Description
Generate and periodically refresh personas from aggregated platform analytics by matrix factorisation, then enrich them with names, photos and quotes and track how stakeholders use them.
APG builds personas from audience interaction data collected through YouTube, Facebook and Google Analytics APIs, keyed by gender, age group and country. A demographic-group by content matrix is decomposed with non-negative matrix factorisation to find latent interaction patterns. The most impactful demographic group for each pattern becomes a skeletal persona, enriched with a census-based name, photo, description, topics, quotes from comments, content and reach. Personas regenerate as data change, keeping them linked to underlying numbers. Design principles include consistency, relevance, non-offensiveness, authenticity and context. Persona analytics extends this by logging stakeholders' mouse movements and webcam gaze on persona profiles to see which elements are used. Limitations: aggregates can hide minorities, demographic buckets drive the results, and automatically chosen text and images need filtering.
Attributes
- Discipline
- Computational & Machine Learning
- Loss type
- Siloed
- Stage
- Discovery, Definition, Post-launch
- Data
- Quantitative
- Effort
- Weeks
Steps
- Collect aggregated interaction data through platform or analytics APIs.
- Build a demographic-group by content (or behaviour) matrix.
- Apply non-negative matrix factorisation to extract latent interaction patterns; choose the number of patterns by interpretability and fit.
- Select the most impactful demographic group for each pattern to create skeletal personas.
- Enrich each persona with a name, photo, description, topics, quotes, content and reach, filtering offensive or inauthentic items.
- Regenerate personas on a schedule as data change.
- Optionally log stakeholders' interaction with persona profiles (mouse, gaze) to learn which elements they use.
Sources
- Quant UX Con 2022 Proceedings (Chapman et al., 2022)
- Persona Focused User Design (Nielsen, 2019)
Also known as
- APG
- Data-driven personas
- Data-driven personas via NMF
- Persona analytics
- Automatic persona generation and persona analytics
Categories
- Analytics & Behaviour
- Computational & Machine Learning