Dataset Viewer
Auto-converted to Parquet Duplicate
_id
stringlengths
1
1
title
stringlengths
6
47
anzsic_version
stringdate
2006-01-01 00:00:00
2006-01-01 00:00:00
parent
stringclasses
0 values
A
Agriculture, Forestry and Fishing
2006
null
B
Mining
2006
null
C
Manufacturing
2006
null
D
Electricity, Gas, Water and Waste Services
2006
null
E
Construction
2006
null
F
Wholesale Trade
2006
null
G
Retail Trade
2006
null
H
Accommodation and Food Services
2006
null
I
Transport, Postal and Warehousing
2006
null
J
Information Media and Telecommunications
2006
null
K
Financial and Insurance Services
2006
null
L
Rental, Hiring and Real Estate Services
2006
null
M
Professional, Scientific and Technical Services
2006
null
N
Administrative and Support Services
2006
null
O
Public Administration and Safety
2006
null
P
Education and Training
2006
null
Q
Health Care and Social Assistance
2006
null
R
Arts and Recreation Services
2006
null
S
Other Services
2006
null

ANZSIC 2006 Company Classification, level 1 (Division), graded

Company descriptions matched against the 19 ANZSIC 2006 division codes with graded relevance, in the BEIR layout (queries, corpus, qrels) used by TechWolf/Skill-extraction-*-graded and loadable with the WorkRB GradedBEIRRankingTask. Sibling repos: ANZSIC-Company-Classification-L1, -L2-, -L3-, -L4-; companion NAICS repos use the same companies.

Config Split Rows Columns
queries train / validation / test 98,604 / 12,344 / 12,391 _id, text
corpus corpus 19 _id, title, anzsic_version, parent
qrels train / validation / test 457,335 / 57,072 / 57,970 query-id, corpus-id, score

corpus._id is the plain ANZSIC code (division letter, 2-digit subdivision, 3-digit group, 4-digit class); parent is the code one level up (null for sections). Documents are titles only. qrels is sparse (only non-zero judgments). Scores: 2 = reached from the company's primary NAICS code, 1 = reached only from secondary NAICS codes. A company may have several grade-2 codes (see below). Identical descriptions in validation/test were merged into one query (union of judgments, max grade; 25 in validation, 25 in test). Train descriptions that also occur in validation or test were removed.

from datasets import load_dataset
repo = "<your-org>/ANZSIC-Company-Classification-L1-graded"
queries = load_dataset(repo, "queries", split="test")
corpus  = load_dataset(repo, "corpus",  split="corpus")
qrels   = load_dataset(repo, "qrels",   split="test")

How the labels were derived (important)

These labels are derived, not human-annotated. Every company in the companion NAICS dataset has a primary and secondary NAICS code. They were translated in two hops, NAICS β†’ ISIC Rev.4 β†’ ANZSIC 2006:

  1. NAICS codes were mapped to ISIC Rev.4 classes (Census crosswalk; see the companion ISIC4-Company-Classification datasets for details).
  2. Each ISIC class was mapped to ANZSIC 2006 classes with the ABS ISIC Rev.4 ↔ ANZSIC 2006 correspondence (union of its two directions). Unlike NACE, ANZSIC does not nest inside ISIC: the correspondence is many-to-many, and the ABS partial-match flags are not used. Half of the 419 ISIC classes map to two or more ANZSIC classes (up to 21).
  3. ANZSIC classes were rolled up to group, subdivision and division, so all levels are consistent.
  4. Grade 2 = ANZSIC codes reached from the company's primary NAICS code; grade 1 = codes reached only from its secondary NAICS codes.

Consequences you should be aware of:

  • These labels are fuzzier than the NAICS, ISIC and NACE ones. The primary is ambiguous (more than one grade-2 code) for 23% of test companies at division level, 53% at subdivision, 66% at group and 72% at class level, and a company has about 25 relevant classes on average. Two lossy hops are chained, and the ANZSIC and NAICS hierarchies cut the economy differently. Use graded, multi-target metrics (nDCG, MAP) at every level, and treat division and subdivision as the most reliable.
  • Weights are an equal split at each hop (the tables carry none), so the grade-2 weights in the combined dataset's classification config are uniform by construction.
  • Errors in the NAICS labels and in the crosswalks carry through. The ANZSIC labels cannot be more accurate than the NAICS labels.
  • Companies whose primary NAICS code mapped to no valid ISIC class have no ANZSIC grade-2 label (primary_status = missing).
  • The same queries, splits and de-duplication as the NAICS, ISIC and NACE datasets are used, so results are directly comparable on the same companies.
  • Documents are titles only.

Provenance and license

TODO before publishing: state the source of the company descriptions and the original NAICS labels, and confirm the license (license: other is a placeholder). ANZSIC 2006 and its ISIC Rev.4 correspondence are published by the Australian Bureau of Statistics and Stats NZ (check their reuse terms; the ABS file is released under a Commonwealth of Australia copyright notice); the NAICS-to-ISIC crosswalk and NAICS 2022-to-2017 concordance are from the U.S. Census Bureau (confirm the exact sources and terms of the files you used).

Downloads last month
66