You may notice that the "Customers who viewed this item also viewed" model recommends products from a category or segment you didn't expect. For example, a customer who has mostly browsed or purchased women's products may still receive recommendations for men's products.
This article explains why this happens, how the model builds its recommendations, and how Customer Preferences can help make the results more relevant for each customer.
How the model works
This model examines which products users have historically viewed together. It learns relationships between items from shared browsing patterns. This means the model isn't focused solely on a single customer's history. Instead, it examines the broader behavioral relationships among products across the project.
Why unexpected categories can appear
Because shared browsing patterns drive the model, the output may include products outside the customer's apparent category preference. If women's and men's items were historically viewed together often enough, the model can still return men's products for a customer whose own history looks mostly women-focused.
This can also happen because customer behavior is often broader than it seems at first glance. Even if one segment is dominant, the customer may still have interacted with other categories, such as unisex, children's, or adjacent product groups, which can widen the behavioral signal the model works with.
How the Customer Preferences feature helps
If you want to make the output feel more aligned with the customer's interests, Customer Preferences can help improve the relevance of the results.
Customer Preferences works as a reordering layer. It doesn't rebuild the recommendation from scratch, and it doesn't strictly exclude items that don't match the preference. However, it moves matching items higher in the list for that customer.
For example, if the recommendation returns a mixed set of products and the customer prefers women's products, Customer Preferences can push women's items to the top of the recommendation.