ALCOMFT-TR-03-196

Gemma Casas-Garriga
Towards a formal Framework for mining Mining General Patterns from Structured Data
Barcelona. Work packages 1 and 4. December 2003.

Abstract: In this paper we present the initial notions of a framework for mining general patterns from complex structured data. To achieve this goal, we start by focusing here on the most basic case: the mining of a plain ordered collection of data, such as sequential databases or time-series data. Our framework develops on the hypothesis of using the closure of the Galois connection as a way to reduce the number of the extracted general patterns to the most representative ones. So, we employ formal concept analysis to characterize the concept lattice of these ordered contexts; our main contribution is a new Galois connection that, in this work, allows to exactly derive the closed sequential patterns found by a recent algorithm (CloSpan, [Clospan]).

Postscript file: ALCOMFT-TR-03-196.ps.gz (72 kb).