User-Centric Metrics and Satisfaction
What You'll Learn
You'll learn how to measure what really matters: whether users are happy with your recommendations through session quality, engagement patterns, and direct feedback.
Beyond Algorithmic Perfection
You can have perfect NDCG scores and excellent coverage metrics, but if users abandon your platform, something's wrong. User-centric metrics focus on human experience rather than algorithmic accuracy alone.
Think of it like restaurant reviews: a chef might execute every technique flawlessly (high precision), but if diners don't enjoy the meal or return, the restaurant fails. Similarly, recommender systems must optimize for actual user satisfaction.
Core User-Centric Measures
Session Quality
Track how users behave during their visit: Did they find what they wanted quickly? Did they explore multiple recommendations? A high-quality session shows smooth navigation and successful discovery, not frustrated clicking or immediate exits.
Engagement Time
Measure meaningful interaction duration. Unlike raw "time on site," this distinguishes between productive engagement (watching videos, reading articles) and frustrated searching. If engagement time increases with positive outcomes (conversions, completions), your recommendations work.
Satisfaction Surveys
Direct feedback through ratings, thumbs up/down, or periodic surveys asking "How satisfied are you with these recommendations?" While subjective, this captures what metrics miss: user perception, trust, and emotional response.
Why This Matters
You might have a model with excellent offline MAP scores, but users feel overwhelmed by too many options or find recommendations repetitive. Satisfaction surveys would catch this disconnect between technical performance and real experience.
Similarly, high CTR doesn't mean users are satisfied—they might click from curiosity but leave disappointed. Session quality reveals whether clicks led to valuable outcomes.
Key Takeaway: User-centric metrics like session quality, engagement time, and satisfaction surveys bridge the gap between algorithmic performance and real-world user happiness—the ultimate success measure for any recommender system.