Large-Scale Online Conversations About Public Health: Predicting Real-World Outcomes

Drug overdose remains one of the most severe public health challenges in the United States. Although online communities contain vast amounts of firsthand accounts, personal experiences, and peer-to-peer discussions about drug use, it is still unclear how these conversations can be analyzed to generate insights that support public health research. This dissertation addresses three core research questions: (1) How can we leverage Large Language Models to extract "gists" (causal language patterns) from decade-long online discussions? (2) What kinds of gists characterize how and why people discuss drugs, and how do these gists evolve over time? (3) Do these discussion themes align with, or predict, changes in real-world health outcomes (specifically overdose mortality)? To address these questions, Study 1 develops and validates an instruction-tuned large language model pipeline for extracting causal gists. Study 2 constructs a thematic taxonomy for these gists and uses multiple NLP models to classify and analyze how major discussion themes evolve over the ten years. Study 3 links these online themes to real-world health outcomes by applying time-series models, including autoregressive distributed lag (ARDL) analyses, to test whether changes in topic prevalence correspond with or precede trends in national, state-level, and drug-specific overdose mortality. Together, these studies demonstrate that large-scale online conversations contain structured, meaningful signals that reflect and anticipate real-world patterns in the overdose crisis.

Name thesis:degree_name

Doctor of Philosophy

Level thesis:degree_level

doctoral

Discipline thesis:degree_discipline

Computer Science & Applications

Department dc:contributor.department

Computer Science and Applications

Grantor dc:publisher

Virginia Tech

Year dc:date.issued

2026

Author dc:creator

• Ding, Xiaohan

Chair dc:contributor.committeechair

• Rho, Ha Rim

Committee members dc:contributor.committeemember

• Ramakrishnan, Narendran
• Lee, Sang Won
• Huang, Lifu
• North, Christopher L.

• Social Media Discourse
• Causal Language Extraction
• Large Language Models
• Public Health Surveillance

Statement dc:rights

• In Copyright

Licence dc:rights.uri

• http://rightsstatements.org/vocab/InC/1.0/

Language dc:language.iso

en

Dc Identifier Other

vt_gsexam:46825

OAI identifier oai:identifier

oai:vtechworks.lib.vt.edu:10919/143320