Optimization of LDA parameters

The aim of topic modeling is to automatically discover topics in large collections of documents. Although it is used in many different fields, the questions of how to eliminate topic instability and how to optimize model parameters are not fully answered yet. In this study, optimization of parameters of Latent Dirichlet Allocation (LDA) model, which has been one of the most preferred topic modeling methods, is investigated. For this purpose, a parallel differential evolution algorithm with two cost functions (LDADE and Word2Vec) has been implemented and applied to two data sets with different properties. The results obtained from the application are discussed in detail.