Presentation
From Transcripts to Personas: A Hybrid Computational-Qualitative Approach for Dementia Caregiver Research
SessionWednesday Poster Session
DescriptionPersonas are widely used in human factors to represent user archetypes, but their development typically relies on subjective researcher synthesis that can miss patterns in large qualitative datasets. This paper presents a hybrid computational-qualitative method that combines latent Dirichlet allocation (LDA) topic modeling with hierarchical clustering to empirically derive personas from interview data, demonstrated using 1,882 first-person "I-statements" extracted from interviews with 15 dementia caregivers.
Researchers manually decomposed deidentified transcripts into discrete I-statements, preserving caregiver voice at consistent granularity. LDA (k=15) generated probabilistic topic assignments across themes including Work-Care Balance, Fear and Protective Behaviors, Cultural Identity, and Financial Management. Hierarchical clustering (Ward's method) on participant-theme profiles produced five empirically-derived persona archetypes: Emotional Support Provider, Cultural Caregiver, Prepared Planner, Hands-On Caregiver, and Professional Caregiver. Researcher validation refined theme labels and reassigned low-confidence statements (probability < 0.30).
Critically, the empirical groupings diverged from researchers' initial intuitions, surfacing a Professional Caregiver persona the team had not anticipated, and clustering demographically similar participants into different archetypes based on thematic profiles. The method augments rather than replaces qualitative expertise, combining machine learning's pattern detection with the contextual judgment rigorous persona development requires.
Researchers manually decomposed deidentified transcripts into discrete I-statements, preserving caregiver voice at consistent granularity. LDA (k=15) generated probabilistic topic assignments across themes including Work-Care Balance, Fear and Protective Behaviors, Cultural Identity, and Financial Management. Hierarchical clustering (Ward's method) on participant-theme profiles produced five empirically-derived persona archetypes: Emotional Support Provider, Cultural Caregiver, Prepared Planner, Hands-On Caregiver, and Professional Caregiver. Researcher validation refined theme labels and reassigned low-confidence statements (probability < 0.30).
Critically, the empirical groupings diverged from researchers' initial intuitions, surfacing a Professional Caregiver persona the team had not anticipated, and clustering demographically similar participants into different archetypes based on thematic profiles. The method augments rather than replaces qualitative expertise, combining machine learning's pattern detection with the contextual judgment rigorous persona development requires.
Event Type
Poster
TimeWednesday, October 21st5:30pm - 6:30pm PDT
LocationSummit Pavilion
Similar Presentations