Uncertainty estimation in the fusion of text-based information for situation awareness

Rapid correlation and analysis of new and existing information is essential to recognizing developing threats and can easily be considered one of the most important challenges for military and security organizations. Automatic preprocessing (fusion) of information can help to "connect the dots" but arguably as important to the intelligence process is the assessment of the combined reliability of the information derived from diverse sources: how can one estimate the overall reliability of a collection of information which may be only partially correct, and is incomplete, vague or imprecise, or even intentionally misleading? In this paper we present a simple heuristics-based model for fusing text-based information from multiple diverse sources which exploits standardized representation of underlying source information. The strength of the model lies in the analysis and evaluation of the uncertainty at three levels: data (source and content), fusion (correlation and evidence), and model (accuracy of representation).