Discovering trends and relationships among rules

Abstract

Data repositories are constantly evolving and techniques are needed to reveal the dynamic behaviors in the data that might be useful to the user. Existing temporal association rules mining algorithms consider time as another dimension and do not describe the behavior of rules over time. In this work, we introduce the notion of trend fragment to facilitate the analysis of relationships among rules. Two algorithms are proposed to find the relationships among rules. Experiment results on both synthetic and real-world datasets indicate that our approach is scalable and effective. © 2009 Springer Berlin Heidelberg.