Our project proposes a new approach to estimate and understand the underlying preference and relationship structure of political actors. In particular, we go beyond traditional text data and focus on videos of speeches at the United Nations as a way to predict emotions for topics that states address at the international stage. The goal of our project is twofold. On the one hand, we provide a novel, open-source data set with more than 450 hours of labelled video data for the past seven years of United Nations General Debate speeches and complement a recently published text corpus on UN speeches. On the other hand, we showcase how this data can help to understand subtle differences in preference structures for specific topics by international actors that are difficult to extract from a standardized diplomatic text and that can only be revealed when using the accompanying video data. Our project was developed and (pre-)tested during the GESIS 2019 CSS Summer School on Computational Social Sciences in Berlin which was awarded with an honorable mention for an outstanding research project and received the CDSS Young Scholar Award 2020. This research project was supported by Tim Hofmann.