Analyzing Climate Change Dialogue During California Wildfires

Abstract

This paper computationally analyzes and classifies social media dialogue on climate change based on the tweets collected and annotated during the California wildfires using a three-pronged approach. Opinions and thoughts of climate change supporters and deniers are mined through word cloud visualizations. This reveals that in the climate change debate politics and science is intertwined, with supporters stressing the imminence of climate change, and deniers deflecting it with conspiracy theories, and alternative explanations. Analysis of the metadata offers insights into how the supporting and denying tweets are being received and how they may be spread. This analysis indicates that tweets supporting climate change are shared from verified accounts, and are liked and retweeted many times, whereas those denying climate change circulate through close, like-minded communities. Sophisticated features that consider sarcasm, offensive language, emotions, and engagement are then built into a classification framework that also accounts for class imbalance. This framework can distinguish between tweets that support and deny climate change with a F1-score and accuracy of around 0.90, outperforming contemporary approaches by over 10%. By the virtue of identifying tweets that deny climate change, along with their associated justifications, the paper opens opportunities to design and disseminate educational, scientific content that can persuade the skeptics to abandon their stance.