Semi-Supervised Recurrent Neural Network for Adverse Drug Reaction mention extraction

Area of Science:

• Computational linguistics
• Health informatics
• Machine learning

Background:

• Social media platforms offer vast reach for health information dissemination, making them valuable for public health monitoring and pharmacovigilance.
• Extracting Adverse Drug Reaction (ADR) mentions from social media presents challenges due to the informal and concise nature of the text.
• Current state-of-the-art methods, often deep learning models like LSTMs, require large annotated datasets, which are costly and difficult to obtain.

Purpose of the Study:

• To address the challenge of labeled data scarcity in ADR mention extraction from social media.
• To develop a novel semi-supervised learning model capable of leveraging abundant unlabeled social media data.
• To improve the efficiency and accuracy of pharmacovigilance through enhanced information extraction.

Main Methods:

• Proposed a novel semi-supervised learning model based on Recurrent Neural Networks (RNNs).
• The model is designed to effectively utilize large volumes of readily available unlabeled social media data.
• Compared the performance against traditional fully supervised learning methods.

Main Results:

• The proposed semi-supervised model demonstrated significant effectiveness in extracting Adverse Drug Reaction mentions.
• Achieved state-of-the-art performance in ADR mention extraction tasks.
• Outperformed a fully supervised learning baseline that relied on extensive manual annotation.

Conclusions:

• Semi-supervised learning offers a viable solution to the labeled data scarcity problem in medical information extraction from social media.
• The developed RNN-based model successfully leverages unlabeled data to enhance ADR mention extraction.
• This approach provides a more practical and efficient method for real-world pharmacovigilance applications.