Novel Topic Models for Parallel Topics Extraction from Multilingual Text

In this work, we propose novel topic models to extract topics from multilingual documents. We add more flexibility to conventional LDA by relaxing some constraints in its prior. We apply other alternative priors namely generalized Dirichlet and Beta-Liouville distributions. Also, we extend finite mixture model to infinite case to provide flexibility in modelling various topics. To learn our proposed models, we deploy variational inference. To evaluate our framework, we tested it on English and French documents and compared topics and similarities by Jaccard index. The outcomes indicate that our proposed model could be considered as promising alternative in topic modeling.