SPBERE: Boosting span-based pipeline biomedical entity and relation extraction via entity information.

Area of Science:

• Biomedical text mining
• Natural Language Processing
• Bioinformatics

Background:

• Triplet extraction is crucial for understanding biomedical relationships.
• Existing joint methods struggle with overlapping entities and relations common in biomedical texts.
• Pipeline approaches can suffer from error propagation.

Purpose of the Study:

• To propose a novel end-to-end pipeline method for biomedical triplet extraction.
• To effectively detect overlapping triplets and relations.
• To improve the accuracy of relation extraction in biomedical literature.

Main Methods:

• A span-based detection strategy enumerates candidate spans and entity pairs to identify overlapping triplets.
• Separate entity and relation models capture contextualized representations.
• Entity information from the entity model is fused early into the relation model without external knowledge.

Main Results:

• The proposed method achieved absolute F1-score improvements of 3.5%-3.7% in relation extraction on DDI and CHEMPROT datasets.
• Demonstrated the effectiveness of the span-based approach for overlapping triplet detection.
• Showcased the benefits of contextualized representations and early entity information fusion.

Conclusions:

• The novel pipeline method enhances biomedical triplet extraction by effectively handling overlapping entities and relations.
• Span-based detection and early fusion of entity information are key to improving relation extraction performance.
• The approach offers a significant advancement for biomedical text mining tasks.