Interview Practice Questions

Which metrics do we need to measure?

Which is more suitable: Recall or Precision@K?

How will we measure metrics on the validation set?

How do we address the cold-start problem?

What validation set can we use?

How will we experiment with different features?

How will we handle deployment and A/B testing?

How will the system interact with the product backend?

Which database would you suggest for the cache?

What time to live would you suggest for precomputed recommendations?

What should we do if the A/B test does not produce a statistically significant improvement?

How can we assess a difference in offline metrics?

How can we solve the cold-start problem?

What can we do if performance suddenly drops?

How can we scale the service?

How can we cache previous recommendations?

Describe a common way to implement a listening mechanism for a mobile application.

What feature hypotheses can you suggest?

Do you think text descriptions would work well in the candidate generator?

What should we do if the model starts producing irrelevant recommendations?