A generalization of Bayesian inference in the Dempster-Shafer belief theoretic framework

In the literature, two main views of Dempster-Shafer (DS) theory are espoused: DS theory as evidence (as described in Shafer's seminal book) and DS theory as a generalization of probability. These two views are not always consistent. In this paper, we employ the generalized probability view of DS theory to arrive at results that allow one to perform Bayesian inference within the DS theoretic (DST) framework. The importance of this generalization is its capability of handling a wider variety of data imperfections, a feature inherited from the DST framework. In the process of developing these results akin to Bayesian inference, we also arrive at an evidence combination strategy which is consistent with the generalized probability view of DS theory, a feature lacking in the popular Dempster's combination rule (DCR). Finally, using the data from a political science survey, we demonstrate the application of our results on an experiment which attempts to gauge the hidden attitude of an individual from his/her observed behavior.