STRIDER: Semantic Targeting of Relevant Individuals, Dispositions, Events, and Relations
STRIDER is collaborative automation for intelligence analysis. STRIDER is designed to automate the gathering, integration, and reporting of new information to ultimately improve the breadth, depth, and efficiency of intelligence analysis.
STRIDER uses the Sparser natural language understanding system to extract information from open-source news articles. It organizes and integrates this information using hierarchical content models, and then displays the information in link diagram interfaces for intelligence analysts, allowing them to focus on their intelligence objectives in a format they understand. STRIDER uses heuristics and equivalence classes to fuse information about entities and events across articles, and retains the source of all of its infrences and conclusions according to U.S. Intelligence Community directives. In our pilot studies, STRIDER extracted 88% of the information that a senior intelligence analyst extracted from news articles with 100% precision, but over 250 times faster, retaining full provenance of all of its content. STRIDER's 12% incompleteness was due to known gaps in its ontology; other SIFT projects are aimed at semi-supervised learning to automatically close these ontological gaps.
STRIDER weighs the credibility of extracted information -- and the sources of information themselves -- by eliciting user approval/rejection of extracted information and using collaborative filtering and preference ranking algorithms. These algorithms will help STRIDER make confidence estimates of extracted information and automatically prioritize conflicting information for human users.
SIFT developed STRIDER with the consultation of a senior U.S. Intelligence Community analyst who served in CENTCOM, Joint Staff, Defense Intelligence Agency, Iraq Survey Group, Joint Task Force Six, and the Afghanistan-Pakistan Task Force. STRIDER has received support from federal and state-level intelligence agencies.