TFEBioNER: A New Method for Biomedical Named Entity Recognition Based on Table Filling for Feature Extraction
Lixing Song, Lin Zhang · 2024
Named Entity Recognition (NER) is a fundamental task in Natural Language Processing (NLP) and is important for information extraction. It is capable of automatically identifying special information and entities from texts. In recent years, biomedicine has become a research hotspot in the NLP field. With the rapid growth of medical literature, huge amount of useful information is stored in the text. Biomedical named entity recognition(BioNER) can extract key information from massive biomedical data and help researchers in related fields. However, the complexity of biomedicine and the lack of attention to nested named entity recognition in current models have led to difficulties in identifying BioNER. To address this problem, a framework for the identification of named entities in biomedical based on the table filling for feature extraction is raised in this paper. which can simplify the data annotation problem, enhance the entity boundary modelling capability, and also provides more powerful feature learning capabilities using a biaffine attention model that incorporates Transformer and CNN. It addresses both nested named entities and flattened named entities. Our model achieves F1 scores of 80.10%, 89.81%, 94.61%, 90.42%, and 91.46% on five corpora, GENIA, BC2GM, BC5CDR-chem, BC5CDR-disease, and NCBI-disease. The results show that this model based on table filling and feature extraction achieves competitive results.