Similar recipes shelf
Your task
Design the "more like this" shelf shown under each recipe on a cooking site. The site has 80,000 recipes, adds about 200 a day, and serves 3 million recipe page views a day. Each shelf shows 8 similar recipes. The data team can produce a text embedding for each recipe from its title, ingredients and steps, and has a year of co-view data (people who viewed this also viewed that). The shelf must not slow down the page, must never show a recipe that was taken down, and a new recipe should get a shelf on the day it is published. Describe how similar recipes are computed, stored and served.
It covers
- Precomputed neighbors
- Cold start
- Embeddings