Event Mining and Pattern Discovery

Abstract

time. Increasingly autonomous systems are being developed to make these sensi\xad tive decisions in critical situations. Big data has ushered in a clear departure from the earlier use of data streams, which was primarily for understanding consumer behavior or for helping people to purchase stocks. An implicit assumption in data systems in the last century was that objects are primary and events are just the properties of an object. In contrast, this century’s data systems place events at the same level as objects. What ancient philosophers [Casati and Varzi 2015] believed about the world being represented by objects and events is finally coming to computers and cyberspaces. With the increasing avail\xad ability of sensor data streams that represent diverse attributes for objects and locations, events are becoming as important as objects. As discussed in Chapter 2, an event has multiple properties, usually recognized at different levels of granularity, and represented with an event model. Depending on the complexity of an application, the event model might either contain all facets (i.e., informational, structural, experiential, spatial, temporal, and causal), or only a few of them. In its simplest form, however, an event model must contain infor\xad mational and temporal facets: the event type or name is needed as a humanand machine-understandable label, and a timestamp is needed because events occur at a certain moment in time or span an interval. Event streams have two main dimensions: (1) temporal sequence, where data are indexed by time, and (2) infor\xad mational segment, where data are encapsulated in events’ properties (such as type, name, location, participants, etc.). As shown in Figure 2.3, event streams contain a Event Mining and Pattern Discovery