Using knowledge for indexing health web resources in a quality-controlled gateway

Area of Science:

• Medical Informatics
• Natural Language Processing
• Information Retrieval

Background:

• Accurate indexing is crucial for effective information retrieval in healthcare.
• Automated term extraction requires refinement to identify the most significant terms.

Purpose of the Study:

• To develop a method for identifying major MeSH terms from automatically extracted document terms.
• To assist indexers by providing a refined list of significant MeSH terms.

Main Methods:

• A hybrid approach combining symbolic knowledge (UMLS Metathesaurus, Semantic Network) and statistical knowledge (term co-occurrences).
• Utilized data mining measures on the CISMeF database, a French health sciences gateway.
• Tested the method on a corpus of 293 resources.

Main Results:

• The proposed method effectively reduces the number of non-descriptive terms.
• It successfully retains the most significant and descriptive MeSH terms for content description.
• The method achieved a proportion of 0.54+/-0.31 for initially identified major terms.

Conclusions:

• The method aids indexers by filtering out irrelevant terms like "check tags".
• It leverages both semantic and statistical knowledge to remove insignificant or non-statistically significant term associations.
• This approach is expected to enhance the efficiency of indexing in systems like CISMeF.