A neural network-based joint learning approach for biomedical entity and relation extraction from biomedical...

Area of Science:

• Biomedical Natural Language Processing
• Computational Biology
• Bioinformatics

Background:

• Joint entity and relation extraction models show promise in general domains.
• Traditional pipelined methods struggle with overlapping relations common in biomedical text.
• Existing joint models are often unsuitable for the complexities of biomedical data.

Purpose of the Study:

• To develop an effective joint learning approach for biomedical entity and relation extraction.
• To address the challenge of overlapping relations in biomedical text.
• To improve the state-of-the-art in biomedical relation extraction.

Main Methods:

• Proposed a novel tagging scheme to accommodate overlapping relations.
• Developed an Attention-based Bidirectional Long Short-Term Memory Conditional Random Field (Att-BiLSTM-CRF) model.
• Utilized ELMo (Embeddings from Language Models) contextualized representations pre-trained on biomedical text.

Main Results:

• The proposed method significantly enhances the performance of overlapping relation extraction.
• Achieved state-of-the-art results on benchmark biomedical corpora.
• Demonstrated the effectiveness of the novel tagging scheme and Att-BiLSTM-CRF model.

Conclusions:

• The neural network-based joint learning approach is highly effective for biomedical entity and relation extraction.
• The novel tagging scheme and model architecture successfully handle overlapping relations.
• Contextualized biomedical ELMo representations further boost extraction performance.