information-extractor is a Python package that combines spaCy, coreferee, and SpanBERT to extract structured relationships between entities in natural language text. It's purpose-built for anyone who wants to bridge NER, coreference resolution, and relation extraction into one streamlined pipeline.

• Uses spaCy with coreferee to resolve pronouns and link entity mentions.
• Flexible support for multiple entity types: PERSON, ORG, LOC, DATE, etc.

• Uses fine-tuned SpanBERT model trained on TACRED.
• Handles subject/object marking and context-aware classification.
• Confidence scoring and de-duplication of extracted relations.
• GPU acceleration supported out of the box.

ie --text "Barack Obama was born in Hawaii." [--deps]

• --deps: Downloads and installs required pretrained models if not present.

pip install information-extractor

Run the following once to download SpanBERT, spaCy model, coreferee model:

ie --deps

Alternatively, you can import and run the dependency script directly:

from information_extractor.dependency import setup_dependencies
setup_dependencies()

from information_extractor.pipeline import RelationExtractor

text = "Sundar Pichai is the CEO of Google. He lives in California."

extractor = RelationExtractor()
results = extractor.extract(text)

for relation in results:
    print(relation)

[
  {
    "subject": "Sundar Pichai",
    "object": "Google",
    "relation": "per:employee_of",
    "confidence": 0.92
  },
  ...
]

information_extractor/
├── assets/
│   └── pretrained_spanbert/
├── dependency.py         # Downloads all model dependencies
├── pipeline.py           # Core logic for NLP + SpanBERT
├── main.py               # CLI entrypoint

Models are downloaded from hosted GitHub release assets:

• ✅ SpanBERT weights & config
• ✅ en_core_web_md spaCy model
• ✅ coreferee_model_en for coreference resolution
• ✅ torch wheel for reproducibility

This project builds on the work of Facebook Research. If you use SpanBERT, please cite:

@article{joshi2019spanbert,
  title={{SpanBERT}: Improving Pre-training by Representing and Predicting Spans},
  author={Mandar Joshi and Danqi Chen and Yinhan Liu and Daniel S. Weld and Luke Zettlemoyer and Omer Levy},
  journal={arXiv preprint arXiv:1907.10529},
  year={2019}
}

MIT. See LICENSE for full terms.
Note: This project redistributes pretrained model weights for convenience under fair use for research.