Automatic evaluation of geopolitical risk

  • John Burns

Student thesis: Doctoral Thesis (PhD)


This thesis aims to construct programs for automatically evaluating geopolitical risks. This project will use machine learning, specifically sentiment analysis, topic modeling and NER, to build new computer programs to evaluate and assess not just scheduled and predictable geopolitical events, but also unpredictable events. The gap in current literature this project intends to fill is the ability to respond to these risk events in real time. Thus, this project’s objective is to build programs that can digest the vast quantities of data generated by Twitter / X, the data source chosen for this project, focusing on keywords indicating a potential geopolitical event or crisis.

With the use of Twitter / X, I was able to find that information appeared quicker through tweets than traditional news sources, thus I was able to identify emerging geopolitical topics, in some cases, hours or days before they became discussed in the mainstream media. I also achieved success with building a geopolitical risk index program with sentiment analysis to relate the index to the trends in the financial markets surrounding the start of the Ukraine War in 2022 at the daily level. Using Granger Causality, I found that the geopolitical risk index I created from the emotions gleaned from the sentiment analysis of the relevant tweets collected, contained predictive information of the movement of various financial assets over time. In addition, with the NER program, I was able to visualize the different geopolitical risks on a world map. While I managed to create a program that created a geopolitical risk index at the real time level, unfortunately, there was little relationship between the real time risk index and the change in financial markets. In combination, the output of these various programs allowed for the automatic evaluation of geopolitical risk.
Date of Award12 Mar 2024
Original languageEnglish
Awarding Institution
  • University of St Andrews
SupervisorTom Kelsey (Supervisor) & Carl Robert Donovan (Supervisor)


  • Geopolitics
  • Machine learning
  • Sentiment analysis
  • NLP
  • Topic modeling
  • NER
  • Twitter / X
  • Multilingual analysis
  • Text analysis

Access Status

  • Full text open

Cite this