Recognition of chemical names in Chinese texts

Abstract

Chemical names recognition is a critical task for search and mining in some science publications and patents. However, most research on chemical names recognition has focused on English texts, e.g., MEDLINE abstracts. This paper is concerned with recognition of chemical substance names in Chinese text, which is regarded as a sequence tagging problem under Conditional Random Field (CRF) framework. In order to achieve a better performance, we make an empirical exploration of several relative parameters, including tagging unit, intervals of feature values, and feature sets. We show that there is a significant variance in performance as different parameters are selected. Based our experiment data, we give some feasibility analysis for further research. © 2012 Springer-Verlag GmbH.