Assigning categorical information to Japanese medical terms using MeSH and MEDLINE

Area of Science:

• Medical Informatics
• Natural Language Processing
• Bibliometrics

Background:

• Accurate indexing of biomedical literature is crucial for information retrieval.
• Existing English-Japanese dictionaries lack comprehensive Medical Subject Headings (MeSH) categorization.
• Previous work mapped 30,000 terms, leaving a substantial portion uncategorized.

Purpose of the Study:

• To develop and evaluate automated methods for assigning MeSH categories to Japanese terms in an English-Japanese dictionary.
• To leverage MEDLINE-indexed articles for calculating term relevancies and assigning MeSH categories.
• To compare the effectiveness of TF*IDF and inner product methods for weight matrix calculation.

Main Methods:

• Identified approximately 20,000 additional Japanese dictionary terms within MEDLINE-indexed article titles and abstracts (2000-2004).
• Employed two distinct approaches for calculating the weight matrix: TF*IDF and the inner product of weight matrices.
• Evaluated the precision and recall of the developed algorithms for both MeSH and non-MeSH terms.

Main Results:

• The automated methods identified an additional 20,000 Japanese terms within the MEDLINE corpus.
• Both TF*IDF and inner product methods demonstrated suboptimal precision and recall.
• The evaluated algorithms did not achieve high accuracy in automated MeSH categorization.

Conclusions:

• Automated assignment of MeSH categories to Japanese terms using the described methods yielded insufficient precision and recall.
• Despite limitations, the developed approach can serve as a supportive tool for manual MeSH categorization efforts.
• Further refinement of algorithms is necessary for more effective automated MeSH assignment in multilingual contexts.