AIONER

AIONER performs biomedical named entity recognition (BioNER) to identify and classify genes, diseases, chemicals, and other biomedical entities in text for downstream information extraction, data mining, and question answering.

Key Features:

• All-in-One (AIO) Scheme: Leverages external annotated resources to enhance accuracy and stability while addressing overfitting and limited generalizability.
• Deep Learning Integration: Employs advanced neural network architectures for high-precision processing of complex biomedical texts.
• General-Purpose Multi-Entity Recognition: Recognizes multiple entity categories (genes, diseases, chemicals, and others) simultaneously rather than focusing on single entity types.
• Scalability and Efficiency: Demonstrated capability to process large-scale datasets such as the entire PubMed database.

Scientific Applications:

• Enhanced Information Extraction: Improves extraction of relevant biomedical facts by accurately identifying diverse entity types in literature.
• Improved Question Answering Systems: Supports biomedical question answering by providing precise entity recognition across literature sources.
• Robustness to Unseen Entity Types: Exhibits adaptability for recognizing entity types not previously encountered during training.

Methodology:

Integrates external annotated datasets to supplement training data, mitigating data scarcity and manual-labeling costs and reducing overfitting to enhance generalizability; evaluated across 14 BioNER benchmark tasks and compared against state-of-the-art methods including multi-task learning approaches.