Adapting ChER for the recognition of chemical mentions in patents

Riza Batista-Navarro, Sophia AnaniadouResearch 2015

Abstract

ChER (Chemical Entity Recogniser) is a pipeline of natural language processing tools optimised for the recognition of chemical names in scientific abstracts. It formed the basis of our submissions to the previous edition of the CHEMDNER track in BioCreative IV, and was one of the top-performing systems both for the chemical document indexing (CDI) and chemical entity mention recognition (CEM) subtasks. As part of our contribution to the new chemical patents-focussed CHEMDNER track in BioCreative V, we adapted ChER for the recognition of chemical names in medicinal chemistry patent documents. As our previous study showed that the incorporation of chemical dictionary features brings about the most significant boost in performance, we focussed our current efforts on investigating the contribution of features drawn from various semantic resources, including RxNorm, US FDA’s Orange Book, Health Canada’s Drug Product Database and BioSemantics’ Chemical Patent Corpus. Results of our methods on the CHEMDNER test corpus demonstrate that the addition of features based on patent-specific dictionaries gives the most optimal performance on the Chemical Entity Mention in Patents (CEMP) task.
BibTeX
@inproceedings{batistanavarro2015adapting,
  author = {Riza Batista-Navarro and Sophia Ananiadou},
  title = {{Adapting ChER for the recognition of chemical mentions in patents}},
  booktitle = {Research},
  year = {2015},
}

Loading related content...

Figure