Using LLMs to Analyze People’s Views On Climate Change In Social Networks

open_access
Open Access
Author:
Wu, Lok Yiu
Area of Honors:
Computer Science
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Wenpeng Yin, Thesis Supervisor
Martin Fürer, Thesis Honors Advisor
Keywords:
Computer Science
Climate Change
Abstract:
This thesis examines how climate change is discussed, opposed, and supported publicly on Twitter through structured tweet metadata and advanced computational techniques. Built upon a dataset of over 15 million English-language tweets from 2006 to 2019, the study highlights regional stance differences, temporal variations in public sentiment, and the role of topical framing, without direct access to tweet text. Leveraging Large Language Models (LLMs), particularly transformer-based architectures, the research provides interpretive insights into emotional tone, stance, and core arguments associated with specific climate topics. The results imply that environmental crises and major policy announcements serve as a stimulus for sudden shifts in sentiment, while regions directly affected by climate extremes often exhibit higher rates of belief in human-driven climate change. Importantly, lack of tweet text did not exclude meaningful analysis, rather structured metadata, e.g., stance labels, sentiment scores, and geolocation, enabled the deployment of an LLM-based framework for summarizing and linking large-scale discourse constructs. This work offers a new approach to combining AI-based qualitative reasoning with statistical analysis for climate communication studies. Future extensions could include cross-lingual analyses, real-time event mapping, and deeper linguistic processing once direct tweet content or external datasets become accessible.