Topic Modeling To Contextualize Event-Based Datasets: The Colombian Peace Process

Abstract

Colombia suffered civil conflict for over five decades resulting in thousands of deaths and kidnappings and millions of displaced citizens. A peace process between the government and the Revolutionary Armed Forces of Colombia (FARC) was negotiated in 2016. Quantifying public sentiment during the process may help us understand the role of social media in shaping opinions and influencing decision makers. Obtaining these viewpoints using traditional survey approaches is costly and logistically challenging. Instead, we used Twitter and news data between 2010-2018 to analyze trends before, during, and after the settlement. We used unsupervised learning methods to identify topics and measure their sentiment over time; we then compare those results to events in the Integrated Crisis Early Warning System (ICEWS) dataset.