| language: en | |
| license: mit | |
| tags: | |
| - stylometry | |
| - authorship-attribution | |
| - bitcoin | |
| - satoshi-nakamoto | |
| - text-classification | |
| pipeline_tag: text-classification | |
| datasets: | |
| - custom | |
| metrics: | |
| - accuracy | |
| - f1 | |
| model-index: | |
| - name: satoshi-stylometry | |
| results: | |
| - task: | |
| type: text-classification | |
| name: Authorship Attribution | |
| metrics: | |
| - name: Accuracy | |
| type: accuracy | |
| value: 0.991 | |
| - name: F1 | |
| type: f1 | |
| value: 0.983 | |
| # Satoshi Stylometry | |
| A ModernBERT-base model fine-tuned to detect Satoshi Nakamoto's writing style. | |
| ## Usage | |
| ```python | |
| from transformers import pipeline | |
| classifier = pipeline("text-classification", model="thestalwart/satoshi-stylometry") | |
| result = classifier("The proof-of-work chain is the solution to the synchronization problem.") | |
| print(result) | |
| # [{'label': 'satoshi', 'score': 0.99}] | |
| ``` | |
| ## Training Data | |
| - **572 Satoshi writings**: BitcoinTalk posts, cryptography mailing list emails, P2P Foundation posts | |
| - **1,546 non-Satoshi writings**: Same era cryptography mailing list and BitcoinTalk forum posts | |
| ## Results | |
| 99.1% accuracy on 350 held-out texts (97.8% Satoshi recall, 99.6% non-Satoshi recall). | |
| ## Links | |
| - [Demo Space](https://huggingface.co/spaces/thestalwart/satoshi-stylometry) | |
| - [GitHub](https://github.com/jnathan9/satoshi-stylometry) | |