MML-Based Approach for Determining the Number of Topics in EDCM Mixture Models

摘要

This paper proposes an unsupervised algorithm for learning a finite mixture model of the exponential family approximation to the Dirichlet Compound Multinomial (EDCM). An important part of the mixture modeling problem is determining the number of components that best describes the data. In this work, we extend the Minimum Message Length (MML) principle to determine the number of topics (clusters) in case of text modeling using a mixture of EDCMs. Parameters estimation is based on the previously proposed deterministic annealing expectation-maximization approach. The proposed method is validated using several document collections. A comparison with results obtained for other selection criteria is provided.