A survey of 120 users of Blendle's news recommendation system investigates whether consumers want explanations for personalized article rankings. The study, by Maartje ter Hoeve, includes an A/B test measuring the impact of providing reasons on user open rates. Results indicate a desire for explanations but no strong preference for presentation format and no significant change in user engagement when reasons were provided.
Use Cases
- Analyzing user demand for explainability in recommendation systems based on survey responses.
- Evaluating the impact of explanation interfaces on user engagement metrics like open rates.
- Studying user preferences for different methods of presenting algorithmic explanations.
Strengths
- Survey includes 120 respondents, providing a quantitative basis for analysis.
- Methodology includes an A/B test to measure behavioral impact of explanations.
Limitations
- Row count and column-level documentation for the underlying data are unknown.
- Data may reflect bias inherent to the specific user base of the Blendle platform.
Provenance
- Source
- paperswithcode
- Collection Method
- User survey and A/B test conducted with users of Blendle's news recommendation system.