On Homeland Security and the Semantic Web: A Provenance and Trust Aware Inference Framework

The advent of the Semantic Web enables distributed publishing mechanisms on the Web as well as a large scale distributed knowledge store for machine agents with various purpose. Currently, significant amount of semantic web data (over 47,000,000 triples from over 130,000 web sites1) is available directly in RDF documents using semantic web vocabularies like Friend Of A Friend (FOAF) 2 and RDF Site Summary(RSS)3, and indirectly through information extraction tools such as Semagix Freedom(B.Hammond, Sheth, & Kochut 2002) and IBM’s tools(Dill et al. 2003). A promising application domain for the Semantic Web is homeland security, in which suspicious activities or associations among individuals and events must be recognized in millions of everyday reports from thousands of sources with varying trustworthiness, relevancy and consistency. When those reports are published in semantic web langauges, human analysts can improve their decision quality and efficiency by using automated tools. Figure 1 shows a simple scenario in homeland security which discovers semantic associations (Sheth et al. 2004) among “Mr. X”, “Terrorist Group” and “Osama Bin Laden” on the merged RDF graph from several sources. The atomic information unit is statement, e.g. (Mr. X, isPresidentOf, Company A). Statements can be grouped by provenance, e.g. (Company A, locatesIn, US) has provenance NASDAQ. This paper focuses on one type of semantic association which is a simple path linking two RDF nodes in an given RDF graph, e.g. Mr. X → Company A→ Organization B, → Mr. Y → Osama Bin Laden. Automating semantic association discovery and evaluation requires (i) augmenting the Semantic Web by extract more information from free text reports or databases; and (ii) discovering and evaluating semantic associations in large scale RDF graph. These are also the research objectives