A technique for semantic classification of unknown words using UMLS resources.

Area of Science:

• Medical Informatics
• Computational Linguistics
• Natural Language Processing

Background:

• Natural Language Processing (NLP) systems require robust semantic lexicons for accurate text analysis.
• Manual creation and maintenance of these lexicons are resource-intensive, time-consuming, and costly.
• The Unified Medical Language System (UMLS) offers a rich source of medical term semantics and syntax.

Purpose of the Study:

• To investigate the feasibility of automating semantic type classification for medical terms using UMLS resources.
• To develop a method for defining semantic types based on syntactic combinations within a clinical text corpus.
• To assess the accuracy of this automated classification method.

Main Methods:

• Utilized UMLS semantic and syntactic information to identify patterns of term co-occurrence in discharge summaries.
• Defined a semantic type by its characteristic syntactic combinations with other types.
• Applied these patterns to classify terms not present in the existing lexicon.
• Ranked newly classified terms based on the number of matching patterns.

Main Results:

• Successfully identified patterns of semantic type combinations within a corpus of discharge summaries.
• Generated a list of 875 candidate words for a specific semantic type.
• Achieved 80% accuracy in correctly classifying the top 95 words based on pattern matching.

Conclusions:

• The UMLS can be leveraged to automate the classification of medical terms, significantly aiding NLP lexicon development.
• Syntactic combination patterns provide a viable method for inferring semantic types of unknown medical terms.
• This approach offers a cost-effective and efficient strategy for expanding and maintaining medical NLP lexicons.