DREAM: Detecting Relations, Entities, and Attributes Misinformation

ABSTRACT: Information fusion and knowledge conflict detection are required for many mission-critical intelligence analysis tasks. Using knowledge extracted from various sources, including entities, relations, and events, intelligence analysts identify relevant documents, integrate facts into summaries about current situation, and augment existing knowledge with inferred information. To deal with large amount of data, analysts require automated solutions to link events, entities and related knowledge across multiple sources. Aptima proposes to develop a system for Detecting Relation, Entity, and Attribute Misinformation (DREAM) to support processing of data with redundant, erroneous, and deceptive information. Our solution combines strong theoretical foundation in cross-document entity resolution with validated solutions for conflict and activity pattern learning and detection. When fully developed, DREAM will provide intelligence analysts with a powerful analysis tool that (1) finds feasible probabilistic association between entity and event mentions across different sources; (2) detect and remove conflicts in relational knowledge across different records; and (3) learn and detect normal, conflicting, and deceptive activity patterns among multiple entities reported in different data sources. BENEFIT: DREAM will allow intelligence analysts and commercial users to reduce the uncertainty in their knowledge bases and detect suspicious and anomalous patterns across multiple documents. For DoD applications, DREAM will support new PCPAD workflow by focusing the analysts on most critical data to reduce the analysis time and increase detection of hostile activities. For commercial applications, DREAM will enable faster and more accurate detection of deception in social media, online fraud, and social engineering activities.