Mark Neumann: ScispaCy: A spaCy pipeline & models for scientific & biomedical text

2019-07-06 · Source: Explosion · Developer tools and consulting for AI, Machine Learning and NLP - Explosion.ai · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Biomedical Natural Language Processing · Depth: Intermediate, long

Summary

SciSpacy is a spaCy package specifically designed for processing scientific and biomedical text, developed at the Allen Institute. It addresses the limitations of general-purpose spaCy models when applied to specialized domains, which often result in incorrect part-of-speech tagging, dependency parsing, and irrelevant named entity recognition (NER) labels. SciSpacy leverages extensive biomedical resources, including ontologies, annotated corpora like MedMentions and CRAFT, and the raw PubMed corpus, to train its models. The core pipeline includes universal dependency parsing, a generic mention detector, and specialized NER models for subfields such as cancer genomics. Notably, SciSpacy offers a significant speed advantage, being approximately 30 times faster than traditional tools like MetaMap, and incorporates custom components for tasks like unsupervised abbreviation detection (achieving 95% precision and 82% recall) and efficient candidate generation for entity linking.

Key takeaway

For NLP Engineers or Research Scientists processing biomedical text, general-purpose spaCy models are insufficient due to domain-specific linguistic challenges. You should adopt SciSpacy for its specialized models, 30x faster performance than tools like MetaMap, and custom components for tasks like abbreviation detection and entity linking. This allows you to build robust, efficient systems without extensive infrastructure work, accelerating your research and application development.

Key insights

SciSpacy provides a fast, domain-specific NLP pipeline for biomedical text, overcoming general spaCy limitations with specialized models and components.

Principles

• Domain-specific NLP requires tailored models.
• Iteration speed is more valuable than thought.
• Simple heuristics can outperform complex ML.

Method

SciSpacy builds on spaCy's architecture, integrating universal dependencies, a generic mention detector, and specialized NER. It uses TF-IDF for entity linking candidate generation and an unsupervised abbreviation detection algorithm.

In practice

• Use SciSpacy for biomedical text processing.
• Develop custom components for unique domain needs.
• Combine heuristic and ML methods.

Topics

• SciSpacy
• Biomedical NLP
• spaCy Pipelines
• Entity Linking
• Abbreviation Detection
• Information Extraction

Best for: AI Engineer, AI Scientist, NLP Engineer, Machine Learning Engineer, Research Scientist

Related on AIssential

• The application of natural language processing for the extraction of mechanistic information in toxicology — spaCy with scispaCy enables robust NER for extracting mechanistic toxicology data from scientific literature.
• David Dodson: spaCy in the News: Quartz’s NLP pipeline — Domain-specific NLP pipelines, like Scio2, excel in time-sensitive news analysis by dynamically adapting to evolving language and context.
• 3 SpaCy Tricks for Efficient Text Processing & Entity Recognition — Optimizing spaCy pipelines requires tailoring component usage, utilizing batch processing, and integrating rule-based entity recognition.
• KliniskVestBERT: BERT Model Specialised to Norwegian Clinical Texts — Domain-specific pre-training significantly enhances BERT models for Norwegian clinical NLP tasks.
• How to Leverage NER and Advanced NLP Techniques for Life Sciences — NER and advanced NLP are essential for structuring and interpreting the vast unstructured text data in Life Sciences.
• EQ-5D Classification Using Biomedical Entity-Enriched Pre-trained Language Models and Multiple Instance Learning — Entity enrichment significantly enhances PLM performance for domain-specific text classification in biomedical literature.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Explosion · Developer tools and consulting for AI, Machine Learning and NLP - Explosion.ai.