Twitter Sarcasm Detection Dataset from ACL Workshop

Shiv213's dataset provides labeled text data for training and testing sarcasm detection models on Twitter posts. The data originates from the 2nd FigLang Workshop Shared Task at ACL 2020. It includes JSONL files with labels for each sample.

Use Cases

• Train a binary text classifier to predict the 'label' field (SARCASM or NOT_SARCASM) from tweet content.
• Fine-tune transformer models like BERT for the specific task of sarcasm detection using the provided 'label' annotations.
• Analyze linguistic patterns and features distinguishing sarcastic 'SARCASM' tweets from literal 'NOT_SARCASM' ones.
• Benchmark model performance against the ACL 2020 FigLang Workshop Shared Task using the official training and test splits.

Strengths

• Data is curated for a peer-reviewed ACL workshop shared task, ensuring a defined benchmark standard.
• Provides a clear binary classification target with the 'label' field for model training.

Limitations

• The dataset size, row count, and sample demographics are unknown, making scale assessment impossible.
• Data is from 2020 or earlier, potentially not reflecting current Twitter language or platform changes.

Provenance

Source

2nd FigLang Workshop at ACL 2020 (Association for Computational Linguistics).

Collection Method

Collected for a shared task on automatic sarcasm detection, specific methodology not detailed.

Time Range

null

Freshness

Dataset page was last updated on 2024-05-06, but the underlying data is from the ACL 2020 workshop.

Geography

null