R&D / Banking AI Evaluation Library / Datasets and sources
BANKING77
Short customer service queries about banking, each labelled with one of 77 intents. It is made for fine-grained intent detection.
Open the publisher page: BANKING77 dataset card (Hugging Face) (opens in a new tab)
Dodone links to the publisher and does not host or copy this data. Read the licence on the publisher page before you use it.
About this source
- Publisher
- PolyAI
- Context
- Online banking customer queries, English
- Version
- Not available (no version shown on the dataset card)
- Last checked
- 2026-11-03
- Labels or answers
- One intent label for each query.
Facts stated by the publisher
- 13,083 queries labelled with 77 intents.
- Training split of 10,003 queries and test split of 3,080 queries.
- English only, single domain.
Licence, access and privacy
- Licence
- Creative Commons Attribution 4.0 International (opens in a new tab) (Licence verified)
- Scope
- The dataset card states this licence for the data. Use needs attribution and the citation the card gives.
- Access
- Open. Public Hugging Face dataset. No sign-in is needed to read the card.
- Privacy
- Privacy not reviewed. The queries are short and general. Dodone has not reviewed them for personal information and does not ingest them.
Limitations
- Intents from one data set may not match the intents of another bank or channel.
- Testing abstention on out-of-scope queries needs extra test cases.
- Single queries do not test a multi-step conversation.
Suggested for these evaluations
- Banking customer query intent identification Direct fit
Why it helps: It has labelled banking customer queries across 77 intents, which matches this task.
What it does not validate: Out-of-scope abstention, other languages and multi-step conversations.