Detecting Climate Change Deniers on Twitter Using a Deep Neural Network

Abstract

Climate change or global warming is a global threat to both human communities and natural systems. In recent years, there is an increasingly public debate on the existence of climate change or global warming, but data describing such discussions are difficult to access. Social media provide a new data source to survey public perceptions and attitudes toward such topics. However, enabling computers to automatically determine users' attitudes towards climate change based on social media contents is still challenging. Taking Twitter data as an example, this study analyzed public discussions about climate change and global warming in year 2016. The objectives are: (1) to develop an optimized Deep Neural Network (DNN) classifier to identify users who are climate change deniers based on tweet contents; (2) to examine the temporal patterns of climate change discussions on Twitter and its driving factors. Results demonstrate that the developed DNN model successfully identified climate change deniers based on tweet contents with an overall accuracy of 88%. There are more climate change discussions during September to December 2016, whereas the percentages of climate change deniers were lower in the same period. Public interests and attitudes on climate change were driven by extreme weather events and environmental policy changes. The developed methodology will shed lights on the utility of deep learning in natural language processing, while the results provide improved understanding of factors affecting public attitudes on climate change.