66. How would you evaluate recommendations for a new item or user with little interaction history?
MediumMachine learning basics
Technical
Show model answer
Create evaluation slices that reflect genuinely limited information rather than letting future interactions leak into a retrospective test. Compare with usable baselines, such as popular or content-based recommendations, while respecting available user preferences.
Measure coverage and useful outcomes for those slices, not only accuracy among established users. Track how performance changes as feedback accumulates, and avoid inferring preferences from sensitive characteristics or treating missing history as a negative signal.
What interviewers listen for
- Simulate the actual information limit
- Use practical cold-start baselines
- Measure coverage as well as quality