if they were running live. In Figure 2, the Fuel expert is logged into the chat system. His current expertise is displayed to the right of his name (Troops-Cargo, Armor-Security). The gold star next to Security's name indicates that this person is the current overall (non-topic weighted) leader of the conversation to that point. In the Chat tab, new messages are displayed on the top along with their sender and the topics of the message. In order to draw attention to important information, messages are highlighted in blue which contain a topic for which Fuel is currently considered an expert. Note that as chat messages stream through the viewer, the expert may change as a participant shows more or less expertise on a particular topic relative to the other team members.

Fig. 2. Prototype interface highlights messages on topics for which the user is the ETL

6 Conclusion

The examination of the study's transcripts suggests that ETLs can be detected with domain-independent phrases representing acts of Reasoning and Uncertainty. Combined with a statistical, domain-dependent topic model, information and communications could be routed to or highlighted for those who can use it most—improving situational awareness and decision-making, and routed away from or de-emphasized for those who can use it least—reducing information overload.

Due to its domain-independence and simplicity, and the ability to get real-time access to chat data in a number of military command and control settings, this approach could be used with current technology for real-time support in actual work settings. For example, following a hurricane, the algorithm could identify a physician volunteer who is an ETL for providing medical services. Others looking for a point of contact to provide medical expertise could thus quickly find such a person efficiently by avoiding a search through a word-of-mouth network. In practice, this might be done only when large differences in p(ETL), and thus a clear leader, among the team members is present. We are currently applying this analysis to new datasets of ad hoc teams, both academic [21] and real-world [24] in order to test the generalizability to new domains and larger team sizes.

Acknowledgments. We thank Jeff Morrison, Ranjeev Mittu, Fernando Bernal, Robert Stephens, Marcela Borge, Sean Goggins, and Carolyn Rosé for insightful discussions about this research. Gabriel Ganberg and Michael Therrien helped develop the software. Funding was provided by the Office of Naval Research's Command Decision Making program (N00014-11-1-0222). The views are those of the authors and do not necessarily represent the views of the Navy.

 
Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >
 
Subjects
Accounting
Business & Finance
Communication
Computer Science
Economics
Education
Engineering
Environment
Geography
Health
History
Language & Literature
Law
Management
Marketing
Philosophy
Political science
Psychology
Religion
Sociology
Travel