A tutorial on hidden Markov models and selected applications in speech recognition
Readings in speech recognition
Topic Detection and Tracking: Event-Based Information Organization
Topic Detection and Tracking: Event-Based Information Organization
Topic Identification in Dynamical Text by Complexity Pursuit
Neural Processing Letters
Bursty and hierarchical structure in streams
Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining
PET: a statistical model for popular events tracking in social communities
Proceedings of the 16th ACM SIGKDD international conference on Knowledge discovery and data mining
Hi-index | 0.00 |
We introduce a new model for detection and tracking of bursts of events in a discrete temporal sequence, its only requirement being that the time scale of events is long enough to make a discrete time description meaningful. A model for the occurrence of events using with Poisson distributions is proposed, which, applying Bayesian inference transforms into the well-known Potts model of Statistical Physics, with Potts variables equal to the Poisson parameters (frequencies of events). The problem then is to find the configuration that minimizes the Potts energy, what is achieved by applying an evolutionary algorithm specially designed to incorporate the heuristics of the model. We use it to analyze data streams of very different nature, such as seismic events and weblog comments that mention a particular word. Results are compared to those of a standard dynamic programming algorithm (Viterbi) which finds the exact solution to this minimization problem. We find that, whenever both methods reach a solution, they are very similar, but the evolutionary algorithm outperforms Viterbi's algorithm in running time by several orders of magnitude, yielding a good solution even in cases where Viterbi takes months to complete the search.