Analyzing the Impact of UMLS Relations on Word-sense Disambiguation Accuracy

Abstract

Word-sense disambiguation (WSD) is the process of finding the correct meaning of words that have multiple meanings. The unsupervised WSD algorithm is the type of WSD algorithm that leverages an external source of knowledge to guide the disambiguation process. The unsupervised WSD algorithm type is attracting more interest in the biomedical domain because of its implementation practicality, especially when it leverages the knowledge sources of the Unified Medical Language System (UMLS), but still the resulted accuracy of the unsupervised WSD algorithm is lower than its supervised alternative. In this study we analyze the impact of using different subsets of the UMLS on the resulted accuracy of the unsupervised WSD algorithm. Our findings show that there are better ways to leverage the UMLS than using it as a monolithic source of knowledge.