Measuring Connection Strengths and Link Strengths in Discrete Bayesian Networks

Abstract

This paper discusses measures for connection strength (strength between any two nodes) and link strength (strength along a specific edge) in Discrete Bayesian Networks. The typical application is to visualize the connections in a Bayesian Network learned from data to learn more about the inherent properties of the system (e.g. in earth sciences, biology or medicine). The paper focuses on measures based on mutual information and conditional mutual information. The goal is to provide an easy-toread document that gives clear reasoning for existing measures, provides some simple extensions (modified measures for different applications), discusses the limitations of the measures, provides enough interpretation to aid a scientist in selecting the most appropriate one and suggests some new uses for link strength.