Named Entity Recognition Model

This SageMaker model package provides a REST API to detect named entities in US English text. Send a JSON array of sentences and receive labeled spans with character offsets and confidence scores.

Eighteen entity types are detected by default (person, organization, location, date, money, and more). Pass an optional "labels" array to extract any entity types you need (drugs, statutes, tickers) without retraining.

GPU is optional. Deploy on a GPU instance (for example ml.g4dn.xlarge) for higher throughput; the same image uses CUDA automatically when a GPU is present, and falls back to CPU otherwise. The API accepts JSON or CSV and supports real-time endpoints and SageMaker batch transform.

We welcome your feedback at [email protected]. Usage notebook: https://colab.research.google.com/drive/1iL1Q0FiYfKoUqYzglpeBFDxzDjV8IF9s

Measured performance - Span-level micro F1 0.691 (precision 0.628, recall 0.768), measured on 1,000 held-out examples covering 1,542 gold entity spans. An entity counts as correct only on an exact character-span and label match. - Measured against the published model package on an ml.m5.large endpoint, not against a local checkpoint, so the number reflects what the endpoint returns. - Precision improved versus the prior package (0.50 to 0.63) after a retrain with hard negatives and per-label score floors. Strongest label is PERSON (F1 0.85); FAC is 0.62; weakest is still PRODUCT (0.48). Raise the per-entity score threshold further if you need more precision. - Throughput about 4 texts per second on ml.m5.large at a batch size of 25 (ml.m5.xlarge remains the larger CPU option).