An evidential model for the analysis of competing hypotheses

Operators in intelligence and situation awareness face difficulties in their work linked to the management of inconsistent pieces of information. Some cognitive biases interfere the decision making process. The Analysis of Competing Hypotheses (ACH) structured procedure helps reduce confirmation bias through a series of steps that guide intelligence analysts. However, it does not allow for a fine representation of uncertainty. We propose an original modeling of the ACH procedure inspired by recent work highlighting the belief functions in the formulation of neural networks. This modeling allows for some automation of the process, with a rich representation of the uncertainty regarding the weight of evidence for or against the hypotheses, as well as the likelihood of the evidence. We illustrate our solution with a case study and conclude with future work aimed at quantifying the conflict between pieces of evidence.