Analyzing theme penetration from specialized to non-specialized discourses with NLP: the case of the IPCC Special Report Global Warming of 1.5 ºC

Catellani, Andrea [UCL] François, Thomas [UCL] Watrin, Patrick [UCL] Fastrez, Pierre [UCL] Matagne, Julie [UCL]

Discourses as scientifically sharp and socially influential as the IPCC reports raise the question of their relationship with non-specialists discourses on climate change. IPCC reports are officially aimed at governments (IPCC, 2019), and their reception by the general public is certainly filtered by legacy media (O’Neill et al. 2015; Painter, 2014; Kunelius, 2017) and shaped through social media (Pearce et al., 2014; O’Neill et al., 2015). Social media discourses are known to be sometimes polarized and over-simplified, and one can expect such discourses to focus only on some of the IPCC reports topics. In this research, we use natural language processing (NLP) techniques as well as a qualitative analysis to identify the main concepts and topics in the Special Report Global Warming of 1.5ºC by the IPCC (2018) and track their presence and recycling in a sample of the general public discourses on social media. Our aim is to understand how topics covered in the last published IPCC report are represented in social media discourses, as a step towards a deeper understanding of the appropriation of specialized discourses on climate change (like IPCC documents) by the general public . The contribution of this communication is twofold. First, whereas other NLP-based investigation of climate discourse have mainly relied on Twitter data (O’Neill et al., 2015; Newman, 2017; Yagodin et al., 2018), our approach will make use of a different type of digital media data, namely Reddit. Reddit is a social platform where news and opinions are shared and debated. As such, it can be a valuable source of data on climate change discourses, and also provides new new perspectives to Twitter-based analyses. Second, previous studies have carried out rather limited topic analyses of Tweets, either manually classifying a small sample of Tweets into a few categories (Newman, 2017) or using the most frequent hashtags as topics (O’Neill et al., 2015). In this study, we have developed a system able to detect all concepts listed in the official glossary of the Special Report Global Warming of 1.5ºC (https://www.ipcc.ch/sr15/chapter/glossary/), which offer a much more representative outline of the topics occurring in IPCC reports than hashtags (which already reflects biases). The distribution of these concepts will first be compared across the whole IPCC 1.5 °C report, the 1.5 °C report’s “Summary for Policymakers”, and the subreddits threads related to climate issues. In a second step, we will automatically cluster these concepts into larger themes, based on the Doc2Vec similarity methods (Le et Mikolov, 2014) and compare their distribution on the three above corpora. Finally, a sample of threads will be manually analyzed in order to shed light on the results obtained by the automatic approach.