The dataset viewer is not available for this split.
Error code: TooLongColumnNameError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
BanglaDeshi Multipurpose Corpus (BMP)
Dataset Description
The BanglaDeshi Multipurpose Corpus (BMP) is a comprehensive, multi-domain Bengali text corpus designed for NLP research and applications. It contains 94+ million rows across 77 CSV files (~16 GB), covering diverse topics from news and entertainment to science, law, and encyclopedic knowledge.
Dataset Summary
| Attribute | Value |
|---|---|
| Language | Bengali (Bangla) + English secondary |
| Total Rows | 94,215,876 |
| Total Size | ~16 GB |
| Files | 77 CSV files |
| Categories | 28 (C01-C62) + C99 |
| License | CC BY-SA 4.0 |
| Version | 1.0 |
Categories
News & Media (C01-C04)
- C01: News articles (National, Local, International, Economy, Politics)
- C02: Entertainment (Bangla Cinema, Music, Showbiz)
- C03: Sports (Cricket, Football, General)
- C04: Opinion & Editorial
Knowledge & Education (C13-C15, C20-C21, C60-C61)
- C13: Education (Universities, Academic)
- C15: Science & Technology
- C20: Knowledge QA (30M+ rows - largest category)
- C21: NLU Tasks (Natural Language Inference)
- C60: Encyclopedia (Wikipedia articles)
- C61: Textbooks (NCTB)
Language Resources (C40-C44)
- C40: Linguistic Corpus (News)
- C41: Conversational AI (Empathetic Dialogues)
- C42: Translation (English-Bengali QA)
- C43: Lexical Resources (Dictionary + 20,719 Idioms)
- C44: Reference Data (Wikidata Facts)
Specialized Data (C50-C51)
- C50: Math Problems (45M+ rows - chain-of-thought)
- C51: Instruction Following (Alpaca-style)
Domain-Specific (C05-C17, C30, C62)
- C05: Health & Wellness
- C06: Literature (Book Reviews)
- C10: Politics
- C11: Crime & Law
- C12: Economy
- C14: Religion
- C16: Environment
- C17: Culture & Society
- C30: Sentiment Analysis
- C62: Legal Documents
Supplementary (C99)
- C99: Scraped data, test enrichment, Wikipedia facts
Key Features
- Multi-domain: Covers 28+ categories from news to science
- Rich Idiom Database: 20,719 Bengali idioms with frequency, sentiment, and cultural metadata
- Massive Math Dataset: 45M+ chain-of-thought math problems
- Knowledge QA: 30M+ question-answer pairs
- Research-ready: Pre-processed CSV format
Use Cases
- Hallucination Detection: RAG retrieval for Bengali LLM verification
- Question Answering: Knowledge-grounded QA systems
- Math Problem Solving: Bengali math reasoning
- Text Classification: NLI, sentiment analysis
- Dictionary/Idiom Lookup: Lexical resources
- Machine Translation: Bengali-English pairs
Data Format
All files are in CSV format with UTF-8 encoding. Column structures vary by category:
| Category | Key Columns |
|---|---|
| C20 (Knowledge QA) | context, prompt_bn, response_bn |
| C43 (Idioms) | idiom, figurative_meaning_bn, frequency, sentiment |
| C50 (Math) | question, solution, chain_of_thought |
| C60 (Encyclopedia) | page_title, content |
Citation
ibtex @dataset{bmp2024, title={BanglaDeshi Multipurpose Corpus}, version={1.0}, author={Shayon Sarker}, license={CC-BY-SA-4.0}, url={https://huggingface.co/datasets/ShayonSarker/Bangladeshi-Multipurpose-Dataset} }
Contact
For questions about this dataset, please open an issue on the GitHub repository.
Acknowledgments
- Built for the Bengali LLM Hallucination Detection Challenge
- Data sourced from news, Wikipedia, dictionaries, and educational materials
- Downloads last month
- 90