scispaCy is a Python library built on spaCy for biomedical natural-language processing and named entity disambiguation. It recognizes medical mentions, proposes knowledge-base candidates, and ranks them using surrounding context. section 8.0 section 7.2
A typical scispaCy disambiguation pipeline proceeds through three stages:
- Select candidate entities from a knowledge base such as UMLS. section 7.2
- Rank candidates using contextual words surrounding each recognized mention. section 7.2
- Integrate the selected entity with ontology identifiers, terminology, classifications, and coding standards. section 7.2
Disambiguating Zika mentions
A pipeline using en_core_sci_md and the UMLS linker recognizes “Zika virus,” “congenital Zika syndrome,” and “Zika virus infection,” returning ranked UMLS candidates for each mention. The selected entities can then be stored with mention positions, aliases, definitions, and types. section 7.6.3
scispaCy is strongest when contextual wording supports disambiguation, but it remains biomedical-domain-specific, is difficult to update with new terminology, and does not exploit relationships or paths between knowledge-base entities. section 8.1
