Modelling asymmetric similarity with prominence

This paper introduces and discusses a geometrically based model, the relative prominence model, which is inspired by Tversky's finding that a factor behind asymmetric similarity seems to be 'relative prominence'. The model proposes that the experienced directed similarity from I to J is proportional to some symmetric similarity measure between I and J, and the quotient between the 'prominences' for J and I. Analysis of several empirical data sets shows that it is possible for a procedure to estimate the parameters of the model quite well. The differences between the relative prominence model and related models that handle asymmetry in terms of 'stimulus bias' are discussed.