Additive similarity and dissimilarity measures (Q2398061)

This paper is continuation of the study on formal relations between diverse formulations of the notions of equality and inequality between fuzzy sets presented in [\textit{I. Couso} et al., Inf. Sci. 229, 122--141 (2013; Zbl 1293.03017)]. Now, the considerations were restricted to the case of additive measures. Many existing notions of measures satisfy the property of additivity or at least they can be written as monotone functions of additive measures. A measure of similarity (dissimilarity) is said to be additive when it can be decomposed as the sum of the particular similarities (dissimilarities) of the compared memberships over the different elements of the universe. The authors note that as a consequence some formal relations between existing formalizations of the notions of similarity and dissimilarity arise under this additional assumption. Among additive measures of similarity and dissimilarity, two main categories of comparison measures in the literature may be distinguished: those that depend on both the intersection and the difference of the compared items and those that only depend on their difference. It is worth to mention that the property of additivity is important from a practical point of view since it usually simplifies the computational effort. In the paper, a list of the properties is given included in the most relevant definitions of equality (similarity, similitude, resemblance) as well as those of inequality (distance, divergence, dissimilarity). These properties are divided into three groups of general properties (that appear in both kinds of definitions), properties of equality and properties of inequality. The authors check that there are no additional dependencies between the different formulations of the notion of similarity apart from those that can be straightforwardly derived. On the other hand, some interesting conclusions concerning the formal relations between different additive measures of inequality have been shown. Namely under additivity assumption, the notions of distance and divergence are equivalent to each other. Moreover, both of them are strictly weaker than the notion of symmetric dissimilarity.