TamilMemes: Troll Classification Dataset for Tamil-Language Memes

TamilMemes is a dataset for identifying offensive or trolling content in region-specific Tamil-language memes. The resource, created by researchers at the University of Galway and presented at WILDRE-5 in 2020, contains meme images categorized as 'troll' or 'not_troll' and includes Latin-transcribed text from the memes. A shared task evaluation found the top-performing system achieved a weighted average F1-score of 0.55.

Use Cases

• Training troll meme classifiers based on annotated 'troll'/'not_troll' labels.
• Developing multimodal models that combine image and Latin-transcribed text features from memes.
• Benchmarking systems for region-specific meme analysis in Tamil.
• Studying the intersection of humor, sarcasm, and offensive content in social media imagery.

Strengths

• Provides a dedicated resource for the computationally challenging task of troll meme classification in Tamil.
• Includes both meme images and corresponding Latin-transcribed text, enabling multimodal analysis.
• Dataset was used in a shared task with 10 system submissions, providing a benchmark context.

Limitations

• Row count, file formats, and column-level documentation are unknown, limiting suitability assessment.
• The top system performance reported an F1-score of 0.55, suggesting the classification task is difficult with current methods.
• Last update date is unknown; freshness unverified.

Provenance

Source

Ollscoil na Gaillimhe – University of Galway

Collection Method

Manually annotated and released by researchers; created for a shared task on troll meme classification.

Time Range

Dataset presented in 2020.

Freshness

Last updated date is unknown.

Geography

Focus on Tamil language, suggesting relevance to Tamil-speaking communities and regions.