Comparison of data development tools for populating cognitive models in social simulation

Abstract

The United States is engaged in a new type of warfare. Defeating the enemy is now predicated on winning over local populations. To win these groups, commanders need to know what responses to expect for various operations in particular locations. Social simulations are a promising means of modeling these reactions, and there are several current methods used to populate these simulations with agents representative of a specific society. These methods, however, often require the input of subject matter experts and are costly in price and time. This thesis examines the simplification and automation of the agent instantiation process by conducting a usability study of two data development tools currently under consideration by the U.S. Army and TRAC-MTRY. The tools, a survey data case file generator developed at TRAC-MTRY and a text analysis tool (STANLEY) developed by Sandia National Laboratory, were examined in separate manners, and the results were encouraging. The survey tool was tested to validate in a practical manner its generated case files with respect to simulation output and real-world surveys. STANLEY was evaluated by scoring sentiment in a document corpus and attempting to correlate those scores to a real world issue. Results of the study indicate that the survey data tool generated case files of adequate quality to instantiate social simulations, potentially minimizing SME requirements and costs. Technical limitations precluded STANLEY from returning enough data for sufficient correlation comparison, although the results indicate the tool has potential.