Date of Publication :5th January 2025

Abstract: In healthcare and biomedical research, enhancing the understanding of medical texts through Named Entity Recognition (NER) is vital for identifying key terms such as diseases, drugs, proteins, and genes. Transfer learning models like BERT have been instrumental in improving the extraction of domain- specific information. NER is critical for supporting medical decision-making and research. However, challenges persist, such as the absence of standardized datasets, limited computational resources, and the inherent complexity of medical data. Advanced techniques like Bidirectional Long Short-Term Memory (BiLSTM), Conditional Random Fields (CRF), and Multi-Task Learning (MTL) are employed to improve NER performance. Preprocessing involves tokenization, annotation, and embedding, while post-processing refines model predictions. The BINER model has demonstrated superior F1-scores for disease detection and has outperformed BioBERT in recognizing proteins and genes, showcasing advancements in biomedical NER applications.

Reference :