UCL PhD candidate Sonja Tilly will be speaking at the ODSC Europe 2020 Virtual Conference that runs from September 17-19 on "Forecasting the Economy with Fifty Shades of Emotions":
Macroeconomic short-term forecasting, often referred to as nowcasting, has emerged in recent years, improving the monitoring of the global economy. However, existing approaches tend to focus on conventional economic indicators for their predictions, few tackle “alternative datasets” such as text. Those works that use text are often restricted to anglophone media and extract negative and positive sentiment only. My research expands the existing body of research, capturing the complexity of emotions contained in global news articles. The Global Database of Events, Language and Tone (GDELT) monitors world media from every possible angle and is arguably the largest database of human society. Version 2 incorporates real-time translation from 65 languages and measures over 2,300 emotions and themes from every news article, updated every 15 minutes. The Global Knowledge Graph (GKG), one of the tables available within GDELT, contains fields such as emotions and themes extracted from newspaper articles since 2015, amounting to 11 terabytes. Learning outcomes: This talk introduces a multi-step methodology for filtering and aggregating emotions from GDELT’s GKG. The predictive power of these emotions will be tested for different economies around the world, showing that emotions add value to short-term economic forecasts. The talk concludes discussing some of the issues encountered during developing the filtering methodology and when incorporating the data into a forecasting framework.