Identifying named entities from PubMed for enriching semantic categories

Area of Science:

• Biomedical Natural Language Processing (NLP)
• Computational Biology
• Bioinformatics

Background:

• Controlled vocabularies like Unified Medical Language System (UMLS) and Medical Subject Headings (MeSH) are crucial for biomedical NLP.
• Standard terminology in these resources has low representation in biomedical literature, with only 13% of UMLS terms appearing in MEDLINE.

Purpose of the Study:

• To develop an efficient and effective method for extracting noun phrases representing biomedical semantic categories from literature.
• To bridge the gap between established controlled vocabularies and the actual terms used in biomedical texts.

Main Methods:

• Utilized simple linguistic patterns to identify candidate noun phrases based on headwords.
• Employed a machine learning classifier to filter out irrelevant or noisy phrases.
• Tested three NLP rules and evaluated them manually by three annotators.

Main Results:

• Achieved high precision, exceeding 93% on average for headwords including 'gene', 'protein', 'disease', 'cell', and 'cells'.
• Demonstrated the effectiveness of the proposed approach in identifying relevant biomedical entities.

Conclusions:

• The developed method offers a practical solution for enriching semantic categories by incorporating terms directly from biomedical literature.
• While manual evaluation is still beneficial, the automated solution provides high precision and convenience for expanding terminological resources.