Bengali & Banglish: A Monolingual Dataset for Emotion Detection in Linguistically Diverse Contexts
Description
This dataset, positioned at the intersection of Bengali and Banglish (an English-character variant of Bengali), is a valuable resource for emotion detection. It encompasses a total of 80,098 data entries, comprising both languages. The dataset is organized into six distinct emotional categories: anger (15,179), disgust (13,098), fear (7,565), joy (17,836), sadness (16,309), and surprise (10,107), aligning with Ekman's six basic emotions framework. Sourced from platforms such as EmoNoBa, UBMEC, MONOVAB, and comments from YouTube and Twitter posts, it offers a diverse and rich dataset for research and analysis. Moreover, given its bilingual nature, this data also holds relevance for neural machine translation tasks.
Files
Steps to reproduce
The dataset was compiled from EmoNoBa, UBMEC, and MONOVAB and enriched with YouTube and Twitter comments across eight Bangladesh-specific domains via official APIs. The original datasets were annotated for YouTube and Twitter data through majority voting, while the original datasets were pre-annotated. After annotation and duplicate removal, the dataset was translated into Banglish, an English-character variation of Bengali.