You are here
Evaluation of System for Selective Indexing Classification.
Finding and accessing scientific literature in PubMed partially relies on indexing articles for MEDLINE® using Medical Subject Headings (MeSH®). Currently, more than 5,000 journals in MEDLINE are fully indexed, where the majority of articles published by these journals have MeSH indexing assigned. However, for approximately 130 selectively indexed journals, articles are only indexed if, after manual review, they are determined to be relevant to biomedicine and the life sciences. In recent years, the number of these articles has increased substantially; for example, in 2018 about 125,000 were added to PubMed. To reduce the number of articles that must be manually reviewed, we have developed the Biomedical Citation Selector (BmCS), a high recall machine learning system that identifies the articles that require indexing. Here, we manually evaluate BmCS’s automated predictions to better understand its performance in the indexing workflow.