Markov logic networks for multi-level fusion support to intelligence analysis

Intelligence analysts are faced with a complex information fusion problem that is characterised by the volume, velocity, variety and veracity of the observed data and information. That is, there are large quantities of data and information (volume) arriving at a high rate (velocity) which are, in general, highly heterogeneous (variety) and of inconsistent fidelity (veracity). Making sound decisions based on these observations is the aim and human decision making is essential, since humans are ultimately held responsible. Current industry tools are designed to support this process, by helping to process and curate the collection of observations automatically, storing it in databases and providing various analytical tools to retrieve and manipulate fragments of the collection. However, industry approaches for managing the inherent uncertainty in these observations and exploiting all the available higher level contextual information are inadequate. What is needed is a practical formalism that can deal with multiple types of uncertainty, can exploit contextual information and can operate at scale to reduce the cognitive burden on analysts. In this paper, we discuss the use of Markov logic networks for handling uncertainty and exploiting higher level contextual information and demonstrate how this provides a framework which is well suited to handling the real-world issues encountered in intelligence-based problems.