Traditional text mining tools rely on large dictionaries and/or simple regular expressions for identifying terms (called entities) in unstructured text. Unfortunately, these approaches are inappropriate for extracting chemical information, where the number of possible systematic chemical names is much larger than any possible dictionary. Almost by definition, the most interesting chemical structures in a patent are those whose names are novel, and therefore unlikely to be in a dictionary or otherwise seen before. To avoid these problems, chemical text mining tools use grammars that understand how chemical names are constructed, and can therefore recognize an infinite number of possible IUPAC names. In this presentation, we show how grammar technology originally developed for chemical nomenclature, can be used to identify many other useful entities with text mining. These include CAS numbers (confirming the correct check digit), dates, journal citations, chemical formulae, SMILES, InChIs, HELM, V2000 Mol files, complex diseases and anatomical terms, URLs, currency, zip codes, people's names, geographic locations, color states (e.g. "off-white powder"), sequence mutations, solvent mixtures, health and safety warnings, Japanese katakana and numerous other entity types.

Advanced grammars for state-of-the-art Named Entity Recognition (NER)

253rd National Meeting (2017)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.