Temporal Trends of International Relations from Event Data

  • Paiju Chang

Student thesis: Master's Thesis


Recent improvements in computational linguistics, data processing and storage have made it possible to automatically extract high resolution data about political events directly from the news. This offers new possibilities for studying international political interaction networks, which have traditionally relied on data about formal alliances and trade among nations. We analyzed 62 million machine-coded political events from the Global Database of Events, Language and Tone (GDELT), spanning the years 1979–2014. The vector of outgoing event counts from a given country towards all other countries characterizes the level of engagement (positive or negative) of that country with the rest of the world. We find that the un-normalized Shannon entropy of this vector increases over time for most countries. This signifies increased connectivity and equality in the behavior employed by an arbitrary nation towards others. Breaking down events by type (e.g. cooperation vs. conflict) reveals the same trend, which may be attributed to increased globalisation and interdependence over time. One way to characterize the attitude of one country towards another is to consider both the quantity and quality of the relevant events. We measure attitude by combining the importance of each event (based on its proportional news coverage) and its impact on the target country’s stability (based on a well-established conflictcooperation scale). By aggregating the incoming attitude edges for a given country, we obtain a measure of the country’s standing. Comparing this measure between the periods (1979-1996) and (1997-2014) reveals that some countries (e.g. China, India, Russia) have experienced increased overall cooperation from the rest of the world). This corresponds to their increased wealth and inclusion in the international community. Other countries seemed to be recipients of decreased aggregate cooperation over time. Using the same metric, we can characterize the behavioral profile of a particular country by investigating the precise manner in which it treats other countries –i.e. its outgoing attitude vector. Computing the cosine similarity between these vectors allows us to measure the similarity between the behavioral profiles of different countries. When combined with cluster analysis, this reveals significant clustering among nations back in the 1980’s. We also observed a marked increase in polarisation in the 2010’s involving countries like Afghanistan, Iraq, Iran and Russia.
Date of AwardMay 2015
Original languageAmerican English
SupervisorTalal Rahwan (Supervisor)


  • Computational linguistics
  • data processing
  • data storage
  • data management
  • international politics
  • political science
  • international relations
  • globalization
  • news coverage.

Cite this