| In this dissertation, we propose an event detection approach to aid in real-time event detection. Social media generates information about news and events in real-time. Given the vast amount of data available and the rate of information propagation, reliably identifying events can be a challenge. Most state of the art techniques are post hoc techniques, which detect an event after it happened. Our goal is to detect the onset of an event as it is happening, using the user-generated information from Twitter streams. To achieve this goal, we use a discriminative model to identify a sudden change in the pattern of conversations over time. We also use a topic evolution model to identify credible events and propose an approach to eliminate random noise that is prevalent in many of the existing topic detection approaches. The simplicity of our proposed approach allows us to perform fast and efficient event detection, permitting discovery of events within minutes of the first conversation relating to an event started. We also show that this approach is applicable for other social media datasets to detect change over the longer periods of time. We extend the proposed event detection approach to incorporate information from multiple data sources with different velocity and volume. We study the event clusters generated from event detection approach for changes in events over time. We also propose and evaluate a location detection approach to identify the location of a user or an event based on tweets related to them. |
