ABSTRACT
Lockheed Martin Advanced Technology Laboratories’ (LM ATL) experience on the Defense Advanced Research Projects Agency’s Integrated Crisis Early Warning System (ICEWS) program demonstrated that there is indeed exploitable information hidden in the massive amounts of news data available to today’s analysts. The ICEWS problem was to use this information to help provide monthly forecasts of national stability across five events of interest (EOIs) ranging from rebellion to international crisis across 53 countries (O’Brien, 2010). We hypothesized that no single model or modeling paradigm could efficiently handle and represent all aspects of this complex problem. A mixed-model, multi-hypothesis approach was developed to enable users to more fully exploit the available information and accurately forecast those stability measures. The success of this approach required meeting significant challenges in ingesting, coding and developing of the individual forecasting models to produce the forecasts. An additional challenge for user acceptance was to expose the results of this complex system to the user in a way that hid the complexity without hiding valuable information.
