CrystalBall: A Visual Analytic System for Future Event Discovery and Analysis from Social Media Data

Abstract

Social media data bear valuable insights regarding events that occur around the world. Events are inherently temporal and spatial. Existing visual text analysis systems have focused on detecting and analyzing past and ongoing events. Few have leveraged social media information to look for events that may occur in the future. In this paper, we present an interactive visual analytic system, CrystalBall, that automatically identifies and ranks future events from Twitter streams. CrystalBall integrates new methods to discover events with interactive visualizations that permit sensemaking of the identified future events. Our computational methods integrate seven different measures to identify and characterize future events, leveraging information regarding time, location, social networks, and the informativeness of the messages. A visual interface is tightly coupled with the computational methods to present a concise summary of the possible future events. A novel connection graph and glyphs are designed to visualize the characteristics of the future events. To demonstrate the efficacy of CrystalBall in identifying future events and supporting interactive analysis, we present multiple case studies and validation studies on analyzing events derived from Twitter data.