BESCAD: A Multi-Label Bangla Emotion and Social Context Annotation Dataset

BESCAD (Bangla Emotion and Social Context Annotation Dataset) is a large-scale multi-label dataset of 17,588 Bangla social media and literary text samples annotated with 15 fine-grained emotion categories and 3 social-context labels (Sarcasm, Mixed Language, Regionalism). Each sample includes original Bangla text, Banglish transliteration, and human-verified English translation, enabling monolingual and cross-lingual research. The dataset was created through a human-first annotation pipeline with LLM-assisted refinement and expert validation, making it suitable for affective computing, NLP, and multilingual emotion analysis.