Disease named entity recognition using long-short dependencies.

Area of Science:

• Biomedical text mining
• Natural Language Processing
• Deep Learning

Background:

• Automatic disease named entity recognition is crucial for biomedical text mining.
• Traditional handcrafted features have limitations in capturing complex linguistic patterns.
• Existing deep learning models often overlook short-range dependencies.

Purpose of the Study:

• To develop an advanced deep learning model for improved disease named entity extraction.
• To address the challenge of imbalanced datasets in biomedical named entity recognition.
• To leverage both local and long-range dependencies for enhanced accuracy.

Main Methods:

• A two-stage deep neural network architecture combining Convolutional Neural Networks (CNNs) for local dependencies and Bidirectional Recurrent Neural Networks (RNNs) for long dependencies.
• Development of a novel Part-of-Speech (POS)-based tagging schema to mitigate dataset imbalance.
• Integration of multiple word embedding techniques and morphological features.

Main Results:

• The proposed model achieved an 85.59 F1-score on the NCBI dataset, outperforming current state-of-the-art methods.
• Demonstrated the effectiveness of capturing both short and long-range dependencies.
• Highlighted the benefits of combining diverse word embeddings and morphological features.

Conclusions:

• The novel two-stage deep learning approach significantly enhances disease named entity recognition.
• The POS-based tagging schema effectively addresses dataset imbalance issues.
• Combining various linguistic features and network architectures offers a promising direction for biomedical NLP tasks.