Full Bayesian Topic Models with Burstiness
Wray Buntine and Swapnil Mishra (NICTA)
NICTA SML SEMINARDATE: 2013-11-21
TIME: 11:15:00 - 12:00:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
Applications of topic modelling, like latent variable modelling generally, often involve adapting the basic model in various ways. For information retrieval, language models exhibiting the phenomena of burstiness are important. Here we show how to combine recent work to develop high-performing non-parametric topic models exhibiting burstiness and present results. The implementation is interesting because it shows a substantial performance improvement over recent variational and online methods for topic models.
