A Bootstrapping Approach for Entity Linking from Biomedical Literature

Abstract

Aim: Entity linking (EL) is a task of aligning literal of a named-entity from an unstructured document to appropriate entities in a knowledge base. The main objective of EL in biomedical domain stems on the construction of efficient computational models. Methodology: The development corpus is a subset of PubMed and Medline abstracts dealing with Huntington disease and its genes. It was annotated with disease and gene relations, based on “etiology†and “clinical biomarker.†The input corpus consists of text related to Huntington disease, gene names with their functions and all words related to neurogenetic disorders. The input corpus which is manually curated has 8998 sentences and 140,481 words. Results and Discussion: A bootstrap approach based on uniformity perception and similarity computation to link entities from unstructured biomedical texts to ontologies. A rich semantic information and structures in ontologies are influenced by the proposed approach for similarity computation and entity ranking. Conclusion: The proposed approach addresses the EL in the biomedical domain. The experiments show that our approach outperforms the existing state-of-the-art algorithms in terms of linkage accuracy.