Similarity searching in sequences of complex events

In this paper we present a generic similarity model for time-stamped sequences of complex business events. It builds upon the idea of deriving similarity from deviations between the pattern sequence and its best-possible representation in the candidate sequence. Which representation is considered optimal solely depends on the analyst's current focus and interest; the model thus foresees highest configurability to adequately balance aspects such as single-event similarities, order, timing, and missing events. The model is furthermore applicable for both sub-sequence searching and full-sequence matching. As an extension to the base model, we discuss enhanced patternmodeling facilities, e.g., to ensure a maximal time interval between two or more candidate events. The proposed tree-search algorithm allows for a seamless integration of such extensions.