Research regarding topic modeling on the coverage of the Nusantara Capital City (IKN) in national mass media remains limited. This study aims to not only model IKN-related topics but also rigorously evaluate the Latent Dirichlet Allocation (LDA) model to ensure its robustness for future implementation. The dataset comprises 1,498 news articles gathered from prominent Indonesian online media, specifically Detik (1,050 articles) and Kompas (448 articles). The methodology involves experimental variations of LDA parameters, including document volume, maximum features, and topic count, utilizing the Scikit-learn library. The results indicate that an increase in data volume and feature dimensions significantly correlates with longer computation times and a higher number of epochs required for convergence. Furthermore, the expansion of variables and data volume resulted in more negative log-likelihood values and increased perplexity, suggesting that model complexity challenges predictive precision. A convergence threshold of $1e^{-2}$ was applied to optimize the training cessation point. While this study establishes a baseline for static topic modeling, future research implies the necessity of Dynamic Topic Modeling (DTM) to capture the temporal evolution of topics, a dimension not addressed by the standard LDA model.