Statistical significance of episodes with general partial orders

摘要: Frequent episode discovery is one of the methods used for temporal pattern in sequential data. An a partially ordered set nodes with each node associated an event type. For more than decade, algorithms existed only when partial order total (serial episode) or trivial (parallel episode). Recently, literature has seen discovering episodes general orders. In frequent mining, threshold beyond which inferred to be interesting typically user-defined and arbitrary. One way addressing this issue mining been based on framework statistical hypothesis testing. This paper presents method assessing significance patterns A proposed calculate thresholds, non-overlapped frequency, would statistically significant. The first explained case injective where event-types are not allowed repeat. Later it pointed out how can extended class all episodes. calculations here also generalize existing results serial Through simulations studies, usefulness these thresholds pruning uninteresting illustrated.