AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Aligning Attention with Human Rationales for Self-Explaining Hate Speech Detection

Brage Eilertsen, Røskva Bjørgfinsdóttir, Francielle Vargas, Ali Ramezani-Kebrya

AAAI 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse and structured attention · 4 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add attention alignment supervision (alpha 0 to 100) (ablation)BERT-base-uncased · HateXplain testIoU F1: 0.019 -> 0.572Improved
add attention alignment supervision (alpha 0 to 100) (ablation)BERT-base-uncased · HateXplain testToken F1: 0.122 -> 0.678Improved
increase alpha from 0 to 100 (ablation)BERT-base-uncased · HateXplain testMacro F1No clear effect
increase alpha from 0 to 100 (ablation)BERT-base-uncased · HateXplain testGMB-Subgroup AUC: range 0.7112-0.7220No clear effect
add attention alignment supervision (alpha 0.05 to 100) (ablation)BERTimbau-base · HateBRXplain testIoU F1: 0.387 -> 0.751; baseline 0Improved
add attention alignment supervision (ablation)BERTimbau-base · HateBRXplain testMacro F1: 0.903 -> 0.921 (alpha=0.5)No clear effect
add attention alignment supervision (ablation)BERTimbau-base · HateBRXplain testGMB-Subgroup AUC: 0.7165 -> 0.7157No clear effect
add attention alignment supervision (ablation)BERT-base-uncased · HateXplain testattention-rationale Pearson correlation: -0.084 -> 0.649Improved
add attention alignment supervision on identity terms (ablation)BERT-base-uncased · HateXplain testtoken precision (identity terms): 0.265 -> 0.938Improved
use SRA instead of BERT-MRPBERT-base-uncased · HateXplain testMacro F1: 0.699 -> 0.682Hurt
use SRA instead of BERT-MRPBERT-base-uncased · HateXplain testGMB-Subgroup AUC: 0.815 -> 0.714Hurt
use SRA instead of BERT-HateXplain supervised attentionBERT-base-uncased · HateXplain testIoU F1: 0.120 -> 0.539Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy