An argumentative knowledge-based model construction approach for Bayesian networks

Abstract

In this paper, an argumentative knowledge-based model construction (KBMC) technique for Bayesian networks is presented. This approach allows an agent to collect and instantiate the most accepted subset of an imperfect knowledge base to dynamically construct a Bayesian network. Arguments are constructed to represent active paths through an agent's knowledge base - paths consisting of information that is computationally relevant in the evaluation of a query Pr(Q|E). Argumentation over paths is used to select the valid or most accepted information according to the preferences of the agent. This information is consequently formed into candidate network structures by accrual. This work is presented as an extension of the KBMC approach of Haddawy [5]. The potential of the approach to be used in multi-agent network construction is discussed.