Presentation: 2024 ND EPSCoR Annual conference
November 21, 2024, Alerus Center, Grand Forks, North Dakota
Nested Named Entity Recognition using Multilayer BERT-based Model
Benu
Bansal
Doctoral Student
University of North Dakota
Co-authors: Hasin Rehana, PhD student, UND; Nur Bengisu Cam, Department of Computer Engineering, Bogazici University; Jie Zheng, Unit for Laboratory Animal Medicine, Department of Microbiology and Immunology, University of Michigan; Yongqun He, Center for Computational Medicine and Bioinformatics, University of Michigan; Arzucan Özgür, Department of Computer Engineering, Bogazici University; Junguk Hur, Associate Professor, Department of Biomedical Sciences, School of Medicine and Health Sciences, UND
Session
Concurrent Presentation Session 1
In natural language processing, named entity recognition (NER) is essential for identifying and categorizing entities in text. The biomedical field poses significant challenges due to complex language and nested entities. This paper presents an innovative approach to Nested NER using a multilayer BERT-based model, specifically leveraging pretrained PubMedBERT. Our model effectively handles nested entities by combining robust contextual embeddings with a multilayer tagging process, enabling precise differentiation of overlapping items common in biomedical literature. We evaluated the Multilayer NER Model (MultilayerNERModel) on the BioNNE English Dataset, part of the BioASQ competition. The results indicate that the multilayer approach enhances the model's ability to detect nested entities, achieving the highest performance in the English-oriented track with an F1 score of 67.30% and a macro F1 score of 56.36%. These findings underscore the effectiveness of a multilayer strategy in Nested NER tasks, especially within the biomedical domain, further enhanced by UMLS dictionaries.