WikiData-based Entity Linking for Medical Applications

This repository provides resources for entity linking task using Wikidata knowledge base. Building on a recently introduced Named Entity Recognition pipeline, this project offers a complementary module for normalizing medical recognized entities across various clinical terminology systems and languages.

Repository Structure

• The named_entity_linking subfolder contains code for dataset acquision, model training and evaluation:

  • Scripts to generate custom entity linking dataset from Wikidata for given SPARQL-defined label classes
  • Model training via self-supervised contrastive learning strategy inspired by the SapBERT paper
  • Evaluation pipeline using FAISS for large-scale similarity retrieval on external datasets

• The demo subfolder includes a web application for querying custom ontologies via semantic similarity search:

  • Flask-based API with support for custom embedding models and knowledge bases
  • PostgreSQL database integration with the pgvector extension for fast vector-based retrieval