Topological Proximity and Relevance in Graphical Decision Models

Eric Horvitz and Kim Leng Poh


We describe qualitative relationships that can be employed to determine the relative value of information for variables in prototypical influence diagrams based solely on a consideration of topological relationships. We also discuss how we can use a qualitative analysis of expected value of information to provide upper and lower bounds on the value of information for chance variables in these decision models. Finally, we describe several general results about the magnitude of the expected value of information associated with chance nodes in terms of their position and relationships in an influence diagram. The results of the qualitative analysis of utility demonstrate that qualitative analysis of utility can be used to assist decision analysts to efficiently identify sources of information that promise the greatest returns. The qualitative utility methods also provide researchers investigating computer-based decision systems with efficient means for controlling model construction and inference for offline and real-time reasoning settings.


Publication typeTechReport
InstitutionMicrosoft Research
> Publications > Topological Proximity and Relevance in Graphical Decision Models