_id stringlengths 1 1 | title stringlengths 9 122 | isic_version stringclasses 1
value | parent stringclasses 0
values |
|---|---|---|---|
A | Agriculture, forestry and fishing | Rev.4 | null |
B | Mining and quarrying | Rev.4 | null |
C | Manufacturing | Rev.4 | null |
D | Electricity, gas, steam and air conditioning supply | Rev.4 | null |
E | Water supply; sewerage, waste management and remediation activities | Rev.4 | null |
F | Construction | Rev.4 | null |
G | Wholesale and retail trade; repair of motor vehicles and motorcycles | Rev.4 | null |
H | Transportation and storage | Rev.4 | null |
I | Accommodation and food service activities | Rev.4 | null |
J | Information and communication | Rev.4 | null |
K | Financial and insurance activities | Rev.4 | null |
L | Real estate activities | Rev.4 | null |
M | Professional, scientific and technical activities | Rev.4 | null |
N | Administrative and support service activities | Rev.4 | null |
O | Public administration and defence; compulsory social security | Rev.4 | null |
P | Education | Rev.4 | null |
Q | Human health and social work activities | Rev.4 | null |
R | Arts, entertainment and recreation | Rev.4 | null |
S | Other service activities | Rev.4 | null |
T | Activities of households as employers; undifferentiated goods- and services-producing activities of households for own use | Rev.4 | null |
U | Activities of extraterritorial organizations and bodies | Rev.4 | null |
ISIC Rev.4 Company Classification, level 1 (Section), graded
Company descriptions matched against the 21 ISIC Rev.4 section 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: ISIC4-Company-Classification-L1, -L2-, -L3-, -L4-; companion NAICS repos use the same companies.
| Config | Split | Rows | Columns |
|---|---|---|---|
queries |
train / validation / test | 98,750 / 12,366 / 12,413 | _id, text |
corpus |
corpus |
21 | _id, title, isic_version, parent |
qrels |
train / validation / test | 323,419 / 40,368 / 41,032 | query-id, corpus-id, score |
corpus._id is the plain ISIC code (section letter, or 2 to 4 digits); 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>/ISIC4-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
(NAICS-Company-Classification-*) has a primary and secondary NAICS code. They were translated to ISIC Rev.4:
- A NAICS code valid in NAICS 2022 goes through the Census 2022 → 2017 concordance, then the NAICS 2017 → ISIC Rev.4 crosswalk. A code that was retired (NAICS 2017 only) goes straight through the crosswalk.
- The ISIC classes (4 digits) reached are rolled up to group, division and section, so all levels are consistent.
- Grade 2 = ISIC 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:
- The crosswalk is many-to-many, so a primary NAICS code often maps to several ISIC codes, all graded 2. Unlike the NAICS datasets, a company can therefore have more than one primary code. The share of test companies whose primary is ambiguous (more than one grade-2 code) is 8% (section), 21% (division), 28% (group), 31% (class). Evaluate with graded, multi-target metrics (nDCG, MAP) rather than single-label accuracy at fine levels.
- Errors in the NAICS labels and in the crosswalks carry through. The ISIC labels cannot be more accurate than the NAICS labels.
- Three ISIC codes in the crosswalk (
0120,0140,0331) are not valid in ISIC Rev.4. They were not used as classes; their valid ancestors (e.g. group012) were kept. - NAICS 2022 code
325315(Compost Manufacturing) is new in 2022 and has no ISIC mapping, so it contributes no ISIC label. - The same queries, splits and de-duplication as the NAICS datasets are used, so NAICS and ISIC results are directly comparable on the same companies.
Provenance and license
TODO before publishing: state the source of the company descriptions and the original NAICS labels, and confirm the license (
license: otheris a placeholder). ISIC Rev.4 is published by the United Nations Statistics Division; the NAICS-to-ISIC crosswalk and NAICS 2022-to-2017 concordance are from the U.S. Census Bureau (confirm the exact source and terms of the crosswalk file you supplied).
- Downloads last month
- 44