Enriching Knowledge Resources with Entity Linking

Abstract

This research project explores methods for enriching domain-specific knowledge resources through entity linking in the context of botanical and historical data. Focusing on the Time Capsule knowledge base, the study uses a variety of entity linking techniques to connect plant names with external databases such as the UMLS Metathesaurus and Dr. Duke’s Ethnobotanical Database. Four entity linking experiments are performed: baseline string matching, rule-based method using synonyms, and the deep learning models LUKE and SapBERT. The approaches are evaluated using precision, recall, accuracy, and coverage. Results demonstrate that SapBERT achieves the best balance between precision and coverage, making it the preferred technique for enriching the Time Capsule knowledge base with external relations. The findings highlight the practical challenges of domain-specific integration and provide directions for future refinement and expansion of knowledge graph enrichment pipelines.