You are here
Semi-Automatic Indexing of Full Text Biomedical Articles
The main application of U.S. National Library ofMedicine's Medical Text Indexer (MTI) is to provideindexing recommendations to the Library's indexingstaff. The current input to MTI consists of the titlesand abstracts of articles to be indexed. This studyreports on an extension of MTI to the full text of articlesappearing in online medical journals that areindexed for Medline. Using a collection of 17 journalissues containing 500 articles, we report on theeffectiveness of the contribution of terms by the wholearticle and also by each section. We obtain the bestresults using a model consisting of the sectionsResults, Results and Discussion, and Conclusionstogether with the article's title and abstract, the captionsof tables and figures, and sections that have notitles. The resulting model provides indexing significantlybetter (7.4%) than what is currently achievedusing only titles and abstracts.