The Frequently Bought Together model looks at users' conversion patterns to recommend items that were either bought together or that might go well together.
Model Variants
It worth noting that our FBT model has two settings which you can choose between:
Relaxed variant (default)
The relaxed variant recommends items that are likely bought together. It uses collaborative filtering to infer relationships from past purchases. For example:
- If users often buy product A with product B, and product B with product C
- The model can suggest product C for product A, even if those two items weren’t bought together.
This approach helps increase catalogue discovery, but can introduce inferred (less direct) pairings.
Strict variant
The strict variant only recommends items that were converted together in the past. It doesn’t infer new pairings, so it typically returns fewer recommendations.
Impact of Events
The quality of recommendations depends on the conversion events used to train the model.
If you are sending an addToCart conversion events, this can introduce some noise. For example, if a user adds an item to cart but does not complete the purchase.
To reduce this "noise", (i.e. only shows products that were purchased together as opposed to added to cart together), you can use the event customization feature to train the model using only purchase events. This should result in the model being more relevant.