Dataset Card for BANKING77

Table of Contents

• Dataset Description
  • Dataset Summary
  • Supported Tasks and Leaderboards
  • Languages
• Dataset Structure
  • Data Instances
  • Data Fields
  • Data Splits
• Dataset Creation
  • Curation Rationale
  • Source Data
  • Annotations
  • Personal and Sensitive Information
• Considerations for Using the Data
  • Social Impact of Dataset
  • Discussion of Biases
  • Other Known Limitations
• Additional Information
  • Dataset Curators
  • Licensing Information
  • Citation Information
  • Contributions

Dataset Description

• Homepage: Github
• Repository: Github
• Paper: ArXiv
• Leaderboard:
• Point of Contact:

Dataset Summary

Dataset composed of online banking queries annotated with their corresponding intents.

BANKING77 dataset provides a very fine-grained set of intents in a banking domain. It comprises 13,083 customer service queries labeled with 77 intents. It focuses on fine-grained single-domain intent detection.

Supported Tasks and Leaderboards

Intent classification, intent detection

Languages

English

Dataset Structure

Data Instances

An example of 'train' looks as follows:

{
  'label': 11, # integer label corresponding to "card_arrival" intent
  'text': 'I am still waiting on my card?'
}

Data Fields

• text: a string feature.
• label: One of classification labels (0-76) corresponding to unique intents.

Intent names are mapped to label in the following way:

Data Splits

Dataset Creation

Curation Rationale

Previous intent detection datasets such as Web Apps, Ask Ubuntu, the Chatbot Corpus or SNIPS are limited to small number of classes (<10), which oversimplifies the intent detection task and does not emulate the true environment of commercial systems. Although there exist large scale multi-domain datasets (HWU64 and CLINC150), the examples per each domain may not sufficiently capture the full complexity of each domain as encountered "in the wild". This dataset tries to fill the gap and provides a very fine-grained set of intents in a single-domain i.e. banking. Its focus on fine-grained single-domain intent detection makes it complementary to the other two multi-domain datasets.

Source Data

Initial Data Collection and Normalization

[More Information Needed]

Who are the source language producers?

[More Information Needed]

Annotations

Annotation process

The dataset does not contain any additional annotations.

Who are the annotators?

[N/A]

Personal and Sensitive Information

[N/A]

Considerations for Using the Data

Social Impact of Dataset

The purpose of this dataset it to help develop better intent detection systems.

Any comprehensive intent detection evaluation should involve both coarser-grained multi-domain datasets and a fine-grained single-domain dataset such as BANKING77.

Discussion of Biases

[More Information Needed]

Other Known Limitations

[More Information Needed]

Additional Information

Dataset Curators

PolyAI

Licensing Information

Creative Commons Attribution 4.0 International

Citation Information

@inproceedings{Casanueva2020,
    author      = {I{\~{n}}igo Casanueva and Tadas Temcinas and Daniela Gerz and Matthew Henderson and Ivan Vulic},
    title       = {Efficient Intent Detection with Dual Sentence Encoders},
    year        = {2020},
    month       = {mar},
    note        = {Data available at https://github.com/PolyAI-LDN/task-specific-datasets},
    url         = {https://arxiv.org/abs/2003.04807},
    booktitle   = {Proceedings of the 2nd Workshop on NLP for ConvAI - ACL 2020}
}

Contributions

Thanks to @dkajtoch for adding this dataset.