Master thesis: Training Algorithms for Multilingual Latent Dirichlet Allocation
Two algorithms are implemented for training MLDA under the standard and the proposed approximate training framework:
- collapsed Gibbs sampling
- variational inference
The Gibbs sampling part is developed from [GibbsLDA++] (http://gibbslda.sourceforge.net/).
Programming languages:
- python, for scripts
- c++, for main algorithms
Extra tools required:
- Stanford segmenter, for Chinese phrase segmentation
- opencc, for traditional-to-simplified Chinese transformation