Eliciting expert beliefs on the structure of a Bayesian Network

The elicitation of prior beliefs about the structure of a Bayesian Network is a formal step of full-Bayesian structural learning which offers the opportunity of exploiting the knowledge accumulated by an expert of the problem domain over years of research in a quantitative way. Motivating applications include molecular biomarkers in gene expression or protein assays, where the use of prior information is often suggested as a promising approach to face the curse of dimensionality. In this paper a general formalization based on propositions describing network features is developed which comprises issues like anchoring and revision. An algorithm is described to estimate the number of structures bearing a-priori relevant features in problem domains characterized by a large number of nodes.