What gets recommended
Commerce recommendations answer questions like what to buy next, what pairs with this, and what similar shoppers bought. Each surface has different intent and different signals.
Common approaches
- Collaborative filtering: shoppers who bought this also bought that.
- Content based: items similar in attributes to ones the buyer liked.
- Complementary: items frequently bought together, useful at cart.
- Hybrid: blend several to cover cold start and sparse data.
System concerns
- Split into an offline model build and an online low latency serving layer.
- Filter out of stock, already owned, and policy violating items before showing.
- Measure with conversion and revenue, not just clicks, to avoid optimizing for bait.
Key idea
Commerce recommendations blend collaborative, content, and complementary signals, built offline and served online, with stock and ownership filters and revenue based evaluation.