Understanding Opinions Towards Climate Change on Social Media

Yashaswi Pupneja, Joseph Zou, Sacha Lévy, Shenyang HuangarXiv 2023 · OpenReview

TL;DR

This paper investigates how real-world events, specifically UN Climate Change Conferences (COP), influence public opinion and community structures on climate change discourse on social media. By analyzing a massive Twitter dataset, the researchers use temporal graph analysis and NLP techniques to map evolving online discussions and identify shifts in sentiment and topic focus around these key global events.

What Problem Does This Paper Address?

Climate change is a critical global issue, yet public opinion is often polarized and influenced by misinformation, particularly on social media. Understanding how major events impact these online discussions is crucial for developing effective communication strategies to foster awareness and action. This paper aims to bridge the gap between real-world climate events and the dynamic, often fragmented, landscape of social media conversations, seeking to understand how these events shape online communities and their opinions.

Methods

The researchers employ a multi-faceted approach:

  • Data Collection: They leverage a pre-existing Kaggle dataset of climate change-related tweets and augment it by re-collecting the raw tweet text and associated metadata (sender, mentions, timestamp, hashtags) using the Twitter API for Academic Research.
  • Temporal Graph Construction: A user-user mention network is built, representing interactions over time. This network is discretized into weekly snapshots, forming a temporal graph. To focus on active users and create a denser network, users with fewer than 100 edges are filtered out.
  • Community Detection: The Louvain algorithm, a standard method for detecting communities in networks, is applied to the temporal graph snapshots. This helps identify evolving groups of users and how their structure changes around COP events.
  • Sentiment Analysis: The TextBlob library is used to determine the emotional tone (positive, negative, neutral) of individual tweets, providing insights into the overall sentiment surrounding climate change discussions.
  • Topic Modeling: A mini-batch Non-negative Matrix Factorization (NMF) algorithm is used to identify latent topics within the tweet corpus, revealing the key themes and issues being discussed.

Datasets

The primary dataset consists of 13.6 million tweets from 3.6 million users, collected from 2006 to 2019. This dataset was initially sourced from a Kaggle dataset and then enriched with raw tweet text and metadata via the Twitter API. The study also references dates for relevant Conference of the Parties (COP) events within this timeframe.

Key Insights and Ideas

  • COP Events as Discussion Catalysts: The study observes that COP events often correlate with local peaks in the number of edges (user interactions) in the temporal graph, suggesting increased discussion activity around these conferences.
  • Shifting Community Structures: While the number of communities detected by the Louvain algorithm shows varied responses to COP events (sometimes increasing, sometimes decreasing), the researchers suggest this indicates changes in community formation, potentially leading to larger, more cohesive groups during periods of heightened discussion.
  • Dominance of Neutral Sentiment: Sentiment analysis reveals that a majority of tweets are neutral, with a significant portion being negative. This highlights a need for more engaging content to foster positive public involvement.
  • Emerging Topics: Topic modeling identifies key themes such as "politics" (related to policy and action), "CO2 emission" (global warming, melting ice, sea level rise), and "weather" (concerns about extreme events), indicating public focus on both the causes and consequences of climate change.

Key Results

  • The analysis of 13.6 million tweets from 2006-2019 reveals that COP events are associated with increased user interactions on Twitter.
  • Sentiment analysis shows that while most tweets are neutral, 19.0% have a negative sentiment, suggesting a significant level of concern or discontent.
  • Topic modeling successfully extracts distinct themes related to climate change, including political responses, the scientific basis of emissions, and observable weather impacts.
  • Community detection around COP events shows fluctuations in the number of communities, with some events leading to a decrease (suggesting larger, more consolidated communities) and others an increase.

Limitations and Future Work

Limitations:

  • Twitter Representativeness: The authors acknowledge that Twitter users may not be representative of the broader population, and discussions on Twitter might not reflect the most salient public concerns.
  • Data Quality: Slang, sarcasm, and the nuances of language can pose challenges for NLP analysis, potentially introducing noise and bias.
  • Community Detection Focus: The Louvain algorithm, used for community detection, relies solely on graph structure and does not incorporate tweet text or other edge features, which could limit its ability to fully capture opinion-based communities.

Future Work:

  • The authors suggest exploring demographic differences in how climate change is perceived and discussed on social media.
  • They also propose tracking shifts in public opinion over time more granularly.
  • Future research could extend this work by applying more recent techniques such as Large Language Models (LLMs) and Graph Neural Networks (GNNs) for deeper analysis.

Summary updated

Abstract

Social media platforms such as Twitter (now known as X) have revolutionized how the public engage with important societal and political topics. Recently, climate change discussions on social media became a catalyst for political polarization and the spreading of misinformation. In this work, we aim to understand how real world events influence the opinions of individuals towards climate change related topics on social media. To this end, we extracted and analyzed a dataset of 13.6 millions tweets sent by 3.6 million users from 2006 to 2019. Then, we construct a temporal graph from the user-user mentions network and utilize the Louvain community detection algorithm to analyze the changes in community structure around Conference of the Parties on Climate Change~(COP) events. Next, we also apply tools from the Natural Language Processing literature to perform sentiment analysis and topic modeling on the tweets. Our work acts as a first step towards understanding the evolution of pro-climate change communities around COP events. Answering these questions helps us understand how to raise people's awareness towards climate change thus hopefully calling on more individuals to join the collaborative effort in slowing down climate change.
BibTeX
@misc{pupneja2023understanding,
  author = {Yashaswi Pupneja and Joseph Zou and Sacha Lévy and Shenyang Huang},
  title = {{Understanding Opinions Towards Climate Change on Social Media}},
  year = {2023},
  eprint = {2312.01217},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2312.01217},
}
Figure