Recommendation Models
ML model types for similar items, frequently bought together, and recently viewed
What it solves
Different recommendation surfaces need different signals. A product detail page needs similar items — products close to what the shopper is looking at right now. A homepage or post-purchase page needs broader signals like best sellers or trending products. A cart page benefits from frequently-bought-together logic.
MXP's recommendation models are pre-configured for these use cases. Each model type is trained on a different combination of behavioral signals and can be independently tuned and evaluated before serving.
When to use it
- Setting up a new recommendation surface — select the model type appropriate for the page context and create a new model
- Reviewing model training status — check whether a model has been trained and is ready to serve
- Tuning optimization objectives — configure whether a model should optimize for click-through rate, conversion, or revenue
Key concepts
Model type — determines what signals the model uses and what it recommends:
| Model type | What it recommends |
|---|---|
| Similar Items | Products similar to a given item based on attributes and co-view behavioral signals |
| Recommended For You | Personalized recommendations based on the current user's session behavior |
| Best Sellers | Top-performing products by purchase count or revenue in a given category |
Optimization objective — what the model is trained to maximize. Options include Click Through Rate, Conversion, and Revenue.
Training status — whether the model has been trained and is ready to serve real-time inference. A model with Paused training status will not update with new behavioral data.
Ready to Query / Data State — flags indicating whether the model has sufficient data to serve recommendations and whether its training data is current.
How it works
Open Recommendations → Generic Models (also labelled Recommendation Models in some views) from the left sidebar. Each row shows:
- Model name
- Ready to Query and Data State indicators
- Model type (Similar Items, Recommended For You, Best Sellers)
- Optimization objective
- Training schedule and tuning status
- When it was last trained
To create a new model, click + New Model and select the model type and optimization objective. After creation, the model needs to complete a training run before it can serve recommendations. Assign it to a Recommendation Container to activate it for a placement.
Quick example
A team sets up a "Similar Items" model for their product detail page. They create the model, set the optimization objective to Conversion, and wait for the first training run to complete. Once the Data State turns ready, they assign it to the pdp-similar container. The PDP now serves model-driven recommendations without any code changes.
Related pages
- Recommendation Containers — assign models to page placements
- Recommendation Rules — override or filter model output with business rules
- Recommendations overview — how the full system fits together