Integrating text-mining approaches to identify entities and extract events from the biomedical literature

s indicate that species names are more likely to be mentioned in the main body. As with the MeSH and Entrez Gene document sets, precision values are of low relevance due to the evaluation set not including all species mentions. 2.4.3.5 PubMed Central linkouts Performance of LINNAEUS compared to PMC linkouts reveals recall levels similar to those obtained on the EMBL document set, but lower than those for MeSH or Entrez Gene, despite the fact that this evaluation set has been constructed with the similar aim of performing species tagging as LINNAEUS (although at a document level). Inspecting a number of false positives and negatives reveals that virtually all were incorrectly tagged in the PMC linkout evaluation set, often for no apparent reason. For some false negative cases, articles have been tagged with species whose names can only be found in the titles of the references. This suggests that species names in the PMC linkouts are detected also in referenced article titles (while in some cases linkouts are missed even when species are mentioned in the main article title). Lower performance for MEDLINE and PMC OA abstracts is due to comparing species names found by LINNAEUS only in abstracts to those found in the full documents in PMC linkouts, and as such are nots is due to comparing species names found by LINNAEUS only in abstracts to those found in the full documents in PMC linkouts, and as such are not