Research map / Large language models / Language, safety and interpretability
Multilingual Abusive Comment Detection at Scale for Indic Languages
Vikram Gupta, Sumegh Roychowdhury, Mithun Das, Somnath Banerjee, Punyajoy Saha, Binny Mathew, hastagiri prakash vanchinathan, Animesh Mukherjee
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Pretrain XLM-R on 5M social media comments (AbuseXLMR vs XLM-R) | XLM-R · MACD Hindi monolingual | F1-macro: 86.11 -> 87.93 | Improved |
| Pretrain XLM-R on social media comments (AbuseXLMR vs XLM-R) | XLM-R · MACD Telugu monolingual | F1-macro: 89.44 -> 91.37 | Improved |
| Unbalanced split (1:5 abusive:non-abusive) instead of random | AbuseXLMR · MACD Hindi | F1-macro: 88.52 -> 85.69 | Hurt |
| Unbalanced split (1:5) instead of random | AbuseXLMR · MACD Hindi | accuracy: 88.53 -> 93.05 | Improved |
| Chronological split instead of random | AbuseXLMR · MACD Malayalam | F1-macro: 88.14 -> 81.82 | Hurt |
| Zero-shot cross-lingual transfer Hindi to Tamil instead of in-language training | AbuseXLMR · MACD Tamil | F1-macro: 88.60 -> 86.70 | Hurt |
| Zero-shot cross-lingual transfer Hindi to Tamil instead of in-language training | XLM-R · MACD Tamil | F1-macro: 87.87 -> 71.82 | Hurt |
| Few-shot with 100 samples: AbuseXLMR vs XLM-R | XLM-R · MACD Hindi | F1-macro: 48.49 -> 77.05 | Improved |
| Few-shot 500 samples vs 5 samples | AbuseXLMR · MACD Telugu | F1-macro: 42.15 -> 88.71 | Improved |
| Joint training on all languages instead of monolingual (ablation) | mBERT · MACD Hindi | F1-macro: 84.31 -> 85.44 | Improved |
| Joint training on all languages instead of monolingual (ablation) | XLM-R · MACD Hindi | F1-macro: 86.11 -> 85.26 | Hurt |
| Joint training on all languages instead of monolingual (ablation) | XLM-R · MACD Malayalam | F1-macro: 85.42 -> 85.87 | Improved |
Closest papers by meaning
- A Multi-Task Benchmark for Abusive Language Detection in Low-Resource Settings (NeurIPS 2025 · similarity 0.94)
- Spanning the Spectrum of Hatred Detection: A Persian Multi-Label Hate Speech Dataset with Annotator Rationales (AAAI 2024 · similarity 0.87)
- Hateful Meme Detection through Context-Sensitive Prompting and Fine-Grained Labeling (Student Abstract) (AAAI 2025 · similarity 0.84)
- Efficient Toxic Content Detection by Bootstrapping and Distilling Large Language Models (AAAI 2024 · similarity 0.84)
- Multi-Modal Discussion Transformer: Integrating Text, Images and Graph Transformers to Detect Hate Speech on Social Media (AAAI 2024 · similarity 0.84)
- TRACE: Textual Relevance Augmentation and Contextual Encoding for Multimodal Hate Detection (AAAI 2026 · similarity 0.83)
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.
