Measuring the Similarity for Heterogenous Data: An Ordered Probability-Based Approach

Abstract

Abstract. In this paper we propose a solution to the similarity measuring for heterogenous data. The key idea is to consider the similarity of a given attribute-value pair as the probability of picking randomly a value pair that is less similar than or equally similar in terms of order relations defined appropriately for data types. Similarities of attribute value pairs are then integrated into similarities between data objects using a statistical method. Applying our method in combination with distance-based clustering to real data shows the merit of our proposed method.

FuzzyProbSim: Fuzzy probability as similarity measure

“…that is, Sim(a, b) is the tail probability of the χ 2 (2n) [LH04], [RPR08]. Figure (7) shows the density and distributions functions, and the tail probability for a χ 2 distribution.…”