Evaluating and combining biomedical named entity recognition systems

Abstract

This paper is concerned with the evaluation of biomedical named entity recognition systems. We compare two such systems, one based on a Hidden Markov Model and one based on Conditional Random Fields and syntactic parsing. In our experiments we used automatically generated data as well as manually annotated material, including a new dataset which consists of biomedical full papers. Through our evaluation, we assess the strengths and weaknesses of the systems tested, as well as the datasets themselves in terms of the challenges they present to the systems.