Combining Facts and Expert Opinion in Analytical Models via Logical and Probabilistic Reasoning

摘要

This report describes work on an integrated system that can assist analysts in exploring hypotheses using Bayesian analysis of evidence from a variety of sources. The hypothesis exploration is aided by an ontology that represents domain knowledge, events, and causality for Bayesian reasoning, as well as models of information sources for evidential reasoning. We are validating the approach via a tool, Magellan, that uses both Bayesian models and logical models for an analyst's prior and tacit knowledge about how evidence can be used to evaluate hypotheses. We also describe how we combine logic information, in the form of proofs provided by the natural deduction system SILK(Semantic Inferencing on Large Knowledge) and probabilistic information, represented by Bayesian networks, in the BRUSE(Bayesian Reasoning Using Soft Evidence) system.