Mining Health Social Media with Sentiment Analysis

Area of Science:

• Computational Linguistics
• Health Informatics
• Social Media Analysis

Background:

• Online health communities are increasingly used for information exchange and social support.
• Analyzing user-generated content in these forums presents challenges due to data volume and complexity.

Purpose of the Study:

• To propose a framework for analyzing user-generated content in health communities.
• To develop an improved topic modeling technique for health-related discussions.

Main Methods:

• A three-phase framework involving medical term extraction, a modified Latent Dirichlet Allocation (LDA) called conLDA for clustering, and sentiment analysis.
• Virtual documents were created for each community question, incorporating extracted medical terms (conditions, symptoms, treatments, effectiveness, side effects).
• conLDA was employed to cluster virtual documents into conditional topics (C-topics) based on medical term distributions, followed by sentiment analysis (polarity, physiological, psychological).

Main Results:

• The proposed conLDA method demonstrated superior performance compared to the original LDA.
• conLDA effectively clustered semantically related medical terms and questions together.
• The resulting C-topics were more coherent and thematic than those generated by standard LDA.

Conclusions:

• The conLDA framework provides a robust method for analyzing health community data.
• Sentiment analysis of clustered C-topics offers valuable insights for patients, caregivers, and doctors.
• This approach enhances understanding of user-generated health information and experiences.