3Department of Informatics Engineering,

Abstract: Traditionally, similarity between two objects is calculated by using only their attribute values, such as number of coincided attributes, Euclid distance, etc. A new concept of similarity dealing with a uniqueness measure is proposed in this paper by which the similarity between two objects is not only considered using their attribute values, but also a subset of objects as an important parameter. Here, the subset of objects may be regarded as knowledge of human. In this concept of similarity, if attribute values of two objects are rare in the subset, and their attribute values are the same, then their degree of similarity is high. On the other hand, if the attribute values of two objects are not rare in the subset, and their attribute values are the same, then their degree of similarity is low. Consequently, the degree of similarity between two objects will be changed depending on the subset of objects. Moreover, we discuss mathematical properties of the concept of similarity dealing with uniqueness measure. Finally, we discuss differences between the concept of similarity based on uniqueness measure and traditional similarity using some examples