Evaluating and Comparing Biomedical Term Identification Systems

In this paper, we propose a term identification system using conditional random fields (CRFs) on two biomedical datasets. Through employing several sets of experiments, we make a comprehensive investigation for different types of features. The final experimental results reflect that with carefully designed features i.e., adding not only the individual and dynamic features but also the combinational features, our system can identify biomedical terms with fairly high accuracy on both datasets, compared with other top systems already published in the literature.