Reclassification Agent for Information Dissemination Security (RAIDS)

We propose to design and demonstrate the feasibility of a Reclassification Agent for Information Dissemination Security (RAIDS) to assist the human review process for multi-level security classifications. RAIDS will suggest an appropriate security classification level (e.g., secret, unclassified) for a derived document obtained via extraction from input documents. Our approach to building RAIDS is to develop a Naïve Bayesian Classifier (NBC) based on training samples of unstructured document portions marked with their classification categories (e.g., military plans, foreign relations, scientific, technical) and associated security levels, where portions in documents include paragraphs, charts, tables, and pictures within the document. The RAIDS agent marks input documents using the trained classifier and then recommends an overall classification level for each derived document along with its confidence on the classification level. We plan to implement a prototype using common services for databases, visualization, and information retrieval available under the existing Service Oriented Architecture (SOA) frameworks. The major benefits of the proposed human-in-the-loop approach are faster document reclassification and avoiding those errors easily overlooked by human eyes. The generic nature of RAIDS provides potential for its transition to existing cross-domain solutions in the field today.