Pipelined biomedical event extraction rivaling joint learning.

Area of Science:

• Biomedical informatics
• Natural Language Processing
• Computational Biology

Background:

• Biomedical event extraction aims to identify triggers and arguments within biomedical text.
• Traditional pipelined approaches face challenges in capturing complex event semantics.
• Existing methods often struggle with accurately constructing events, especially those involving multiple participants.

Purpose of the Study:

• To propose an n-ary relation extraction method utilizing the BERT pre-training model.
• To enhance the extraction of Binding events by capturing semantic context and participant information.
• To improve the overall accuracy and efficiency of biomedical event extraction.

Main Methods:

• Developed an n-ary relation extraction model leveraging BERT.
• Applied the model to construct Binding events from biomedical literature.
• Evaluated the method on the GE11 and GE13 corpora from the BioNLP shared task.

Main Results:

• Achieved F1 scores of 63.14% on the GE11 corpus and 59.40% on the GE13 corpus.
• Demonstrated significant improvement in the performance of Binding event extraction.
• The proposed method shows competitive or superior performance compared to current joint learning methods.

Conclusions:

• The BERT-based n-ary relation extraction method effectively captures semantic information for Binding events.
• This approach offers a promising alternative to traditional pipelined and current joint learning methods.
• The study highlights the potential of pre-trained language models for advanced biomedical text mining.