Two-perspective Biomedical Named Entity Recognition with Weakly Labeled Data Correction

Abstract

Biomedical Named Entity Recognition (BioNER) is one of the most essential tasks in biomedical information extraction. Previous studies suffer from inadequate annotation datasets, especially the limited knowledge inside. This paper proposes a two-perspective named entity recognition method with Weakly Labeled (WL) data correction. Firstly, from the perspective of coverage and accuracy, we utilize PubTator and multiple knowledge bases to construct two large-scale WL datasets, which are then revised by their corresponding label correction models respectively, obtaining two high-quality datasets. Finally, we compress the knowledge in the two datasets into a BioNER model with partial label integrating. Our approach achieves new state-of-the-art performances on three BioNER datasets.