Recommendation Filters

When setting your recommendation strategy, you can add dynamic filtering rules to customize the algorithm. These filters are based on inclusion/exclusion rules in addition to the algorithm chosen.

Filters based on the page type properties

Category filters

Category filters enable you to show only items from the same category or parent category as the currently-viewed product page or category page. For example, use this filter if the user is viewing a shirt and you want recommendations to only show shirts.

Filter logic: Category vs. Parent category

• Only include items that match the viewed item's category:
  Returns only products that match the full category path. It does not "loosen" the filter if no products are available that match the complete path.
  Example: If the current product is Men | Shoes | Sneakers, only men's sneakers are recommended, and if none are available to recommend, no other products are shown.
• Include items that match the viewed item's parent category:
  Returns products that match the next level up in the path, and then the level above that. 
  Example: If the current product is Men | Shoes | Sneakers, the recommendation filter will begin with men's shoes, and if it can't fill the slots with men's shoes, it will return other men's products. 
  • Note that the first category in the tree is never ignored, even if it means serving empty slots. So the filter in this example stops at Men. 

Product property filters

Product property filters ensure that recommendations match or do not match the currently viewed item's properties, using any attribute of the feed. For example, use this filter to show items of the same brand or color.

Note: If the Categories filter is used, it loosens the restrictions (see the Keywords example in Filter Logic). We advise using this filter sparsely and using the designated category filter instead.

Filter logic

• Exclude/Include items that match viewed item properties such as:

  Brand | Color | Gender | Keywords
  • Brand | Color | Gender:
    Returns products that match all item properties, but loosens the restrictions if no products are available that match all of the properties. 
    Example:
    • Recommends: Brand | Color | Gender
    • If there are not enough products to complete the widget, the recommendation logic loosens the last property – Gender – and recommends items that match Brand | Color.
    • If there are still no products to show, it ignores the Color attribute and fills the widget with Brand products. The first category in the tree is never ignored, even if it means serving empty slots.
  • Keywords:
    Returns products that match all item keywords, but loosens the restrictions if no products are available that match all keywords. 
    Example:
    • Recommends: Material | Pattern | Size
    • Loosens the restrictions in the same way as the previous example, leaving only Material, which is never ignored.

Filters based on past user behavior

User behavior filters ensure that visitors are not being recommended items that they have recently viewed or purchased, online or offline. To exclude offline purchases, you need to identify them.

Filter logic

The filter does not loosen if no matching products are available, and in this case empty slots are displayed.