Logo Logo
Switch Language to German

Schönfeld, Mirco; Eckhard, Steffen; Moiseeva, Alena; Patz, Ronny and Meegdenburg, Hilde van (2018): The UN Security Council Transcripts. From Automated Topic Modeling to Speeches-as-Network. Panel From Texts to Networks: Semantic, Socio-Semantic and Discourse Networks. European Consortium for Political Research General Conference, Hamburg, 22. - 25. August 2018.

Full text not available from 'Open Access LMU'.


The United Nations Security Council (UNSC) is the highest authority in the domain of international security. It brings together the global powers China, France, Russia, the United Kingdom and the United States as Permanent Five members and changing constellations of ten other, non-permanent members. Most meetings are registered, transcribed, and translated to all official UN languages in public protocols. Each meeting covers one major topic of geopolitical importance, often related to single countries or conflicts. We propose a novel method of analyzing political topic networks based on these UNSC protocols. Distilling a corpus of speeches together with information about each speaker's nationality from meeting protocols through automated speech extraction, we model the topics discussed by the speaker with LDA-based topic modeling techniques. Based on these country-topic dyads, we construct two-mode political topic networks. From meeting to meeting, different topical closeness between participants becomes apparent, and the resulting network portrays international diplomatic relations between participants of UNSC meetings. For this paper, and selecting from the huge online database of UNSC protocols and speeches, we focus on all meetings regarding Afghanistan in the period from 2001 until 2017. This confined subset allows to qualitatively validate the automated extraction, topic-modeling and speeches-as-network display: The timeframe of external events as well as actors’ preferences are known from previous studies and can be juxtaposed with the results of the automated analysis. The paper demonstrates first that the automated content analysis is able to identify sub-topics in the Afghanistan debate as well as the trends in their relative importance over time. Second, the network analysis confirms that in this debate Permanent Five and non-permanent members emphasize different sub-topics. These findings underline how automated topic modeling is able to represent key structures of differentiation in the international system alongside overall sub-topics trends over time. We discuss these findings as well as open methodological challenges in view of scaling up our analysis to a complete corpus of UN Security Council speeches.

Presented by Mirco Schönfeld.

Actions (login required)

View Item View Item