Metrics question

How do you find the top 10 brunch places in the USA, assuming you have all the Yelp ratings and review data?

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Tests defining a ranking methodology from raw rating and review data, requiring the candidate to go beyond a simple average-rating sort.

Clarify the definition of top: highest quality, most popular, or best value, since the ranking method differs.
State the naive approach and its flaw: sorting by raw average rating fails because a place with five 5-star reviews would outrank one with five hundred 4.8-star reviews.
Propose a Bayesian or confidence-adjusted average that weights rating by review volume, pulling low-volume ratings toward the category mean.
Add a recency weight, since a place's quality can change, so older reviews should count less than recent ones.
Filter to brunch-specific signals: require reviews mentioning brunch-relevant terms or filter by the brunch category tag if Yelp data supports it.
Combine into a single ranking score, then take the top 10 by that score, sanity-checking the list against well-known, obviously good brunch spots.

Identifies the exact flaw in a naive average-rating sort, showing understanding of small-sample bias.
Proposes a specific, named technique, a Bayesian-adjusted or confidence-weighted average, rather than a vague weight by reviews.
Adds recency weighting, addressing that restaurant quality changes over time, which a static average misses.
Includes a sanity-check step against known-good results, showing practical rigor beyond pure math.

Sorting by raw average rating without addressing the small-sample-size bias problem.
Ignoring recency, treating a five-year-old review the same as a recent one.
Not filtering specifically for brunch relevance before ranking.

Chapters of the AI PM course that teach what this question tests.