Integrating Language Model and Reading Control Gate in BLSTM-CRF for Biomedical Named Entity Recognition

Area of Science:

• Computational Biology
• Bioinformatics
• Natural Language Processing

Background:

• Biomedical Named Entity Recognition (Bio-NER) is crucial for biomedical text mining.
• Current neural network methods for Bio-NER overlook sentence-level semantic and general linguistic features.
• Complex hand-designed features are often avoided in mainstream NER approaches.

Purpose of the Study:

• To propose a novel Long Short Term Memory (LSTM) Networks model for Bio-NER.
• To integrate sentence-level semantic information and general linguistic features into Bio-NER models.
• To improve the accuracy and performance of Bio-NER systems.

Main Methods:

• Developed a Long Short Term Memory (LSTM) Networks model integrating a language model and a sentence-level reading control gate (LS-BLSTM-CRF).
• Incorporated a sentence-level reading control gate (SC) to capture implicit sentence meaning.
• Utilized character-level embeddings to handle out-of-vocabulary words.

Main Results:

• The proposed LS-BLSTM-CRF model achieved an F-score of 89.94% on the BioCreative II GM corpus.
• The model outperformed all existing state-of-the-art Bio-NER systems.
• Achieved a 1.33% improvement over the best-performing neural networks.

Conclusions:

• The novel LS-BLSTM-CRF model effectively integrates sentence-level semantics and language model features for Bio-NER.
• The proposed approach significantly enhances Bio-NER performance compared to current methods.
• Character-level embeddings are beneficial for handling unseen words in biomedical text mining.