A Bayesian Paradigm for Dynamic Graph Layout

Abstract

Dynamic graph layout refers to the layout of graphs that change over time. These changes are due to user interaction, algorithms, or other underlying processes determining the graph. Typically, users spend a noteworthy amount of time to get familiar with a layout, i.e. they build a mental map [ELMS91]. To retain this map at least partially, consecutive layouts of similar graphs should not differ significantly. Still, each of these layouts should adhere to constraints and criteria that have been specified to improve meaning and readability of a drawing.In [BW97], we introduced random field models for graph layout. As a major advantage of this formulation, many different layout models can be represented uniformly by random variables. This uniformity enables us to now present a framework for dynamic layout of arbitrary random field models. Our approach is based on Bayesian decision theory and formalizes common sense procedures. Example applications of our framework are dynamic versions of two well-known layout models: Eades´ spring embedder [Ead84], and Tamassia´s bend-minimum orthogonal layout model for plane graphs [Tam87].