Datasets:

LiquidAI
/

structured-extraction

license: other
language:
  - en
  - ar
  - zh
  - fr
  - de
  - it
  - ja
  - ko
  - pt
  - es
  - pl
  - vi
  - th
  - id
  - hi
  - ru
task_categories:
  - text-generation
tags:
  - information-extraction
  - structured-output
  - json
  - json-schema
  - synthetic
  - multilingual
size_categories:
  - 1K<n<10K
configs:
  - config_name: default
    data_files:
      - split: train
        path: train.parquet
      - split: test
        path: test.parquet
      - split: test_unseen
        path: test_unseen.parquet
      - split: showcase
        path: showcase.parquet

Structured Extraction

📓 Tutorial: This is the dataset used in SFT with TRL

Documents paired with a JSON Schema and the JSON object extracted from them. The documents (emails, receipts, bookings, forms, letters, listings, ...) cover 35 document types in 16 languages. All people, companies, addresses and identifiers are fictional.

Splits

Split	Rows	Content
`train`	6,000	30 document types
`test`	500	the same 30 document types, other documents
`test_unseen`	300	5 document types that are not in `train`
`showcase`	6	examples of the 5 unseen document types

Fields

Field	Description
`id`	row identifier
`doc_type`	document type
`language`	ISO 639-1 code of the document
`schema`	JSON Schema of the target object, as a JSON string
`document`	document text
`extraction`	target object, keys in schema order, as a JSON string

Conventions

Strings are copied from the document as written, without field labels or honorifics.
Dates are YYYY-MM-DD, times HH:MM (24-hour), amounts are JSON numbers, and currencies are ISO 4217 codes.
Information absent from the document is null ([] for arrays).
When the document contains several candidate values, the field description says which one to extract.