Template Based Semantic Similarity for Security Applications

Today’s search technology delivers impressive results in finding relevant documents for given keywords. However many applications in various fields including genetics, pharmacy, social networks etc. as well as national security need more than what traditional search can provide. Users need to query a very large knowledge base (KB) using semantic similarity, to discover its relevant subsets. One approach is to use templates that support semantic similarity-based discovery of suspicious activities, that can be exploited to support applications such as money laundering, insider threat and terrorist activities. Such discovery that relies on a semantic similarity notion will tolerate syntactic differences between templates and KB using ontologies. In this paper, we describe our approach on querying large KBs using template-based similarity performed as part of the SemDIS (Semantic Discovery) project. We also described the associated mechanism for ranking results that are complex relationships between objects (rather than documents in a typical Web search). This approach is prototyped in a system named TRAKS (Terrorism Related Assessment using Knowledge Similarity) and explained using scenarios involving potential money laundering.