Automated detection and analysis of ad-hominem attacks in climate misinformation
This project was a successful recipient of CAIDE's 2023 seed funding round 'Automated Expertise.'
Overview
This project will develop novel natural language processing (NLP) methods to model ad-hominem attacks on experts—a rhetorical technique which has been shown to dominate climate misinformation. [1]
Our aims are:
- (i) Analyse ad-hominems with NLP models in more detail and at a larger scale than previous work
- (ii) Explore how their exposure can combat misinformation by raising readers’ awareness and critical evaluation of content
The theoretical part of this project will expand on CIs Cook and Samoilenko’s ad-hominem taxonomy and annotation scheme [1] and ad-hominem attack types [4]. Methodologically, we extend CI Frermann’s models of narrative framing of climate change [3] and entity roles [2,3] to ad-hominem attacks. To train and evaluate models for aim (i), we will consolidate and extend two annotated data sets: adhominem attacks in contrarian blogs [1,4], and climate news labeled with entities' roles (eg. victim, villain), types (eg. scientists, politicians), and framing (political conflict vs. individual impact) [3].
We will develop NLP models which, given a news article, identify an ad-hominem attack and highlight its victim, attack type, and framing. We will apply our models to larger data sets to study changes of attacks and victims over time. To address aim (ii), we will run a user study, exposing participants to ad-hominem attacks exposed in varying detail, and measure user trust in model predictions, and change in sensitivity to ad-hominem based misinformation.
[1] Coan, Boussalis, Cook, Nanko, (2021). Computer-assisted detection and classification of misinformation about climate change. Scientific Reports, 11(22320).
[2] Khanezar, Mikolajczak, Cohn, Frermann (2023). Probing Power by Prompting: Harnessing Pre-trained Language Models for Power Connotation Framing. In: EACL 2023.
[3] Frermann, Li, Khanehzar, Mikolajczak. Conflicts, Villains, Resolutions: Towards models of Narrative Media Framing. ACL 2023.
[4] Samoilenko, Cook (in revision). Content analysis and deconstruction of ad-hominem attacks on climate scientists. Global Environmental Change.
Research Team
- Dr Lea Frermann
School of Computing and Information Systems
Faculty of Engineering and IT
University of Melbourne
- Dr John Cook
Melbourne School of Psychological Sciences
Faculty of Medicine, Dentistry, and Health Sciences
University of Melbourne
- Assistant Professor Sergei Samoilenko
College of Humanities and Social Sciences
George Mason University