Abstract Latent Dirichlet Allocation (LDA) has been used to support many software engineering tasks. Previous studies showed that default settings lead to sub-optimal topic modeling with a dramatic impact on the performance of such approaches in terms of precision and recall.