Attention, sentiments and emotions towards emerging climate technologies on Twitter

Highlights • We analyze and compare tweets on geoengineering and 16 related climate technologies.• Attention has shifted from general geoengineering to specific carbon removal methods.• Sentiments are more positive for carbon removal than solar radiation management.• Methods perceived closer to nature have the highest shares of positive sentiments.• Our social media analysis is consistent with survey results and qualitative research.


Conspiracy-related
Figure S8: Share of sentiments and emotions and share of conspiracy-related tweets for each technology category by user.To compute this metric, we first took the average sentiment of all tweets by each user and then averaged over all users.This is equivalent to a weighted average over tweets with a weight of one over the number of tweets by the user posting the tweet.

S2 List of subqueries for data gathering
The following list of keyword searches (Table S2) is based on queries developed for identifying scientific literature on carbon dioxide removal techniques Minx et al., 2018.Furthermore, we use other reviews on the topics of geoengineering, solar radiation management and greenhouse gas removal to extend the list of relevant keywords Cummings et al., 2017;Vaughan and Lenton, 2011;Caldeira et al., 2013.These keyword collections were adapted to return high shares of tweets relevant to the covered topics.We iteratively screened random samples of the results for each subquery to ensure that tweets are relevant to the covered topics.During this process, we dropped some subqueries that yielded mostly irrelevant results or modified the queries to include climate-related or exclude keywords not related to climate.
During the refinement of queries, we made an effort to make sure that we do not miss large parts of tweets on geoengineering-related topics.This is why we extended some of subqueries for techniques that initially returned only few results, which resulted in some imbalances in the number of subqueries per technique.97521 As CCS is also an acronym for other things, the query needed to be specified to also include climate related terms.

S3 Filter for conspiracy-related tweets
We mark a tweet as conspiracy-related if either one of the strong keywords is matched in the text or two of the weak keywords.

Figure S1 :
Figure S1: Map of tweets using a two-dimensional projection of embedded tweets, colored by technology category.Automatically generated labels indicate regions of high concentrations of tweets from technology categories.

Figure S2 :
Figure S2: Comparison matrix between labels from the two sentiment classifiers (left panel) and the two best performing emotion classifiers (left panel).Note that we mapped emotions from the scheme of each classifier to {positive, neutral, negative} to make the very different schemes comparable.

Figure S4 :
Figure S4: Histogram showing the number of users that posted a certain number of tweets.

Figure S5 :
Figure S5: Number of English-language tweets over time.The numbers have been estimated by searching for the 100 most common English words in tweets via the Twitter count API (https://api.twitter.com/2/tweets/counts/all).

Table S1 :
Average number of retweets, replies and likes per user.
Comparison of labels from different sentiment and emotion classifiers.Note that we mapped emotions from the scheme of each classifier to {positive, neutral, nega-tive} to make the very different schemes comparable.

Table S2 :
Table of subqueries for data gathering, count of unique tweets and explanation.