README.md

metadata

license: mit
task_categories:
  - table-question-answering
  - token-classification
language:
  - en
tags:
  - finance
  - xbrl
  - numeric-tagging
  - fintagging
size_categories:
  - 1K<n<10K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train.parquet
      - split: test
        path: data/test.parquet

FinTagging 800/200 Split

This dataset is a sampled train/test subset of the labeled FinTagging data. Only examples with at least one annotated numeric entity are included.

Splits

Split	Samples	Entity triples	Unique datatypes	Unique concepts
train	800	11,687	5	1,248
test	200	2,546	5	395

Columns

• source_sample_idx: original row index from the source FinTagging parquet file.
• split: split label, either train or test.
• text: HTML table/text context.
• numeric_entities: list of annotated numeric entities. Each entity contains concept, type, and value.
• context_id: original context identifier.

Split Construction

The split was constructed from 3,641 labeled examples. The training split was selected first, and test examples were restricted so every test concept and datatype also appears in the training split. The split was also checked to ensure there is no overlap between train and test by source_sample_idx or context_id.

Validation checks:

Check	Value
source sample index overlap	0
context ID overlap	0
test concepts missing from train	0
test datatypes missing from train	0

Datatype Distribution

Datatype	Train count	Train %	Test count	Test %
monetaryItemType	10,448	89.40	2,285	89.75
percentItemType	574	4.91	116	4.56
sharesItemType	375	3.21	70	2.75
perShareItemType	219	1.87	58	2.28
integerItemType	71	0.61	17	0.67

Files

• data/train.parquet: training split.
• data/test.parquet: testing split.
• metadata/split_report.json: split-generation report.
• metadata/statistics_summary.json: summary statistics and validation checks.