Characterizing Twitter Discussions About HPV Vaccines Using Topic Modeling and Community Detection

Area of Science:

• Public Health Surveillance
• Computational Social Science
• Computational Linguistics

Background:

• Online information spread influences health decisions and outcomes.
• Understanding online community dynamics is crucial for public health surveillance.

Purpose of the Study:

• To evaluate community structure and topic modeling for analyzing human papillomavirus (HPV) vaccine opinions on Twitter.
• To characterize opinion clusters related to HPV vaccines using computational methods.

Main Methods:

• Analyzed 285,417 tweets about HPV vaccines (Oct 2013-Oct 2015).
• Employed Latent Dirichlet Allocation and Dirichlet Multinomial Mixture for topic inference.
• Utilized Louvain and Infomap algorithms for community detection based on social connections.
• Assessed alignment between detected communities and inferred topics.

Main Results:

• Louvain community detection with Dirichlet Multinomial Mixture showed higher alignment between topics and user communities.
• Lower numbers of topics generally resulted in higher alignment values.
• Identified distinct clusters of tweets related to evidence/advocacy versus personal experiences regarding HPV vaccines.

Conclusions:

• Community detection combined with topic modeling is effective for public health opinion surveillance on social media.
• This methodology can identify online communities at risk of negative influence regarding public health interventions like HPV vaccines.