Modeling events in continuous time.
Aleksandr Simma and Michael Jordan
Many processes in the real world can be best thought of in terms of events. An event occurs in a particular instant in time, has certain attributes defining properties of the specific event and may cause other events to occur in the future. As one example, consider the historical evolution of Wikipedia -- each revision is an event that has attributes (the page that was edited, who made the revision, words changed) and may cause further revisions to occur in the future. Other examples of events are packets in a computer network, text messages in a conversation and accesses to a computer system.
Using Wikipedia revisions as the setting, we build a statistical model of revisions that addresses both when revisions are expected to occur and what content they are expected to carry. The work consists of defining the general models, specializing them to the specific task and developing effective inference algorithms and parametrizations that allow both statistical and computational scaling to very large amounts of data.