Lannisters@DravidianLangTech 2026: A Comparative and Ablation Study of Multilingual Transformers for Gender-Targeted Abuse Detection in Tamil Social Media Platforms

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, quick

Summary

The "Lannisters@DravidianLangTech 2026" study addresses the critical need for automated detection of gender-targeted abuse in Tamil social media, aiming to foster safer online communication. Researchers developed a binary classification model to identify abusive and non-abusive content, experimenting with several multilingual transformer models. Among DistilBERT, mBERT, and XLM-RoBERTa, the XLM-RoBERTa model demonstrated superior performance, achieving an accuracy of 91.17% and a macro F1 score of 0.8865. Ablation experiments further revealed that structured preprocessing, effective balancing of the minority class, and meticulous hyperparameter tuning are crucial factors contributing significantly to the model's overall performance in this specific task.

Key takeaway

For NLP Engineers developing abuse detection systems for Dravidian languages like Tamil, this research highlights XLM-RoBERTa as a strong baseline. You should prioritize structured preprocessing and minority class balancing to significantly boost model accuracy and F1 scores. Consider fine-tuning hyperparameters rigorously, as these steps are proven to enhance performance for gender-targeted abuse identification in social media contexts.

Key insights

Multilingual transformers, particularly XLM-RoBERTa, effectively detect gender-targeted abuse in Tamil social media.

Principles

• Preprocessing enhances model performance.
• Class balancing improves minority detection.
• Hyperparameter tuning is critical.

Method

A binary classification model identifies abusive content using multilingual transformers, optimized via structured preprocessing, minority class balancing, and hyperparameter tuning.

In practice

• Implement XLM-RoBERTa for Tamil abuse detection.
• Apply structured text preprocessing.
• Balance datasets with minority classes.

Topics

• Gender-Targeted Abuse Detection
• Tamil Language Processing
• Multilingual Transformers
• XLM-RoBERTa
• Social Media Analysis
• Binary Classification

Best for: Research Scientist, NLP Engineer, AI Scientist, Machine Learning Engineer

Related on AIssential

• mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection — Finetuning mid-sized LLMs with QLoRA and augmented data improves multilingual online polarization detection.
• MORPHOGEN: A Multilingual Benchmark for Evaluating Gender-Aware Morphological Generation — MORPHOGEN evaluates LLM gender-aware morphological generation across French, Arabic, and Hindi, revealing performance gaps and biases.
• RacismoBR: A Manually Annotated Dataset for Racist Discourse Detection in Brazilian Portuguese — Culturally grounded annotation is more critical than model architecture for effective racism detection.
• Multi-Stage Training for Abusive Comment Detection in Indic Languages — Multi-stage training with language preprocessing and model ensembles improves abusive comment detection while minimizing false positives.
• REDACT: A Systematically Controlled Multilingual Benchmark for Personal Information Detection — REDACT offers a controlled multilingual PII benchmark revealing LLMs outperform rule-based systems on high-sensitivity data.
• Cross-lingual Relation Extraction with Large Language Models: Zero-Shot, Few-Shot, and Fine-Tuned Evaluation on Romanian — QLoRA fine-tuning significantly boosts cross-lingual relation extraction performance for low-resource languages, outperforming prompt-only LLMs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.