A simple view of the Dempster-Shafer theory of evidence and its implication for the rule of combination

Abstract

The emergence of expert systems as one of the major ar-eas of activity within AI has resulted in a rapid growth of interest within the AI community in issues relating to the management of uncertainty and evidential reasoning. Dur-ing the past two years, in particular, the Dempster-Shafer theory of evidence has att,ract,ed considerable attention as a promising method of dealing with some of the basic prob-lems arising in combination of evidence and data fusion. To develop an adequate understanding of this theory re-quires considerable effort and a good background in proba-bility theory. There is, however, a simple way of approach-ing the Dempster-Shafer theory that only requires a min-imal familiarity with relational models of data. For some-one with a background in AI or database management, this approach has the advantage of relating in a natural way to the familiar framework of AI and databases. Fur-thermore, it clarifies some of the controversial issues in the Dempster-Shafer theory and points to ways in which it can be extended and made useful in AI-oriented app1ications.l The Basic Idea The basic idea underlying the approach in question is that in the context of relational databases the Dempster-Shafer theory can be viewed as an instance of inference from second-order relations, that is; relations in which the en-tries are first-order relations. ’ To clarify this point, let