TSUNAMI: Topical Social Understanding via Natural-Language Analytics For Megacity Interventions

Conducting operations in a Megacity will require new tactics involving cooperation with the local population. This necessitates gaining an understanding of the socio-cultural features of an area. The proposed TSUNAMI system will provide commanders with geographically referenced socio-cultural feature layers for publication on a Common Operational Picture display. These features will be derived from existing Civil Affairs data sources fused with open source data such as news feeds, Twitter, and Wikipedia. To achieve this, TSUNAMI will partition the city into a set of zones based on the existing spatial data. Textual data from the sources described earlier will then be processed by a novel, deep-learning based, Topic Modeling system. Each of the zones will be evaluated against the resulting topics. Each topics layer will be evaluated to determine how spatially relevant the topic is, and topics with high spatial relevance will be published to the Common Operational Picture. A spatial predictive model will also be run on these published layers to improve predictive accuracy and demonstrate the importance of these features.