Moirae is a continuous monitoring system that has the ability to extract useful historical information in near-real-time. Previous work on Moirae has investigated techniques to provide the top k most similar past events to users. However, the top k results are often too similar to each other to be useful. To cope with this limitation, we propose to modify Moirae to return the top k clusters of similar past events. In our work, we revise the similarity metric used in Moirae to use a variant of the Earth Mover's Distance. This new metric is applied with several clustering algorithms to cluster our data sets. We show that our approach better presents extracted historical information while preserving Moirae's power of finding similar past events efficiently.
Advised by Magda Balazinska
CSE 203
Wednesday
May 28, 2008
3:30 - 4:20 pm