Bidirectional Long Short-Term Memory-Based Detection of Adverse Drug Reaction Posts Using Korean Social Networking...

Area of Science:

• Pharmacovigilance
• Computational Linguistics
• Artificial Intelligence

Background:

• Social networking services (SNS) generate vast amounts of user data, offering insights into modern life.
• Previous research has utilized SNS data for drug information extraction, highlighting the potential for pharmacovigilance.
• Early detection of adverse drug reactions (ADRs) is crucial for public health, necessitating advanced surveillance systems.

Purpose of the Study:

• To develop a deep learning model for classifying posts related to adverse drug reactions (ADRs) using Korean social media data.
• To investigate the efficacy of recurrent neural network-based models in identifying drug-induced adverse events from user-generated content.
• To establish a feasible process for leveraging social data in automated drug safety monitoring.

Main Methods:

• Collected Korean SNS data related to ketoprofen, a cautionary drug, and applied natural language processing (NLP) techniques.
• Filtered relevant posts using association analysis between drug names and ADR keywords, followed by manual labeling to create training data.
• Developed a Bidirectional Long Short-Term Memory (Bi-LSTM) classification model using word2vec embeddings and validated its performance against other machine learning models.

Main Results:

• Generated Korean-specific lexicons for drug names, ADRs, and stop words to enhance model accuracy.
• The developed Bi-LSTM model achieved 85% accuracy for ketoprofen and 80% for aceclofenac in classifying ADR posts.
• The study confirmed the feasibility of using social data for automated ADR detection and monitoring.

Conclusions:

• A novel process for developing ADR post classification models using SNS data was proposed.
• The model effectively filters high-quality data by identifying posts with known ADR patterns.
• This research validates the potential of social data analytics for automated pharmacovigilance.