AUST-Sarc (Emoji-Aware Bangla Sarcasm & Offensiveness)

AUST-SARC is a Bangla text dataset for sarcasm and offensiveness detection with explicit emoji context.
Each entry is a short user-generated sentence containing ≥1 emoji, labeled with Sarcasm (1/0) and Offensive (1/0).

Size: 2,649 sentences

Sarcasm distribution: 1 = 1,496, 0 = 1,153

Offensive distribution: 1 = [fill], 0 = [fill]

Language: Bangla (with natural Bangla–English code-mixing)

Preprocessing: emojis preserved; light normalization of URLs/mentions/hashtags; whitespace cleanup.

Intended use: sarcasm detection, offensive/abusive language detection, emoji-aware pragmatics, multitask learning.
Ethics: personally identifying info removed; texts may contain sensitive or offensive content due to the task.