Reproduce “A Group-Specific Approach to NLP for Hate Speech Detection” 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
Karina Halevy
The authors propose group-specific hate-speech detection by attaching historical and linguistic context from a reusable knowledge base before classification. KnowledJe matches tweet phrases to an antisemitism-specific graph. On Echo Corpus, mean F1 rises from 0.321 to 0.336 for Logistic Regression but falls from 0.680 to 0.657 for DistilBERT. This preliminary result from an English-language echo-associated dataset does not establish general benefit.
Key insights
- Historical and linguistic context can provide useful signals for detecting explicit and coded antisemitic language, but exact n-gram matching limits coverage of paraphrases, novel code words, and implicit references.
- Knowledge infusion helped the simpler Logistic Regression classifier but did not consistently improve DistilBERT, indicating that augmentation quality and model architecture strongly affect outcomes.
- Detection and bias auditing are complementary: a model must identify antisemitic content without learning or amplifying stereotypes such as conspiracy beliefs, dual loyalty, greed, or Holocaust denial.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.