Synonym set extraction from the biomedical literature by lexical pattern discovery

Area of Science:

• Biomedical informatics
• Computational linguistics
• Natural Language Processing

Background:

• Existing biomedical thesauri often have limited coverage of terms and their variations.
• Automatic thesaurus construction methods, while proposed, face challenges in pattern generation for specific semantic relations and domains.
• The reliance on syntactic analysis and external resources like parsers limits applicability across languages.

Purpose of the Study:

• To develop a method for automatically generating patterns to identify synonymous terms in the biomedical domain.
• To construct synonym sets from identified term pairs without requiring syntactic analysis.
• To evaluate the performance of the automated method against established resources.

Main Methods:

• Heuristically expanding seed patterns to generate regular expression patterns.
• Developing feature vectors based on term pair co-occurrence within generated patterns.
• Classifying term pairs as synonymous or non-synonymous and modeling results as a probability graph for synonym set identification.

Main Results:

• Achieved 73.2% precision and 29.7% recall in identifying synonymous terms.
• Demonstrated superior performance compared to manually curated resources such as MeSH and Wikipedia.
• Validated the effectiveness of the probability graph approach for optimal set cover in synonym set formation.

Conclusions:

• Automatic methods are practical for developing and expanding biomedical thesauri with minimal training data and no need for parsers.
• Grouping terms into synonym sets enhances the accuracy of automated thesaurus construction.
• The proposed method offers a scalable and efficient alternative for building comprehensive biomedical terminologies.