Personalised product recommendation engines provide recommendations to each contact based on a taste profile which has been calculated for that contact.
The taste profile is calculated based on the weight of product attributes that a contact has interacted with.
E.g. If I have shown interested in products with the attributes ‘women’s clothing', ‘dresses’ and 'floral’, my taste profile might recommend me floral dresses.
This is called attribute affinity, which you can check for individual contacts in the contact details screen:
The weight of an attribute is calculated by multiplying the attribute frequency by the attribute entropy:
- Attribute frequency - the measure of the number of times that the contact has interacted with the attribute - either ordered a product with that attribute or viewed a product with that attribute. Orders have a higher score than views.
- Attribute entropy - If a product attribute is applied to a large number of products it doesn’t tell us as much about the contact’s affinity for the product itself, so it is given a lower weighting than an attribute which applies to fewer products. Low entropy can lead to attributes being excluded from the customer’s taste profile.
The customer's taste profile comprises the top 10 highest scoring attributes based on these factors, and the personalised recommendation engine recommends products with these attributes.