Grepator: accents & case mix for thesaurus

Area of Science:

• Medical Informatics
• Information Retrieval
• Natural Language Processing

Background:

• Researchers and students require effective pedagogical resources for information retrieval.
• French MeSH (Medical Subject Headings) indexing faces challenges with diacritics and user query mismatches.
• Existing information retrieval systems need improvement for handling specialized vocabularies and user input variations.

Purpose of the Study:

• To present the Grepator project, a tool designed to enhance medical information retrieval.
• To address inadequacies between user requests and controlled vocabularies like MeSH.
• To handle specific challenges posed by languages with diacritic characters, such as French.

Main Methods:

• Grepator transforms thesauri, standardizing case and diacritic characters for specific domains.
• It completes MeSH terms using their natural language forms found in a corpus.
• A statistical approach using regular expressions is employed, trained on a French medical corpus from CISMeF.
• The system corrects user spelling mistakes.

Main Results:

• Grepator successfully finds 75% of MeSH terms within the corpus using the statistical method.
• The tool demonstrates high accuracy, with fewer than one mistake per hundred words.
• The effectiveness of the Grepator tool in standardizing and correcting terms was evaluated.

Conclusions:

• Grepator offers a viable solution for improving the accuracy of information retrieval in specialized medical domains.
• The project demonstrates the potential of statistical methods and corpus analysis for enhancing thesaurus utilization.
• Further development and application of Grepator are recommended for broader use in medical education and research.