TKE: Mining Top-K Frequent Episodes

Abstract

Frequent episode mining is a popular data mining task for analyzing a sequence of events. It consists of identifying all subsequences of events that appear at least minsup times. Though traditional episode mining algorithms have many applications, a major problem is that setting the minsup parameter is not intuitive. If set too low, algorithms can have long execution times and find too many episodes, while if set too high, algorithms may find few patterns, and hence miss important information. Choosing minsup to find enough but not too many episodes is typically done by trial and error, which is time-consuming. As a solution, this paper redefines the task of frequent episode mining as top-k frequent episode mining, where the user can directly set the number of episodes k to be found. A fast algorithm named TKE is presented to find the top-k episodes in an event sequence. Experiments on benchmark datasets shows that TKE performs well and that it is a valuable alternative to traditional frequent episode mining algorithms.