Learning to Classify: UN Communication of Climate Change on Twitter

• In Event: Cutting-Edge Research Methods in Climate Governance Research

Abstract

International organizations (IOs) have an important role to play in global climate governance coordinating fragmented efforts of actors at different scales. These efforts by IOs are well-communicated on social media platforms, like Twitter, that have become a legitimate tool for IOs' public communication with stakeholders. As the volume of climate texts published on social media by IOs rapidly grows, social scientists face a serious challenge related to coding of these data. For instance, manual coding of the available textual data in a sample that would at least vaguely resemble a representative one becomes infeasible. To address these challenges, scholars turn to automated labeling techniques. Methodologically, this article contributes to climate governance literature by systematically reviewing available approaches to text classification for climate research. In doing so, the paper compares the performance of various approaches frequently used in existing climate governance research that builds on text analysis methods. Empirically, the paper builds on data from UN organizations that are actively engaged into climate governance. Using Twitter data collected from the official accounts of the UN organization, we demonstrate that supervised machine learning methods (in particular, neural networks with pretrained embeddings) outperform lexicon-based classifiers.

Authors

• Lisa Maria Dellmuth, Stockholm University

• Karina Shyrokykh, Stockholm University