NAICS-GH

A multi-region, GPT-4.1-labeled corpus of GitHub repositories tagged with 2-digit NAICS industry sectors.

NAICS-GH contains 6,588 public GitHub repositories drawn from the USA, the European Union, and Australia, each labeled with one of the 19 retained 2-digit codes from the North American Industry Classification System (NAICS 2022). Sector 55 ("Management of Companies and Enterprises") is absent because too few repositories passed verification under the 80-sample minimum-class-size threshold.

How the labels were produced

A two-stage retrieve-and-verify pipeline:

1. A Presto/Trino query against GitHub's data warehouse yields ~1.37M public repositories (filters: num_stars >= 1, non-fork, non-spammy-owner, README ≥ 750 bytes, ≥ 6 commits, ≥ 2 contributors, non-empty description).
2. BAAI/bge-large-en-v1.5 embeddings indexed in FAISS retrieve the top-20 most semantically similar repositories per NAICS subindustry query, producing 28,341 candidate (repository, sector) pairs.
3. GPT-4.1 (snapshot gpt-4.1-2025-04-14) scores each candidate against a structured rubric on a 1–10 scale; only repositories scoring at least 8 are kept.

A stratified random sample of 2,421 USA repositories was manually re-checked by research assistants, confirming 96.98% label precision on the USA portion (Wilson 95% CI [96.23%, 97.59%]).

Splits

Split	Rows
train	4,611
validation	659
test	1,318
total	6,588

Splits are stratified by NAICS sector with random_state = 42. All 19 classes are present in every split.

Schema

Column	Type	Description
name_repo	string	Repository short name (no owner prefix)
description	string	Repository description from GitHub
topics	string	Semicolon-joined topic tags; empty if none
readme_content	string	Cleaned README text
label	int64	Integer class encoding 0–18
code	string	2-digit NAICS sector code

Class labels

label	code	NAICS sector
0	11	Agriculture, Forestry, Fishing and Hunting
1	21	Mining, Quarrying, and Oil and Gas Extraction
2	22	Utilities
3	23	Construction
4	31-33	Manufacturing
5	42	Wholesale Trade
6	44-45	Retail Trade
7	48-49	Transportation and Warehousing
8	51	Information
9	52	Finance and Insurance
10	53	Real Estate and Rental and Leasing
11	54	Professional, Scientific, and Technical Services
12	56	Administrative and Support and Waste Management and Remediation Services
13	61	Educational Services
14	62	Health Care and Social Assistance
15	71	Arts, Entertainment, and Recreation
16	72	Accommodation and Food Services
17	81	Other Services (except Public Administration)
18	92	Public Administration

Usage

from datasets import load_dataset

ds = load_dataset("aquiro1994/naics-gh")
print(ds)

# Inspect a sample
row = ds["train"][0]
print(row["name_repo"], "->", row["code"])

Each repository is most easily passed to a classifier by serializing its four text columns:

text = (
    f"Repository: {row['name_repo']} | "
    f"Description: {row['description']} | "
    f"Topics: {row['topics']} | "
    f"README: {row['readme_content']}"
)

This is the input format used by the published RoBERTa-large checkpoint at alexanderquispe/naics-github-classifier.

License

• Labels and metadata: CC-BY-4.0.
• Underlying repository content (READMEs, descriptions) remains governed by each repository's own license; the code column is a derived label, not an extract of any single repository's content.

Citation

@inproceedings{xu2026naicsgh,
  title  = {Industry Classification of GitHub Repositories
            Using the North American Industry Classification System (NAICS)},
  author = {Xu, Kevin and Quispe, Alexander},
  year   = {2026},
  note   = {Dataset available at https://huggingface.co/datasets/aquiro1994/naics-gh}
}

Limitations and caveats

• Validation coverage: The 96.98% precision figure is established on the USA portion of the corpus only. EU and AU repositories were labeled by the same GPT-4.1 pipeline but have not yet been included in the manual gold sample.
• English-only retrieval: BGE-large-en-v1.5 is trained on English text; non-English READMEs are under-represented.
• NAICS is a North American taxonomy: Applying NAICS to EU and AU repositories assumes that economic activities map cleanly across jurisdictions. Sector 22 (Utilities) is particularly US-centric.
• Label noise is not uniform: At score ≥ 8, sectors 31–33 (Manufacturing) and 42 (Wholesale Trade) have ~73% precision in the USA gold sample; raise the inclusion threshold to score ≥ 9 for stricter applications.
• Repository content licensing: The dataset releases labels about public repositories, plus excerpts of their READMEs. The underlying repositories remain governed by their own licenses.