JMIR Publications

ABSTRACT

Background:

Adverse drug reactions (ADRs) are an important public health issue. Traditional ADR reporting systems often suffer from reporting delays. These delays hinder the timely identification of ADRs. In China, more and more people are sharing information about health issues and drug use on online platforms, providing an opportunity for real-time detection of ADRs.

Objective:

To address the critical public health challenge of ADRs and the limitations of traditional pharmacovigilance systems, this study explores the potential of user-generated content on Chinese social media platforms for early ADR identification. The research aims to develop a machine learning-based method to detect potential ADRs from unstructured social media text.

Methods:

The study collected and tagged 75,000 posts from three major Chinese social media platforms: Weibo, Zhihu, and Xiaohongshu. The posts were categorized into ADR-related and non-ADR content. Nine machine learning techniques, including Support Vector Classification (SVC), Logistic Regression (LR), Random Forest (RF), Decision Tree (DT), XGBoost (XGB), and Ensemble models, were evaluated. Preprocessing and feature extraction were performed using the Term Frequency-Inverse Document Frequency (TF-IDF) method.

Results:

The XGB model demonstrated the highest performance, achieving F1-scores of 91.4% on Weibo, 94.3% on Zhihu, and 93.0% on Xiaohongshu. The results indicate that machine learning, particularly ensemble models, can effectively detect potential ADRs from unstructured social media text.

Conclusions:

Sentiment analysis of Chinese social media data can complement traditional pharmacovigilance systems, enabling faster and more accurate detection of drug safety issues. This study highlights the potential of leveraging user-generated content for early identification of ADRs, suggesting a promising direction for future pharmacovigilance efforts.