banking10 / README.md
timpearce's picture
Update README.md
9670482 verified
|
raw
history blame
3.15 kB

Bankin10

Banking10 Dataset

Overview

This dataset is a modified version of the original Banking77 dataset created by PolyAI. While the original dataset contains 77 fine-grained banking intents for customer service queries, this version provides a higher-level categorization by grouping these intents into 9 macro-categories.

Original Dataset

The Banking77 dataset was created to help train and evaluate banking customer service dialogue systems. It contains 13,083 customer service queries labelled with 77 different intents. The original dataset focuses on fine-grained intent detection with clusters of semantically similar intents but differ in their banking-specific resolution.

Modifications

We have regrouped the 77 original intents into 10 broader categories to facilitate higher-level categories classification:

CARD_MANAGEMENT

  • Card ordering, activation, and delivery
  • Lost, stolen, or malfunctioning cards
  • Card linking and support

CARD_TRANSACTIONS

  • Card payments
  • Transaction recognition and disputes
  • Payment status and declines
  • Exchange rate issues

CASH_OPERATIONS

  • ATM withdrawals
  • Cash withdrawal charges
  • ATM-related issues
  • Cash amount discrepancies

TRANSFERS_AND_PAYMENTS

  • Money transfers
  • Direct debits
  • Payment reception
  • Transfer status and issues

TOP_UP_OPERATIONS

  • Account top-ups
  • Balance updates
  • Top-up verification
  • Failed top-ups

ACCOUNT_VERIFICATION

  • Identity verification
  • PIN and passcode management
  • Security measures
  • Phone-related security

REFUNDS_AND_DISPUTES

  • Refund requests
  • Refund tracking
  • Dispute resolution

ACCOUNT_SETTINGS

  • Personal details management
  • Account termination

FEES_AND_CHARGES

  • Transaction fees
  • Exchange charges
  • Transfer and top-up fees

LIMITS_AND_RESTRICTIONS

  • Account limits
  • Currency support
  • Country restrictions
  • Payment method support

Dataset Structure

The dataset maintains the same structure as the original Banking77:

Training set: 10,003 examples Test set: 3,080 examples

The only difference is in the label space, which has been reduced from 77 to 10 categories.

Citation

If you use this dataset, please cite the original Banking77 dataset: Copy@article{banking77, title={Banking77 - A Fine-Grained Banking Customer Service Dataset}, author={Paul, S. and Ganchev, H. and Henderson, M.}, journal={arXiv preprint arXiv:2003.04807}, year={2020} }

License

This dataset follows the same license as the original Banking77 dataset.

Acknowledgments

We thank PolyAI for creating and sharing the original Banking77 dataset. This modified version aims to provide an alternative perspective on the data by offering a more consolidated view of banking intents.


dataset_info: features: - name: text dtype: string - name: label dtype: string splits: - name: train num_bytes: 856872 num_examples: 10003 - name: test num_bytes: 247690 num_examples: 3080 download_size: 384208 dataset_size: 1104562 configs: - config_name: default data_files: - split: train path: data/train-* - split: test path: data/test-*