Soft concept hierarchies to summarise data streams and highlight anomalous changes

Abstract

A hierarchical approach is natural when managing large volumes of information, from both static (database) and dynamic (datastream) sources. Hierarchies allow progressively finer division into more specific categories, but frequently the categories are fuzzy rather than crisp. In this paper, we use fuzzy formal concept analysis to extract soft hierarchies from data. The hierarchies are used to classify data and to monitor changes over time by means of a fuzzy confidence measure for association analysis. A (simulated) stream of terrorism incident data is used as proof of concept. © Springer-Verlag Berlin Heidelberg 2010.