Bursty Event Detection Model for Twitter

Huge amount of diversified information in the form of multimedia data gets uploaded to Online Social Network platform every second. This eventually gets a sudden burst during high impact events. Twitter platform plays a very important role during these events in the process of diffusion of this information across the entire social network of users. The real challenge is in the analysis of tweet during these bursty events when data gets generated in large volume with high arrival rate. Under this circumstances, near real-time detection of bursty event should be implemented to match up the speed of the information diffusion which demands efficient algorithms. In this paper a bursty event detection algorithm is proposed which considers a dynamic set of tweets in every time window and generates optimal k topics per window of a bursty event. This research has also studied the goodness of the topics produced across the different time windows. Our proposed model is successful in creating better semantically coherent and contextual topics for bursty event as compared to the other state of the art techniques such as Latent Dirichlet Allocation Model, Gibbs Sampling Dirichlet Mixture Model and Gamma-Poisson Mixture Topic Model.