Chemical named entity recognition: Improving recall using a comprehensive list of lexical features

Abstract

As the number of published scientific papers grows everyday, there is also an increasing necessity for automated named entity recognition (NER) systems capable of identifying relevant entities mentioned in a given text, such as chemical entities. Since high precision values are crucial to deliver useful results, we developed a NER method, Identifying Chemical Entities (ICE), which was tuned for precision. Thus, ICE achieved the second highest precision value in the BioCreative IV CHEMDNER task, but with significant low recall values. However, this paper shows how the use of simple lexical features was able to improve the recall of ICE while maintaining high levels of precision. Using a selection of the best features tested, ICE obtained a best recall of 27.2% for a precision of 92.4%.