BDEmo-CodeMiX a Bengali Code-Mix Dataset for Emotion Analysis
Description
The BDEmo-CodeMix dataset is a manually curated, raw multilingual code-mixed corpus to advance emotion analysis in the relatively unexplored field of Bangla Natural Language Processing (BNLP). Consisting of 17,483 lines, 11,807 paragraphs, and 156,387 words distributed across .jsonl files, the dataset captures positive and negative linguistic expressions from YouTube viewer comments where users constantly transition between Bengali and English. The data is structurally preprocessed by mapping unique words to a vocabulary index and applying sequence padding alongside special token insertions—specifically the classification token [CLS] at the beginning and the separator token [SEP] at the end—to prepare the code-mixed text for machine learning models. Ultimately, the dataset maps binary emotional dispositions, labeling negative sentiment as 0 and positive sentiment as 1.
Files
Institutions
- American International University-BangladeshDhaka Division, Dhaka