Nonparametric Bayes Pachinko Allocation for super-event detection in Twitter

Event detection in social media has been an area of considerable research in the past five years. Most models used in event detection, however, assume that events will be bursty and discrete. What does not exist, as of yet, is a model for discovering overlapping events. This paper offers a model for automatically detecting long-lasting super-events that contain smaller sub-events. This model refines Diao and Jiang's unified model for events and topics in Twitter, supplemented with the Nonparametric Pachinko Allocation Method.