Knowledge Enhanced Fine-grained Biomedical Named Entity Recognition

Abstract

While Named Entity Recognition (NER) in universal domains has made significant progress, there is still room for improvement in Biomedical Named Entity Recognition (BioNER) as it pertains to specific domains. Unlike the universal domain, NER in specialized domains requires a more detailed categorization of entities. Based on the idea by incorporating knowledge, we propose a method that is better suited for this task. Our method initially employs comparative learning to thoroughly grasp entity category information as additional knowledge. Subsequently, we seamlessly integrate this category information into the NER process. By enhancing knowledge through the incorporation of category information into entity extraction, our method improves the F1 score by 1% compared to existing methods.