A framework for data mining pattern management

Abstract

To represent and manage data mining patterns, several aspects have to be taken into account: (i) patterns are heterogeneous in nature; (ii) patterns can be extracted from raw data by using data mining tools (a-posteriori patterns) but also defined by the users and used for example to check how well they represent some input data source (a-priori patterns); (iii) since source data change frequently, issues concerning pattern validity and synchronization are very important; (iv) patterns have to be manipulated and queried according to specific languages. Several approaches have been proposed so far to deal with patterns, however all of them lack some of the previous characteristics. The aim of this paper is to present an overall framework to cope with all these features. © Springer-Verlag Berlin Heidelberg 2004.