STAR: Semantic Temporal Association Retrieval

Intelligence analysts gather and synthesize information from multiple sources in response to both high priority requests and standing topics. Their expertise enables them to evaluate and process information effectively, but they spend too much time searching for relevant information. As ops tempo increases, information overload will only increase. The STAR (Semantic Temporal Association Retrieval) system will support information triage and synthesis and enable analysts to allocate more time to evaluation and genuine analysis. STAR will systematically compare traditional Latent Semantic Analysis and two of its emerging alternatives – Probabilistic Latent Semantic Analysis and Latent Dirichlet Allocation – structured around meaningful analyst tasks. We will also combine topic and time-course models to create a tool that automatically compares recent and historical patterns, enhancing triage and increasing the predictive power of intelligence analysis. Both systematic comparison of statistical language processing methods and incorporation of an explicit temporal component are innovations.