Peak Detection

Summary

Peak Detection is a method for finding bursty structure in streams. Example of peak detection strategies are Rapid Rising and Rise and Fall.

In Rapid Rising, the count of documents containing an entity of interest in a time window c is compared with the counts of documents containing the entity in the leading k windows. The entity is said to be bursting iff c is more than l standard deviations away from the mean count of those leading windows.

In Rise and Fall, both leading and trailing k windows are compared. The entity is said to be bursting iff c is more than l standard deviations away from the mean count of both leading and trailing windows.

References / Links

• Kleinberg, J. Bursty and hierarchical structure in streams. KDD 2002.