specifind 0.1.0

Specifind: A Natural Language Processing Tool for Automating Species Occurrence (Re-)Discovery from Scientific Literature

Specifind

Specifind is a Python toolkit built to automatically extract species occurrence information from unstructured ecological literature. It identifies scientific species names, geographic entities, and the relations connecting them—unlocking occurrence data hidden in text.

The toolkit integrates OCR, layout analysis, Named Entity Recognition, Coreference Resolution, and Relation Extraction into a unified and traceable pipeline. It is powered by a newly developed, expertly annotated dataset of 1,000+ ecological abstracts spanning biogeography, botany, entomology, mycology, and zoology.

---

🌿 Key Features

• 📄 Science-OCR for domain-optimized OCR of scientific papers
• 🔍 NER for scientific species names & geographic entities
• 🌍 Relation Extraction connecting species to locations
• 🧠 FastCOREF for high-performance coreference resolution
• 🧩 Built on spaCy for extensibility, speed, and NLP interoperability
• 🧱 Full pipeline for text & PDF extraction
• 🧭 Traceability that links extractions back to the original text

---

📦 Installation

pip install specifind

---

🚀 Quick Start

Basic Usage

from specifind import Specifind

s = Specifind()

s.analyze("Upupa epops is an exotic bird. It is widely extended over Spain.")

# or

s.analyze_file("path/to/file.pdf")

# Output:
# {
#     "species": [
#         "Upupa epops"
#     ],
#     "geography": [
#         "Spain"
#     ],
#     "occurrences": {
#         "Upupa epops": [
#             "Spain"
#         ]
#     },
#     "evidence": {
#         "Upupa epops": {
#             "Spain": [
#                 "It is widely extended over Spain."
#             ]
#         }
#     }
# }

---

📘 API Reference

analyze_file(...)

Process and extract information from a PDF file.

Parameters

Name	Type	Default	Description
path	str	—	Path to the file to analyze.
first_page	int	0	First page to process (inclusive).
last_page	int	PDF page length	Last page to process (exclusive).
coref	bool	True	Enable coreference resolution.
dpi	int	192 if GPU available else 96	Rendering DPI for PDF pages. Consider lowering the value if running out of memory (OOM).
return_doc	bool	False	If True, return Spacy Doc object with the annotations available in doc.ents and doc._.relations.

Returns

• Dictionary including parsed entities, relations and evidences.
• (optional) internal doc object (if return_doc=True)

---

analyze(...)

Process and extract information directly from raw text.

Parameters

Name	Type	Default	Description
text	str	—	Raw text to analyze.
coref	bool	True	Enable coreference resolution.
return_doc	bool	False	If True, return Spacy Doc object with the annotations.

Returns

• Dictionary including parsed entities, relations and evidences.
• (optional) internal doc object

---

🚀 Benchmarks

Named Entity Recognition (NER)

Species & Locations

🔍 Match Type	🎯 Precision	📈 Recall	🏆 F1
Exact	0.904	0.935	0.919
Partial/Intersect	0.938	0.969	0.958

---

Relation Extraction (RE)

Occurrences

🎯 Precision	📈 Recall	🏆 F1
0.964	0.993	0.978

---

🤝 Contributing

Contributions, issue reports, and feature suggestions are welcome. Feel free to open a Pull Request or discussion.

---

📄 License

Specifind is licensed under AGPL-3.0. See LICENSE for details.

---

📚 Citing Specifind

If you use Specifind in your research, please cite our pre-print:

BibTeX

@article{specifind2025,
  title   = {Specifind: A Natural Language Processing Tool for Automating Species Occurrence (Re-)Discovery from Scientific Literature},
  author  = {Golomb Durán Tomas, Díaz Anna, Barroso María, Far Antoni Josep, Roldán Alejandro, Cancellario Tommaso},
  year    = {2025},
  journal = {BioRxiv},
  url     = {https://github.com/ToGo347/specifind}
}