Step-Up Program, University of Utah – School of Computing
Prachi Aswani | Collaborators: Di Wang (Postdoctoral Fellow), Marina Kogan (Assistant Professor)
Overview:
Public responses to crisis communication shape trust, compliance, and perception during health emergencies. My research examined how message topics (risk vs. protective action) and sources (government vs. politicians) influence reactions and discussions on Twitter during the COVID-19 pandemic.
Methods:
Dataset: 1.27M tweet replies related to COVID-19 crisis communication.
Classification:
Labeled 2,000 tweet replies as reaction (emotional/expressive) vs. discussion (analytical/informative).
Built a Gaussian Process Classifier (62% accuracy).
Improved performance using a LLAMA-based classifier with 90% accuracy via one-shot learning.
Analysis Dimensions:
Source: Government (federal, state, party-led) vs. Politicians (Democrat, Republican).
Topic: Risk (primary, secondary) vs. Protective Action (primary, secondary).
Key Findings:
Risk-related topics generated more discussion, while protective action messages triggered more emotional reactions.
Politicians’ tweets prompted more money-related responses compared to government sources.
Developed interactive visualizations (network graphs, tweet walls) to show how public engagement patterns differ by source and topic.
Impact:
This work provides a framework to understand public engagement during crises, helping communicators tailor messaging strategies for greater clarity, trust, and effectiveness.