Twitter-based election prediction in the developing world

Abstract

Elections are the main instrument of democracy. Citizens decide which entity or entities (a political party or a particular politician) should represent them. Traditionally, preelection polls have been used to learn about trends and likely election outcomes. Predicting an election outcome based on user activity on Twitter has been shown to be a cheap alternative. While past research has focused on election prediction in the developed world (where its use is debatable), in this paper we provide a comprehensive argument for the use of Twitter-based election forecasting in the developing world. For our use case of Indonesia’s presidential elections 2014, the most basic Twitter-predictor outperforms the majority of traditional polls, while the best performing predictor outperforms all traditional polls on the national level.