Quality of Information Sources in Information Fusion

Abstract

Pieces of information can only be evaluated if knowledge about the quality of the sources of information is available. Typically, this knowledge pertains to the source relevance. In this chapter, other facets of source quality are considered, leading to a general approach to information correction and fusion for belief functions. In particular, the case where sources may partially lack truthfulness is deeply investigated. As a result, Shafer's discounting operation and the unnormalised Dempster's rule, which deal only with source relevance, are considerably extended. Most notably, the unnormalised Dempster's rule is generalised to all Boolean connectives. The proposed approach also subsumes other important correction and fusion schemes, such as contextual discounting and Smets' a-junctions. We also study the case where pieces of knowledge about the quality of the sources are independent. Finally, some means to obtain knowledge about source quality are reviewed.

Distributed combination of belief functions

“…Since its development by Shafer [35] following Dempster's seminal work on statistical inference [8], Dempster-Shafer (DS) theory has been widely used as a formal framework for uncertain reasoning [10,11]. In the past thirty years, it has been used extensively in a large number of applications including information fusion [7,33,44], classification [14,13], clustering [15], scene perception [47], etc. DS theory is essentially a theory of evidence: it consists in representing elementary pieces of evidence pertaining to a question of interest using belief functions, i.e., completely monotone set functions [35,11], and pooling them using some appropriate combination rule.…”