We’re a small business specializing in:
-> near-real time data extraction
-> behavioral analysis
-> threat forecasting
- Conflict and cooperation outcomes are a function of the strategic decisions that governments, dissidents, and the larger population make within constrained environments.
- Ultimate victory resides with who gains the support of the larger population/society.
- Government & dissident behavior is a function of their desire to win support and gain/remain in power (This theory transports well to businesses and competitors trying to win support of consumers)
- They use a variety of tactics to do so (actions & rhetoric)
Data Needed to Specify/Test Theory
- Events Data – “who does what to whom”
- Sentiment/Emotions Data – “who says what to whom or about what ” and “feels what about whom/what”
- Discourse Data – “how dissidents communicate ”
Automated Data Extraction Tools
- Praxis: Automated Events Data Engine
- Pathos: Automated Sentiment & Emotions Data Engine
- Logos: Automated Rhetorical Discourse Data Engine
- Taxis: Automated Document and Topic Classifier
- HuGo: Heuristic Geo-location tool
Models Needed to Specify/Test Theory
Models deliberately simply reality to highlight key causal relationships, filter out extraneous/secondary effects, and focus on how motivations and actions are interrelated.
Statistical Models estimate parameters (effects) for the impact that:
- strategic interactions among key players in a situation
- emotions and support for those actions/actors
- ) the rhetoric and intent of actors have on outcomes (4) within a constrained environment.
- Once we understand the relationships, we can use those parameters to forecast future outcomes.
- There is no silver bullet model; thus we create multiple models that highlight and weight accordingly different aspects of a situation to produce the best ensemble forecast.
- estimate parameters of a model that describe the stochastic process underlying a temporal series of data
Hierarchical Mixed Effects Models
- Allows the estimation of different effects for variables across spatial units (cities, regions, countries, continents, etc.)
Random Forest Models
- Ensemble learning method for classification and regression which constructs a multitude of decision trees to optimize the fit of a variety of variables
- Spatial autoregressive models
- Distance decay models
Ensemble Bayesian Model Aggregator (eBMA)
- Uses outputs from multiple models to better forecast behavior (e.g., similar to hurricane track forecasting)