Using Data Visualization to Explore International
Trade Agreements

Esteban Serrano - Oliver Ford - Xinyu Du

For this work we have been using data visualization as a means to gain insight into the underlying structure and relationships of international trade agreements. Our dataset is comprised of 450 preferential international trade agreements with the majority (424) being in english. The original dataset is open to the public and can be found here.

Using data mining and visualisation techniques is interesting in this context for two reasons. First, trade agreements have meaningful impact on the world, however, gaining an overview of these relationships using traditional methods is difficult. Thus, using data mining in combination with visualisation techniques to show how countries interact through trade agreements may uncover new insights into the landscape of international economic governance. Second, there has been little previous work towards using computational analysis and visualisation in the domain of international trade, making this work especially exciting.

For this work we have two main areas of inquiry.

  1. First, is it possible to uncover the relationships between countries by analysing the legal documents and, if so to what extent?
  2. Second, the relationships between the documents themselves, for example, do previous trade agreements impact later ones?

Topic Modeling

For out first series of visualisations we investigate the similarity and possible relationships between trade agreement documents. We do this using topic modeling in combination with clustering as well as techniques to visualise these clusters.

Network Analysis

We used network analysis to show international trade relationships between countries and to explore the link between geopolitics and trade agreement in different regions. With it we can visualize the number of trade agreements between countries and the trade signing status of each region.

Word Embeddings

Word embeddings is a technique for representing words based on the context in which they appear. It is capable of capturing a measure of similarity between words by associating those that appear frequently. In this case we use an algorithm called Word Mover Distance (WMD) for measuring the distance (similarity) between trade agreements. The model we are using has been pretrained on wikipedia documents (glove-wiki-gigaword-300).