Pattern set kernel

Abstract

Frequent pattern mining has been used in many applications of data mining. One of the reasons for the effectiveness of frequent pattern methods is that frequently occurring patterns can capture crucial aspects of the underlying semantics of the data. Thus, a good set of frequent patterns obtained from a data set can serve as a high level representation for the data. Hence, an interesting question is that of quantifying the similarity between sets of patterns. Such a similarity measure allows us to compare different data sets by comparing the sets of patterns mined from the data. In this paper we address this problem of quantifying similarity between two sets of patterns.