Fusion of data from sources with different levels of trust

In the context of this paper, trust is defined to be “a measure of to what degree an information source is believed to be capable of producing information that conforms to fact.” No standard method has been adopted by the intelligence community for fusing data from sources with different levels of trust. This paper proposes an approach that extends the standard application of Bayesian inference to allow for the fact that any piece of intelligence data may be less than fully trustworthy. Based on a prototypical intelligence scenario from which synthetic data was generated, results indicate that trust models produce results which are closer to the ground truth than those for a model containing no trust variables, exhibit less variability and which provide a better basis for making correct decisions.