DoDo Learning: Domain-Demographic Transfer in Language Models for Detecting Abuse Targeted at Public Figures

Angus Redlarski Williams, Hannah Rose Kirk, Liam Burke-Moore, Yi-Ling Chung, Ivan Debono, Pica Johansson, Francesca Stevens, Jonathan Bright, Scott Hale


Abstract
Public figures receive disproportionate levels of abuse on social media, impacting their active participation in public life. Automated systems can identify abuse at scale but labelling training data is expensive and potentially harmful. So, it is desirable that systems are efficient and generalisable, handling shared and specific aspects of abuse. We explore the dynamics of cross-group text classification in order to understand how well models trained on one domain or demographic can transfer to others, with a view to building more generalisable abuse classifiers. We fine-tune language models to classify tweets targeted at public figures using our novel DoDo dataset, containing 28,000 entries with fine-grained labels, split equally across four Domain-Demographic pairs (male and female footballers and politicians). We find that (i) small amounts of diverse data are hugely beneficial to generalisation and adaptation; (ii) models transfer more easily across demographics but cross-domain models are more generalisable; (iii) some groups contribute more to generalisability than others; and (iv) dataset similarity is a signal of transferability.
Anthology ID:
2024.trac-1.15
Volume:
Proceedings of the Fourth Workshop on Threat, Aggression & Cyberbullying @ LREC-COLING-2024
Month:
May
Year:
2024
Address:
Torino, Italia
Editors:
Ritesh Kumar, Atul Kr. Ojha, Shervin Malmasi, Bharathi Raja Chakravarthi, Bornini Lahiri, Siddharth Singh, Shyam Ratan
Venues:
TRAC | WS
SIG:
Publisher:
ELRA and ICCL
Note:
Pages:
134–154
Language:
URL:
https://preview.aclanthology.org/build-pipeline-with-new-library/2024.trac-1.15/
DOI:
Bibkey:
Cite (ACL):
Angus Redlarski Williams, Hannah Rose Kirk, Liam Burke-Moore, Yi-Ling Chung, Ivan Debono, Pica Johansson, Francesca Stevens, Jonathan Bright, and Scott Hale. 2024. DoDo Learning: Domain-Demographic Transfer in Language Models for Detecting Abuse Targeted at Public Figures. In Proceedings of the Fourth Workshop on Threat, Aggression & Cyberbullying @ LREC-COLING-2024, pages 134–154, Torino, Italia. ELRA and ICCL.
Cite (Informal):
DoDo Learning: Domain-Demographic Transfer in Language Models for Detecting Abuse Targeted at Public Figures (Williams et al., TRAC 2024)
Copy Citation:
PDF:
https://preview.aclanthology.org/build-pipeline-with-new-library/2024.trac-1.15.pdf
Optionalsupplementarymaterial:
 2024.trac-1.15.OptionalSupplementaryMaterial.zip