In this paper, we introduce a diamond episode of the form s1 -> E -> s2, where s1 and s2 are events and E is a set of events. The diamond episode s1 -> E -> s2 means that every event of E follows an event s1 and is followed by an event s2. Then, by formulating the support of diamond episodes, in this paper, we design the algorithm FreqDmd to extract all of the frequent diamond episodes from a given event sequence. Finally, by applying the algorithm FreqDmd to bacterial culture data,we extract diamond episodes representing replacement of bacteria.
Recently, knowledge discovery in large data increases its importance in various fields. Especially, data mining from time-series data gains much attention. This paper studies the problem of finding frequent episodes appearing in a sequence of events. We propose an efficient depth-first search algorithm for mining frequent serial episodes in a given event sequence using the notion of right-minimal occurrences. Then, we present some techniques for speeding up the algorithm, namely, occurrence-deliver and tail-redundancy pruning. Finally, we ran experiments on real datasets to evaluate the usefulness of the proposed methods.
CiteSeerX - Document Details (Isaac Councill, Lee Giles): Sequences of events describing the behavior and actions of users or systems can be collected in several domains. We consider the problem of discovering frequently occurring episodes in such sequences. An episode is defined to be a collection of events that occur relatively close to each other in a given partial order. Once such episodes are known, one can produce rules for describing or predicting the behavior of the sequence. We give efficient algorithms for the discovery of all frequent episodes from a given class of episodes, and present extensive experimental results. The methods are in use in telecommunication alarm management.
Event Sequence arises naturally in many applications. Episode mining can discovery the knowledge hidden in the event sequence. Currently, the most influential algorithm for episode mining is WINEPI. However, it is likely to suffer from the tendency of generating too many of candidate episodes. In this paper, a novel algorithm named DRE for mining frequent episodes is presented. It studied the conditions for the events which can be pruned from the database, so the size of database is reduced gradually. The performance of algorithm DRE was evaluated and compared with WINEPI algorithm. The results demonstrate that the DRE has better performance.