A Dempster-Shafer theoretic conditional approach to evidence updating for fusion of hard and soft data
Kamal Premaratne, Manohar N. Murthi, Jinsong Zhang, Matthias J. Scheutz, Peter H. Bauer · 2009
Abstract – Fusion of hard data with soft data is an issue that has attracted recent attention. An effective fusion strat-egy requires an analytical framework that can capture the uncertainty inherent in hard and soft data. For instance, computational linguistic parsing of text-based data gener-ates logical propositions that inherently possess significant semantic ambiguity. An effective fusion framework must ex-ploit the respective advantages of hard and soft data while mitigating their particular weaknesses. In this paper, we describe a Dempster-Shafer theoretic approach to hard and soft data fusion that relies upon the novel conditional ap-proach to updating. The conditional approach engenders a more flexible method that allows for tuning and adapting update strategies. When computational complexity concerns are taken into account, it also provides guidance on how ev-idence could be ordered for updating. This has important implications in working with models that convert proposi-tional logic statements from text into Dempster-Shafer theo-retic form.