NEROC : Named Entity Recognizer of Chemicals

We describe a pipeline system, Named Entity Recognizer of Chemicals (NEROC), that aims to identify chemical entities mentioned in free texts. The system is based on a machine learning approach, a Conditional Random Field (CRF), and a selection of feature sets that are used to capture specific characteristics of chemical named entities. In this paper, we report results that produced by CRF models trained with the training dataset (3500 PubMed abstracts), and the best performance for chemical named entity recognition (NER), as assessed on the development data, is precision of 0.81, recall of 0.72, and F-score of 0.76. For our final system, training is based on 7,000 PubMed abstracts released for a task of BioCreative IV.