RiskWatch: Linking Environmental Stressors to Security Outcomes

Many environmental factors can potentially affect a communitys security. In this project, we propose to develop an open source system, called RiskWatch, for harvesting data from multiple sources and linking that data to security outcomes. Relevant data can potentially be found in a wide variety of sources, including structured, numeric data sets and images published by governments and NGOs, in natural language in reports and news articles, or even in informal text aggregated from social media. The difficulty of reliably forecasting security outcomes is amplified by the fact that there are so many types of relevant data, and so many ways to express that data. RiskWatch will include capabilities for extracting and interpreting data about environmental stressors published in a wide variety of sources. To forecast security outcomes, we will employ Probabilistic Soft Logic (PSL), a recently developed machine learning approach that has been employed successfully for many applications, including forecasting civil unrest from open source indicators. In Phase I, we will develop an end-to-end design of the RiskWatch architecture and show that our approach is feasible, using sample data sources.