A document level neural model integrated domain knowledge for chemical-induced disease relations

Area of Science:

• Biomedical Natural Language Processing
• Computational Linguistics
• Bioinformatics

Background:

• Existing chemical-induced disease (CID) relation recognition systems often overlook the impact of diverse knowledge sources.
• Current neural network models typically focus on sentence or mention-level analysis, limiting their scope for document-wide context.

Purpose of the Study:

• To develop a document-level neural model for enhanced CID relation extraction from biomedical literature.
• To effectively integrate and weigh domain knowledge to improve the accuracy of CID relation identification.

Main Methods:

• Proposed a document-level neural network architecture incorporating domain knowledge.
• Introduced a knowledge attention mechanism to differentiate the influence of various knowledge components on CID pairs.
• Integrated textual and knowledge representations for final CID classification using a softmax classifier.

Main Results:

• The proposed model achieved strong performance on the BioCreative V chemical-disease relation corpus.
• Demonstrated superior results compared to existing state-of-the-art systems in CID relation extraction.
• Validated the effectiveness of integrating domain knowledge for improved accuracy.

Conclusions:

• The developed attention mechanism on domain knowledge is crucial for accurately assessing the impact of different knowledge sources on CID relation judgment.
• The document-level approach offers a more comprehensive understanding of CID relations within biomedical articles.