Figure 3.

Word cloud (http://www.wordle.net/create) representations of MeSH terms found in each corpus: Ab3P (top left), BIOADI (top right), MEDSTRACT (bottom left) and Schwartz and Hearst (bottom right). The MeSH terms confirm each corpus’ original intent: Ab3P was intended as a representation of all biomedical literature in PubMed, BIOADI is the corpus used in the BioCreative II gene normalization challenge, half of MEDSTRACT documents were a result of the search term ‘gene’ on MEDLINE restricted to a small group of biomedical journals and Schwartz and Hearst was a selection of documents returned as a result of the search term ‘yeast’ applied to PubMed.