Rodrigues, Paulo Manuel MarquesFadda, Pietro2023-08-252023-08-252022-06-012022-05-20http://hdl.handle.net/10362/156874Communication is set to become vital for policy transmission. Growing textual data availability requires the development of more sophisticated tools. This report investigated the empirical application of Latent Dirichlet Allocation (LDA), a Machine Learning Topic Model, with the aim of revealing topics in ECB speeches for the period 1997-2022. Results corroborate comparable studies and produce significant novelties: Topic dynamics captured structural macroeconomic events; ECB speakers differ in topic allocation and frequency coverage; Topic weights correlate with underlying variables and the model well perform on unseen documents. However, outcomes should be treated with caution as LDA lies between correlation and causality.engMachine learningEcbTopic modelingLda modelCentral bank communicationmaster thesis203064445