Collaborative Recommender System for Spatio-Temporal Intelligence Documents

US military and intelligence agencies have invested significant resources in data collection and effective search and analytics tools. However, due to increasing amounts of data, finding relevant information has become more difficult. Thus, there is an important need for recommender system technology that pushes relevant un-queried data to analysts through automation and machine learning techniques. Numerica proposes a novel recommender system for spatio-temporal intelligence documents that (1) implements a multi-level graph-based recommender to accommodate different data models and algorithms, while supporting real-time updates and computations at scale, (2) leverages deep learning for probabilistic linking of documents based on content to enable relevant document recommendations, (3) identifies users performing similar tasks to enhance collaboration, and (4) exploits user feedback for persistent improvements of recommendations over time. We believe this type of system is of general interest to companies that are developing information systems and intelligence agencies leveraging such systems. This approach blends cutting edge algorithms for graph processing, information flow analysis, and machine learning, developed by Numerica through years of government funded research, and leverages data and customer relations gained through Numerica's development of the state-of-the-art Lumen search and analytics system for law enforcement.