Event trigger identification for biomedical events extraction using domain knowledge.

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Biomedical event extraction aims to identify molecular interactions in research articles, crucial for understanding drug reactions and diseases.
• Event trigger identification, a key step, classifies words indicating molecular event types within sentences.
• Effective feature sets are vital for accurate event trigger identification in biomedical text.

Purpose of the Study:

• To propose a novel framework for improving biomedical event trigger identification.
• To enhance word features by embedding domain knowledge from a large text corpus using neural language modeling.
• To integrate these enhanced features with syntactic and semantic context features for a more discriminative classifier.

Main Methods:

• Learning biomedical domain knowledge from a large Medline text corpus.
• Embedding this knowledge into word features using neural language modeling.
• Combining embedded features with syntactic and semantic context features via multiple kernel learning for classifier training.

Main Results:

• The proposed framework achieved over 2.5% improvement in F-score compared to state-of-the-art methods.
• Experimental results on a golden standard corpus demonstrate the framework's effectiveness.
• The integration of domain-specific embeddings and multiple kernel learning significantly boosts performance.

Conclusions:

• The novel framework effectively enhances biomedical event trigger identification.
• Neural language modeling for domain knowledge embedding is a powerful technique.
• The proposed approach represents a significant advancement in biomedical text analysis.