Topic modelling of ukraine war-related news using latent dirichlet allocation with collapsed Gibbs sampling

The context of this research is the application of topic modeling to war-related news in the context of the Ukraine war. The objective of the research is to use Latent Dirichlet Allocation (LDA) with Collapsed Gibbs sampling to identify distinct content groups in war-related news. The method used in the research involves data scraping from a Ukrainian news website, data preprocessing, and applying the LDA with Collapsed Gibbs algorithm to infer the latent topics within the corpus. The results of the research include the identification of twelve distinct topics and the corresponding keywords that characterize each topic. The analysis of the results provides insights into the context of each topic, such as discussions on safety measures during wartime, consequences of military actions, and reports on military casualties. The research concludes that the application of LDA with Collapsed Gibbs is a valuable tool for identifying and understanding the context of war-related news. However, there may be discrepancies between the results of the model and human interpretation, which may be due to limitations in the results, model parameters, and the presence of noise data. Future research should focus on optimizing model parameters, filtering noise data, and improving the analysis of topic context to enhance the reliability and interpretability of the results.