Method for identifying patterns

TL;DR: In this article, an edge-weighted graph is generated in which each event type is represented by a vertex and a window of pre-determined width is then moved through an event stream and the edges in the graph are weighted in response to the co-occurrence of events in the window.

Abstract: The method identifies repeating patterns in sequential event streams, in which each event (A 1 , B, C, D, A 2 ) is of a pre-determined event type (A, B, C, D) and occurs at a specified position within an event stream. In the method, an edge-weighted graph is generated in which each event type is represented by a vertex. A window of pre-determined width is then moved through an event stream and the edges in the graph are weighted in response to the co-occurrence of eventsin the window. Thus, if a window contains an event A 1 and moves on through the event stream such that it also contains a second event B, the weighting of the edge linking the vertices representing the corresponding event types A, B in the graph is incremented. Repeating patterns in the event stream thus emerge as clusters of higher edge weights in the graph and can be detected by a graph clustering method. Individual instances of a pattern identified by the method can optionally be located by re-scanning the event stream for individual patterns.