Visualizing social media discussions of policy-relevant issues surrounding the opioid crisis in the United States
More details
Hide details
Central Michigan University, United States
Northwood University, United States
Publication date: 2023-04-26
Popul. Med. 2023;5(Supplement):A1885
Background and Objective:
We present a novel method of network visualization with computer-assisted text processing. What do visualizations add, above and beyond a qualitative study, to our understanding of social media users’ policy discussions?

We apply a network analysis and natural language processing to the corpus of 8,761 manually-coded social media user comments on opioid crisis-related videos (N=20) on CNN and Fox News YouTube channels. Manual codes on pain patients’ experiences and crisis solutions are combined with computer-scored social and psychological states from Linguistic Inquiry Word Count (LIWC-22). VOSviewer visualizations of opioid-relevant terms extracted from YouTube comments are overlayed with manual codes and computer-generated LIWC scores.

Our colorful map – a network of 196 interconnected terms extracted from 8,761 comments included three thematic clusters of terms: 1) patients’ stories of pain management/relief/addiction; 2) cannabis legalization; and 3) organizations/entities involved in the opioid crisis. Manually-coded and computer-generated overlays indicate some overlap. Cluster 1 has manually-coded pain patients’ stories and stigma and also scores high on several LIWC dimensions: 1st person pronouns, feeling, and emotion. User-proposed solutions to the opioid crisis (represented by such terms as kratom, insurance, and suboxone) are concentrated around cluster 2. Cluster 2 has high audience engagement (counts of likes/comments per view) and above average LIWC scores for negation, cause-and-effect cognition, and discrepancy cognition. Legal solutions (manually-coded) and power drive (LIWC scores) are high for terms in Cluster 3.

Networks of terms that frequently occur across multiple comments add value to the manual codes from content analyses. To make data analysis more efficient, some human coding may be substituted for automated text scoring to identify discussion areas with specific linguistic features. Most important, eye-catching term networks and LIWC overlays help researchers visualize and communicate about policy issues, as they are deliberated in social media.

Journals System - logo
Scroll to top