Automatic indexing errors in PubMed: Sometimes the apple does fall far from the tree

Objective: The objective of this study was to explore the prevalence and characteristics of precision errors in a sample of automatically indexed records in PubMed. Specifically, the sample was comprised of records automatically indexed with the MeSH term, Malus (the genus name for apple tree). Methods: A search was performed in PubMed to capture records automatically indexed with the MeSH term, Malus. The search retrieved 1,705 records, which underwent screening to determine whether they were correctly or incorrectly indexed. For records labeled as being incorrectly indexed, the quotations that used variations of the term, Malus, were extracted from each record. These quotations were then labeled with the type of indexing error. Results: Of the 1,705 records, 1,562 (91.6%) were correctly indexed, 135 (7.9%) were incorrectly indexed, and 8 (0.5%) could not be determined (i.e., the full text could not be accessed to determine indexing correctness). Of the 135 records incorrectly indexed, 80 (59%) were due to the use of figurative language (e.g., "comparing apples to oranges"); 50 (37%) were due to names (e.g., Apple, Inc.); 4 (3%) were due to acronyms (e.g., "Access to Post Partum LARC in Edinburgh South (APPLES)"); and 1 (1%) was due to a passing reference to Sir Isaac Newton. Conclusions: As demonstrated by this study, automatic indexing algorithms can experience errors when encountering words used in non-literal contexts. Until automatic indexing errors can be resolved, librarians should be mindful of their existence, and instruct researchers on how to avoid indexing errors in their own manuscripts.