Weak models for insider threat detection

This paper describes the design for a content-based approach to detecting insider misuse by an analyst producing reports in an environment supported by a document control system. The approach makes use of Hidden Markov Models to represent stages in the Evidence-Based Intelligence Analysis Process Model (EBIAPM). This approach is seen as a potential application for the Process Query System / Tracking and Fusion Engine (PQS/TRAFEN). Actions taken by the insider are viewed as processes that can be detected in PQS/TRAFEN. Text categorization of the content of analyst's queries, documents accessed, and work product are used to disambiguate multiple EBIAPM processes.