CollaboNet

CollaboNet enhances biomedical named entity recognition (BioNER) by integrating multiple deep neural network models trained on distinct annotated datasets to share information, reduce false positives, and disambiguate polysemous entity labels.

Key Features:

• Collaborative Model Architecture: Interconnected NER models are each trained on specific annotated datasets, allowing specialization on particular biomedical entity types.
• Dynamic Role Assignment: Models alternate between target and collaborator roles during training to exchange learned representations and reduce false positives and misclassifications.
• Handling Polysemy: Leveraging multiple datasets and cross-model collaboration to disambiguate polysemous biomedical entities across contexts.

Scientific Applications:

• Bio-entity relation extraction: Improves entity recognition precision for relation extraction tasks by reducing misclassifications.
• Biomedical text mining: Enhances downstream text mining and information extraction workflows that depend on accurate entity annotation.

Methodology:

Train multiple NER models on different annotated datasets for specific entity types, connect the models to share information, and alternate models between target and collaborator roles during training.