Related Experiment Video
Updated: Feb 17, 2026
05:47
Evidence-based Knowledge Synthesis and Hypothesis Validation: Navigating Biomedical Knowledge Bases via Explainable AI and Agentic Systems
Published on: June 13, 2025
1.6K
Supervised Learning and Knowledge-Based Approaches Applied to Biomedical Word Sense Disambiguation
1.
Journal of Integrative Bioinformatics
|December 14, 2017
Summary
Supervised word sense disambiguation (WSD) achieved 95.6% accuracy using word embeddings and bag-of-words features. This approach significantly improved biomedical concept disambiguation compared to knowledge-based methods.
Area of Science:
- Biomedical text mining
- Natural Language Processing
- Computational Linguistics
Background:
- Word sense disambiguation (WSD) is crucial for accurate biomedical information extraction.
- Ambiguous terms in biomedical texts require precise concept assignment for effective data mining.
Purpose of the Study:
- To evaluate supervised and knowledge-based approaches for biomedical word sense disambiguation.
- To assess the impact of word embeddings and UMLS definitions on WSD accuracy.
Main Methods:
- Implemented supervised WSD using bag-of-words (local) and word embeddings (global) features.
- Developed a knowledge-based WSD method combining word embeddings, UMLS definitions, and MeSH co-occurrence data.
- Tested various word embedding averaging functions to enhance context vector representation.
Main Results:
- Supervised WSD achieved a top accuracy of 95.6% on the MSH WSD dataset.
- The best knowledge-based WSD method reached an accuracy of 87.4%.
- Word embedding models demonstrated a significant improvement in disambiguation accuracy.
Conclusions:
- Supervised methods, particularly with word embeddings, are highly effective for biomedical WSD.
- Word embeddings are a powerful resource for enhancing the accuracy of WSD tasks in the biomedical domain.
