Reproduce “Make Satire Boring Again: Reducing Stylistic Bias of Satirical Corpus by Utilizing Generative LLMs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Asli Umay Ozturk, Recep Firat Cekinel, Pinar Karagoz
Researchers from Middle East Technical University propose using generative Large Language Models to reduce stylistic bias in satirical corpora. This approach, implemented through a debiasing pipeline, creates less stylistically marked versions of satirical texts, leading to improved cross-domain and cross-lingual generalizability of satire detection models, particularly for Turkish.
Key insights
- Stylistic bias in satirical datasets limits the generalizability of satire detection models across different domains and languages.
- Generative LLMs can be effectively used to reduce stylistic markers in satirical texts while preserving their satirical nature.
- Models trained on debiased or hybrid datasets improve cross-domain and cross-lingual satire detection, even if same-domain performance sees a slight decrease.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.