An adverse drug effect mentions extraction method based on weighted online recurrent extreme learning machine

Area of Science:

• Biomedical Text Mining
• Pharmacovigilance
• Machine Learning

Background:

• Automatic extraction of adverse drug effect (ADE) mentions from biomedical texts is crucial for pharmacovigilance.
• Current deep learning methods struggle with accurately identifying the boundaries of ADE mentions.
• A novel Weighted Online Recurrent Extreme Learning Machine (WOR-ELM) is proposed to address this limitation.

Purpose of the Study:

• To develop and evaluate a WOR-ELM based method for improved ADE mention extraction.
• To enhance the accuracy of identifying both the boundaries and types of ADE mentions in biomedical texts.

Main Methods:

• A two-stage approach using WOR-ELM for span detection and ADE mention classification.
• Utilizing a combination of character-level and word-level embeddings as features.
• Character-level embeddings are generated via a modified online recurrent extreme learning machine; word-level embeddings are derived from a pre-trained model.

Main Results:

• The proposed WOR-ELM method achieved an F-score of 87.5% on a standard ADE corpus.
• This performance surpasses current state-of-the-art methods in ADE mention extraction.
• Experiments validated the effectiveness of the combined feature approach and the IOU segment representation.

Conclusions:

• The WOR-ELM method significantly improves performance for adverse drug effect mention extraction.
• Integrating word-level and character-level embeddings enhances the efficacy of WOR-ELM.
• The use of IOU segments is beneficial for representing ADE mentions.