Co-occurrence graphs for word sense disambiguation in the biomedical domain
作者：
Highlights：
• Co-occurrence graphs provide competitive results in biomedical WSD.
• The proposed approach reduces the use of external resources to a minimum.
• Evaluation on the NLM-WSD test dataset overcomes knowledge-base state-of-the-art systems.
• Less restrictive graphs, which avoid the removal of important edges, lead to better results.
• Convergence is fast as the number of documents used for building the graph increases.
摘要
•Co-occurrence graphs provide competitive results in biomedical WSD.•The proposed approach reduces the use of external resources to a minimum.•Evaluation on the NLM-WSD test dataset overcomes knowledge-base state-of-the-art systems.•Less restrictive graphs, which avoid the removal of important edges, lead to better results.•Convergence is fast as the number of documents used for building the graph increases.
论文关键词：Word sense disambiguation,Graph-based systems,Unsupervised machine learning,Unified medical language system,Natural language processing,Information extraction
论文评审过程：Received 17 July 2017, Revised 23 January 2018, Accepted 11 March 2018, Available online 21 March 2018, Version of Record 28 May 2018.
论文官网地址：https://doi.org/10.1016/j.artmed.2018.03.002
