A Simple but Useful Multi-corpus Transferring Method for Biomedical Named Entity Recognition

In the biomedical field, understanding biomedical texts requires domain-specific knowledge, and the annotation of biomedical texts often requires a lot of human involvement, which makes annotation costly and time-consuming. Therefore, how to effectively use the existing public-available corpus resources is meaningful for performance improvement in biomedical NLP. In this paper, we present a multi-corpus transferring method for biomedical named entity recognition task. We clearly define the target criteria by adding two artificial tokens at the beginning and end of each input sentence. A comprehensive evaluation was conducted to prove the efficiency of our method. The results illustrate that the multi-corpus transferring method could benefit the current methods and improve its performance. Our method provides a potential solution for biomedical NER enhancement from data perspective, and it could further improve biomedical information extraction with the help of increasingly public available corpus.