Context-Dependent Similarity

Numerical similarity measures are used to describe the relative ranks of the similarity of objects or cases in many artificial intelligent systems. These measures are usually absolute and contextindependent. On the other hand, humans perceive context-dependent similarity. That is, the ranking of similarity between pairs of objects is varying under a changing context. We consider this problem as the construction of numerical formulas that satisfy certain axioms and criteria. An entropy-related formula is proposed and its implementation in a changing environment is considered. A demonstration of this formula on a well-known context-dependent similarity assessment is given.