Shallow semantic analysis to estimate HUMINT correlation

This paper proposes a new solution to estimate the correlation of HUMINT data. From a more general point of view, the underlying problem tackled concerns the evaluation of information, as, according to military doctrine, the more correlated data are, the more credible, therefore valuable, they should be considered. The solution proposed is completely automatic, and it is based on shallow semantic analysis. HUMINT data are not enriched by semantic annotations, as it is the case in many approaches treating such data, but they are processed by a chain of treatments allowing the identification of context specific features and ontological entities. Therefore, a measure is defined to express the correlation degree of HUMINT data by keeping the distinction between ontological entities and their evolution context. A military intelligence ontology supports this solution. It models domain entities, whose properties are exploited by shallow analysis procedures. The ontology also allows us to cope with linguistic variability, an inherent problem of HUMINT information processing. Going beyond keywords spotting and analysis, the proposed solution provides a more accurate estimation of HUMINT correlation and its output can be exploited by different applicative scenarios. Therefore the paper briefly addresses the use of shallow semantic analysis to assess uncertainty of HUMINT data.