Chemical entity extraction using pattern based bootstrapping

Abstract

This paper describes a Named Entity Recognition system for identifying the chemical entities in free text documents. The
existing approaches use dictionary and rule-based methods to extract the chemical entities in text. Rather, we have evolved
a chemical NER system, based on semi-supervised machine learning approach called bootstrapping technique which
involves minimal human oversight. To better the accuracy, scoring has been performed for both entities and forms. In
accession to that, a 10-fold cross validation has been performed for different lots of testing and preparation data. The
scheme has been proven on a corpus consisting of 10000 chemical research articles. The outcomes demonstrate the
feasibility of the bootstrapping technique and the valuation demonstrates that our method achieved approximately the F
score of 90.94%. This approach is also compared with the existing approach to extract IUPAC chemical names and has
achieved better outcomes.