A1273
Title: A statistical framework for topic modeling
Authors: Anna Staszewska-Bystrova - University of Lodz (Poland) [presenting]
Peter Winker - University of Giessen (Germany)
Victor Bystrov - University of Lodz (Poland)
Abstract: A new statistical model of sparse corpora generation based on topics and a framework for comparing observed document term matrices with model assumptions are presented. The performance of the testing procedure and its limitations in the context of high sparsity are investigated by means of Monte Carlo experiments. Parameter identifiability and estimation methods based on non-negative matrix factorization are also discussed.
