Visualization of Preference Matrices for Labeled Objects

Thomas A. Runkler

Abstract: Preference matrices are used to quantify the pairwise degrees of preference of each object over each other object. Visualization of preference matrices helps to understand the underlying preference structures, for example to identify the most and least preferred objects, groups of objects with similar preference patters, or inconsistencies in the preference structure. A recent method for visualizing preference matrices is PrefMap. Often objects are not only associated with pairwise preferences, but also with continuous or discrete labels such as price or size. In this paper we extend the PrefMap visualization method to data which possess such labels, which we call LPrefMap. LPrefMap allows to analyze how preference is affected by the corresponding object features. Experiments with two real world preference data sets indicate that LPrefMap is a very useful tool to visually gain valuable insights in the preference structure and to understand how the object features influence the preference.