96 ◾ Acc elerat ing Disc over y
keywords + 2. e two additional categories are “miscellaneous,” for
those snippets that do not mention any entity (if any), and “ambiguous,”
for those snippets that mention more than one entity name. We found an
“ambiguous” class approach is sometimes preferable to the multimember-
ship approach for two reasons:
1. Analyst does not have to read the same snippets again in dierent
categories when perusing the category.
2. e mentions of multiple entities could be a qualitatively dierent
event than the mention of a single entity by itself (e.g., to indicate a
protein complex).
In addition to physical entities of interest, like organism or proteins,
keyword taxonomies can capture nearly any other issue of ...