ICICLE-AI/organization-sic-code_smart-foodsheds

Organization SIC Code Dataset

This dataset is released as part of the ICICLE project and is used for training models to classify organizations into Standard Industrial Classification (SIC) codes. The dataset contains multiple textual representations of organizational descriptions collected via web search and large language model (LLM) summarization.

Dataset Structure

The dataset consists of three different versions of organization descriptions:

1. 1.gsnip: Google search snippet-based text
2. 2.gptsummary: LLM-generated summaries using GPT-4o mini
3. 3.llamasummary: LLM-generated summaries using LLaMA 3.1 8B Instruct

Each version is divided into standard splits: train, dev, and test.

Each CSV contains:

• —organization: Organization name
• —label: Ground truth SIC code label
• —google_snippet / gpt_response / llama-summary: Task-specific text description

How to Use

Clone or download the dataset using:

git lfs install
git clone https://huggingface.co/datasets/ICICLE-AI/organization-sic-code_smart-foodsheds

Then unzip the data archive:
unzip organization-sic-code_smart-foodsheds/data.zip -d organization-sic-code_smart-foodsheds/data

You will get the following structure:
data/
├── gsnip/
│   ├── train.csv
│   ├── dev.csv
│   └── test.csv
├── gptsummary/
│   ├── train.csv
│   ├── dev.csv
│   └── test.csv
├── llamasummary/
│   ├── train.csv
│   ├── dev.csv
│   └── test.csv