| --- |
| license: apache-2.0 |
| language: |
| - en |
| - hi |
| - ta |
| - te |
| - bn |
| - kn |
| size_categories: |
| - 100K<n<1M |
| task_categories: |
| - token-classification |
| - text-generation |
| tags: |
| - finance |
| - banking |
| - entity-extraction |
| - indian-banking |
| - sms |
| - synthetic |
| pretty_name: FinEE Dataset - Indian Financial Entity Extraction |
| --- |
| |
| # FinEE Dataset |
|
|
| <p align="center"> |
| <img src="https://img.shields.io/badge/Samples-152K%2B-blue" alt="Samples"> |
| <img src="https://img.shields.io/badge/Languages-6-orange" alt="Languages"> |
| <img src="https://img.shields.io/badge/Banks-25%2B-green" alt="Banks"> |
| </p> |
|
|
| ## Dataset Description |
|
|
| A comprehensive dataset for training financial entity extraction models on Indian banking messages. Contains 152,000+ samples covering SMS, emails, and transaction notifications from major Indian banks. |
|
|
| ### Languages |
|
|
| - English (en) - 86% |
| - Hindi (hi) - 3% |
| - Tamil (ta) - 3% |
| - Telugu (te) - 3% |
| - Bengali (bn) - 3% |
| - Kannada (kn) - 2% |
|
|
| ### Supported Transaction Types |
|
|
| - UPI payments (PhonePe, GPay, Paytm) |
| - NEFT/IMPS/RTGS transfers |
| - Credit card transactions |
| - Debit card transactions |
| - EMI payments |
| - Refunds and reversals |
| - Salary credits |
| - Bill payments |
|
|
| ### Covered Banks |
|
|
| HDFC, ICICI, SBI, Axis, Kotak, PNB, BOB, Canara, Union, IDBI, IndusInd, Yes Bank, Federal, South Indian, Karur Vysya, and more. |
|
|
| ### Covered Merchants |
|
|
| - Food: Swiggy, Zomato, Zepto, BigBasket |
| - Shopping: Amazon, Flipkart, Myntra, Meesho |
| - Travel: Uber, Ola, IRCTC, MakeMyTrip |
| - Investment: Zerodha, Groww, Upstox, Angel One |
| - Bills: Airtel, Jio, electricity, gas |
| - Entertainment: Netflix, BookMyShow, Hotstar |
|
|
| ## Dataset Structure |
|
|
| ### Data Fields |
|
|
| ```json |
| { |
| "input": "HDFC Bank: Rs.2,500 debited from A/c XX1234...", |
| "output": { |
| "amount": 2500.0, |
| "type": "debit", |
| "account": "1234", |
| "bank": "HDFC", |
| "merchant": "Swiggy", |
| "category": "food", |
| "is_p2m": true |
| } |
| } |
| ``` |
|
|
| ### Instruction Format (ChatML) |
|
|
| ```json |
| { |
| "messages": [ |
| {"role": "system", "content": "You are a financial entity extraction assistant..."}, |
| {"role": "user", "content": "Extract financial entities from: ..."}, |
| {"role": "assistant", "content": "{\"amount\": 2500.0, ...}"} |
| ] |
| } |
| ``` |
|
|
| ### Splits |
|
|
| | Split | Samples | Description | |
| |-------|---------|-------------| |
| | train | 137,267 | Training data | |
| | valid | 7,625 | Validation data | |
| | test | 7,627 | Test data (held out) | |
|
|
| ## Data Sources |
|
|
| 1. **Real Data** (2,419 samples) |
| - Anonymized ICICI Bank SMS messages |
| - Manually verified labels |
|
|
| 2. **Synthetic Data** (100,000 samples) |
| - Grammar-based generation |
| - Covers all bank templates |
| - Realistic amount distributions |
|
|
| 3. **Multilingual Synthetic** (50,100 samples) |
| - Hindi, Tamil, Telugu, Bengali, Kannada |
| - Markov chain for realistic flow |
| - Edge case oversampling |
|
|
| ## Usage |
|
|
| ### Load with Datasets |
|
|
| ```python |
| from datasets import load_dataset |
| |
| dataset = load_dataset("Ranjit0034/finee-dataset") |
| |
| # Access splits |
| train = dataset["train"] |
| valid = dataset["valid"] |
| test = dataset["test"] |
| |
| # Iterate |
| for example in train: |
| print(example["input"]) |
| print(example["output"]) |
| ``` |
|
|
| ### Load for Fine-tuning |
|
|
| ```python |
| from datasets import load_dataset |
| |
| # Load instruction format |
| dataset = load_dataset("Ranjit0034/finee-dataset", data_files={ |
| "train": "instruction/train.jsonl", |
| "valid": "instruction/valid.jsonl" |
| }) |
| ``` |
|
|
| ## Output Schema |
|
|
| | Field | Type | Description | |
| |-------|------|-------------| |
| | amount | float | Transaction amount in INR | |
| | type | string | "debit" or "credit" | |
| | account | string | Last 4 digits of account | |
| | bank | string | Bank name | |
| | date | string | Transaction date (YYYY-MM-DD) | |
| | time | string | Transaction time (HH:MM) | |
| | reference | string | UPI/NEFT reference number | |
| | merchant | string | Merchant name (P2M) | |
| | beneficiary | string | Person name (P2P) | |
| | vpa | string | UPI VPA address | |
| | category | string | Transaction category | |
| | is_p2m | boolean | true if merchant, false if P2P | |
| | balance | float | Balance after transaction | |
| | status | string | success/failed/pending | |
| |
| ## Categories |
| |
| - `food` - Restaurants, delivery |
| - `grocery` - Supermarkets |
| - `shopping` - E-commerce, retail |
| - `transport` - Cab, fuel |
| - `travel` - Flights, hotels |
| - `bills` - Utilities, recharge |
| - `entertainment` - Movies, streaming |
| - `healthcare` - Medical, pharmacy |
| - `investment` - Stocks, mutual funds |
| - `transfer` - P2P transfers |
| - `salary` - Income |
| - `emi` - Loan payments |
| |
| ## Citation |
| |
| ```bibtex |
| @dataset{finee_dataset, |
| title={FinEE Dataset: Indian Financial Entity Extraction}, |
| author={Ranjit Behera}, |
| year={2026}, |
| url={https://huggingface.co/datasets/Ranjit0034/finee-dataset} |
| } |
| ``` |
| |
| ## License |
| |
| Apache 2.0 |
| |
| ## Related |
| |
| - 🤖 [FinEE Llama 8B](https://huggingface.co/Ranjit0034/finee-llama-8b) - Fine-tuned model |
| - 📦 [FinEE Package](https://pypi.org/project/finee/) - Python library |
| - 💻 [GitHub](https://github.com/Ranjitbehera0034/Finance-Entity-Extractor) |
| |
