Extracting Location-stamped Events from Textual Data for Persistent Situational Awareness

ABSTRACT: Automated extraction of event location information enables intelligence analysts to rapidly visualize information contained in large volumes of unstructured textual data. Although natural language text analysis software already has the ability to extract locations to some degree, there still exist deficiencies that we will address in this project. We will expand geocoding to include facilities rather than only geocoding locations. Linking location mentions with event mentions or each other has not received an adequate treatment in existing literature. We will address this issue by implementing and benchmarking modules for these tasks. We will also study the prospect of automated extraction of implicit location-event extraction, namely determining the location of an event mentioned in an input sentence even if that location is not explicitly mentioned there. BENEFIT: The main anticipated benefit of this work involves advancements in methods to automatically extract and geocode locations corresponding to events as they occur in unstructured data. These are especially useful in enhancing event visualization from such data. Some of the methods to be studied involve extracting information that is only implicitly mentioned in the text, which is one step beyond what most systems can produce.