| This paper describes an approach toward the efficient optimization of hyperparameters in Latent Dirichlet Allocation (LDA) topic modeling under stringent computational constraints. The main aim is to improve performance and outcomes by fine-tuning the LDA model's alpha, beta, and topic parameters in the Mallet implementation. The difficulty comes from the time-consuming task of manually adjusting hyperparameters and the high computational expense of experimenting with different parameter combinations. To get around this, we suggest an automated hyperparameter tuning approach based on the recursive binary search algorithm, aiming to reduce the model's complexity and hence improve performance. Our process, which was created for the competition Predicting Good Configurations for Topic Models at the Genetic and Evolutionary Computation Conference, offers a quick way to optimize the hyperparameters while avoiding the use of heuristics and manual labor. The study provides an extensive description of our technique, findings, and an in-depth evaluation, making a significant contribution to the field of topic modeling's automated hyperparameter optimization. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.