Word Sense Disambiguation by Selecting the Best Semantic Type Based on Journal Descriptor Indexing: Preliminary...

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Information Retrieval

Background:

• The National Library of Medicine's (NLM) MetaMap system faces challenges in resolving word sense disambiguation (WSD) when mapping free text to Unified Medical Language System (UMLS) Metathesaurus concepts.
• Ambiguity arises when MetaMap assigns multiple concepts with high confidence, lacking a mechanism to determine the correct mapping.

Purpose of the Study:

• To introduce and evaluate the Journal Descriptor Indexing (JDI) methodology for improving WSD accuracy within NLM's biomedical text processing.
• To compare the performance of JDI against a baseline disambiguation method.

Main Methods:

• The JDI methodology utilizes statistical associations between words in MEDLINE citations and journal descriptors assigned to journals.
• It correlates these associations with UMLS Semantic Types (STs) to select the most appropriate meaning for ambiguous terms.
• An experiment compared four JDI versions against a baseline using 45 ambiguous strings from the NLM WSD Test Collection.

Main Results:

• The highest-scoring JDI version achieved an overall average precision of 0.7873, significantly outperforming the baseline's 0.2492.
• JDI demonstrated high precision for individual ambiguities, with over 51% achieving >0.90 and 79% achieving >0.65.

Conclusions:

• The JDI methodology offers a substantial improvement for word sense disambiguation in biomedical contexts.
• Further research aims to enhance JDI performance and explore its application in various NLM systems.