| --- |
| license: apache-2.0 |
| language: |
| - en |
| library_name: transformers |
| pipeline_tag: text-classification |
| base_model: distilroberta-base |
| datasets: |
| - aitrainingdatabroker/synthetic-document-sensitivity-tiers |
| tags: |
| - data-classification |
| - sensitivity |
| - data-governance |
| - text-classification |
| - synthetic |
| --- |
| |
| # Document Sensitivity Tier Classifier |
|
|
| Reads a short description of a document and suggests a sensitivity tier: `public`, `internal`, `confidential` or `restricted`. |
|
|
| Every name, number and sentence it was trained on is made up. Nothing comes from a real person, customer or business. |
|
|
| Related tool: [document-sensitivity-keyword-sorter](https://huggingface.co/spaces/aitrainingdatabroker/document-sensitivity-keyword-sorter) (a Space that runs in your browser). |
|
|
| ## Why it exists |
|
|
| Before data leaves a company, someone sorts what it holds by how sensitive it is. This model gives that sort a first draft. |
|
|
| ## Use it |
|
|
| ```python |
| from transformers import pipeline |
| |
| classify = pipeline("text-classification", model="aitrainingdatabroker/document-sensitivity-tier-classifier", top_k=3) |
| print(classify("Customer database export with emails, phone numbers and home addresses.")) |
| ``` |
|
|
| ## How it was trained |
|
|
| Fine-tuned from `distilroberta-base` for 6 epochs on one RTX 3070. Training data: 305 rows from [synthetic-document-sensitivity-tiers](https://huggingface.co/datasets/aitrainingdatabroker/synthetic-document-sensitivity-tiers), all invented, with test rows built from templates the model never saw in training. The scripts that built the data and trained the model are in `source/`; the full scores are in `scores.json`. |
|
|
| ## Test scores |
|
|
| On 77 held-out rows: accuracy 0.831, macro F1 0.823. |
|
|
| | Label | F1 | Count | |
| |---|---|---| |
| | `confidential` | 0.857 | 19 | |
| | `internal` | 0.816 | 20 | |
| | `public` | 0.974 | 20 | |
| | `restricted` | 0.645 | 18 | |
|
|
| ## Limits |
|
|
| - Trained only on invented text from templates, with test sentences built from templates it never saw. Real text varies far more. |
| - Tiers differ between companies and laws. A model's tier is a suggestion; the owner of the data sets the tier. |
| - This model is not legal advice and not a certification. A person should review its output before anything is shared or decided, and your counsel decides what a rule requires. |
|
|
| ## Read more |
|
|
| - [De-identification before data is licensed](https://aitrainingdatabroker.com/de-identification/?utm_source=huggingface&utm_medium=model&utm_campaign=document-sensitivity-tier-classifier) |
| - [Privacy rules by region](https://aitrainingdatabroker.com/regions/?utm_source=huggingface&utm_medium=model&utm_campaign=document-sensitivity-tier-classifier) |
| - [Can you license SOPs and knowledge bases?](https://aitrainingdatabroker.com/data/sops-and-knowledge-bases/?utm_source=huggingface&utm_medium=model&utm_campaign=document-sensitivity-tier-classifier) |
|
|
| ## License and credit |
|
|
| Apache 2.0. Please credit **AI Training Data Broker** and link to https://aitrainingdatabroker.com/ when you share or build on it. |
|
|
| ## About |
|
|
| AI Training Data Broker is a broker and technical advisor for companies that sell or license their data to AI labs. It guides the technical side, including export, format, de-identification and compliance, and publishes free guides, checklists and tools. |
|
|
