Industry Classification of GitHub Repositories Using the North American Industry Classification System (NAICS)

2026-07-08 · Source: cs.SE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Data Science & Analytics · Depth: Expert, quick

Summary

Kevin Xu and Alexander Quispe introduce NAICS-GH, a new corpus of 6,588 GitHub repositories classified by 2-digit North American Industry Classification System (NAICS 2022) sectors. This dataset addresses the lack of standardized industry mapping for GitHub repositories, which hinders research on innovation geography and open-source production. The classification relies on a retrieve-and-verify pipeline that combines BAAI/bge-large-en embeddings, FAISS retrieval, and GPT-4.1 rubric scoring. This pipeline processed approximately 1.37 million source repositories, narrowing them to 31,178 candidates before selecting 6,588 high-confidence labels. The released labels demonstrate 96.98% precision on a 2,421-repository human-validated sample. Benchmarking six pretrained encoders on NAICS-GH shows RoBERTa-large achieving 86.45% F1 and 86.35% accuracy. The dataset, code, and fine-tuned checkpoint are publicly available.

Key takeaway

For research scientists studying innovation or open-source economics, NAICS-GH offers a critical resource for analyzing industry-specific software development. You can integrate this 6,588-repository dataset into your empirical work to map technology diffusion or industrial composition. Consider adapting the retrieve-and-verify pipeline for classifying other unstructured code data, leveraging its 96.98% precision for robust results.

Key insights

NAICS-GH provides a high-precision, AI-generated industry classification for 6,588 GitHub repositories using NAICS 2022.

Principles

• Standardized industry classification aids research.
• Hybrid AI pipelines enhance data labeling.
• Public datasets foster innovation studies.

Method

A retrieve-and-verify pipeline uses BAAI/bge-large-en embeddings for FAISS retrieval, followed by GPT-4.1 rubric scoring to assign 2-digit NAICS 2022 labels to GitHub repositories.

In practice

• Use NAICS-GH for innovation geography studies.
• Apply the pipeline for large-scale repository labeling.
• Benchmark encoders on industry-specific codebases.

Topics

• GitHub Repositories
• Industry Classification
• NAICS 2022
• Large Language Models
• Data Labeling
• Open-Source Research

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• A Multi-Plant Machine Learning Framework for Emission Prediction, Forecasting, and Control in Cement Manufacturing — Data-driven ML models can significantly reduce cement plant NOx emissions and costs without hardware changes.
• Accelerating researchers and developers building multilingual AI with a new open dataset — The GitHub Multilingual Repositories Dataset provides metadata to discover non-English content in over 40 million public repositories.
• Most Influential NAACL Papers (2026-03 Version) — Citation analysis from papers and patents reveals influential NAACL research, highlighting trends in NLP and AI agent development.
• General Supervised Learning Framework for Open World Classification — A data- and model-agnostic framework automates unknown class identification, categorization, and model updates in open-world classification.
• Learning the chemical language of natural products — The NaFM foundation model enhances natural product mining by learning their chemical language.
• So you’ve heard these AI terms and nodded along; let’s fix that — This glossary demystifies key AI terms, from AGI to Weights, for technical and professional readers.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.