Specifying nonspecific evidence

TL;DR

This paper extends previous work on partitioning nonspecific evidence within Dempster-Shafer theory by specifying the potential event references for each piece of evidence and developing methods to handle false or less credible evidence.

Contribution

It introduces a new methodology to specify evidence to particular events and to incorporate falsity and credibility assessments into reasoning processes.

Findings

Developed a method to specify evidence to potential events.

Created a technique to assess evidence falsity and credibility.

Enhanced reasoning by discounting evidence based on these assessments.

Abstract

In an earlier article [J. Schubert, On nonspecific evidence, Int. J. Intell. Syst. 8(6), 711-725 (1993)] we established within Dempster-Shafer theory a criterion function called the metaconflict function. With this criterion we can partition into subsets a set of several pieces of evidence with propositions that are weakly specified in the sense that it may be uncertain to which event a proposition is referring. Each subset in the partitioning is representing a separate event. The metaconflict function was derived as the plausibility that the partitioning is correct when viewing the conflict in Dempster's rule within each subset as a newly constructed piece of metalevel evidence with a proposition giving support against the entire partitioning. In this article we extend the results of the previous article. We will not only find the most plausible subset for each piece of evidence as was done in the earlier article. In addition we will specify each piece of nonspecific evidence, in the sense that we find to which events the proposition might be referring, by finding the plausibility for every subset that this piece of evidence belong to the subset. In doing this we will automatically receive indication that some evidence might be false. We will then develop a new methodology to exploit these newly specified pieces of evidence in a subsequent reasoning process. This will include methods to discount evidence based on their degree of falsity and on their degree of credibility due to a partial specification of affiliation, as well as a refined method to infer the event of each subset.