ADENER: A syntax-augmented grid-tagging model for Adverse Drug Event extraction in social media

Area of Science:

• Natural Language Processing
• Pharmacovigilance
• Computational Linguistics

Background:

• Extracting Adverse Drug Events (ADEs) from social media is challenging due to semantic similarities and complex mention patterns.
• Existing methods struggle with the noisy and irregular nature of social media text.

Purpose of the Study:

• To develop a robust model, ADENER, for accurate ADE extraction from social media.
• To address challenges like semantic similarity, overlapping, and discontinuous mentions.

Main Methods:

• ADENER employs a grid-tagging architecture for multi-label word-pair classification.
• It utilizes a convolutional capture layer for feature fusion and dependency analysis.
• A syntactic affine layer integrates path-level dependency information for enhanced logic understanding.

Main Results:

• ADENER achieved high F1 scores (74.64% on CADEC, 77.97% on CADECv2, 61.73% on SMM4H).
• The model outperformed existing baseline models on multiple datasets.
• It demonstrated effectiveness in handling noisy social media data and complex ADE mentions.

Conclusions:

• ADENER provides a unified solution for ADE extraction from social media.
• The model accurately distinguishes ADEs from therapeutic indications and handles various mention types.
• This approach enhances pharmacovigilance and real-world drug safety monitoring.