Design Yelp: nearby business search and reviews

Design a local business review service like Yelp. Users find businesses near a location by category and rating, read and write reviews, and upload photos. Reads far outnumber writes.

Requirements

• Functional: search businesses near a point or in a map area, filtered by category, rating, price and open now; business pages with details, ratings, reviews and photos; write and edit reviews; upload photos; business owners update their details.
• Non-functional: low-latency search; read-heavy traffic; tens or hundreds of millions of businesses and reviews; new reviews visible within minutes; ratings resistant to fake reviews.

Clarifying questions

• How many businesses, users and searches per second?
• Is search by radius, by visible map area, or both?
• How fresh must ratings and new reviews be?
• How are results ranked: distance, rating, relevance?
• Do we handle fake review detection in scope?

What a strong answer covers

• A data model: businesses (with location), reviews, photos, users; average rating and review count kept on the business and updated as reviews arrive.
• A geospatial index (geohash, quadtree or a search engine's geo queries) combined with filters, and why businesses, which rarely move, suit a precomputed index.
• Search ranking from distance, rating, review count and text relevance; pagination.
• Read scaling with caches for business pages and popular searches, read replicas, and a CDN for photos.
• The write path for reviews and photos, moderation, and updating ratings and the search index asynchronously.
• Fake review detection signals, and monitoring.

Follow-ups

• Show only businesses open now, across time zones and holidays.
• A business gets thousands of reviews in an hour after going viral.
• Search a dense city center where a small radius holds thousands of results.
• Personalize results for each user.