For online retailers with large product catalogs, or publishers with large content feeds, affinity-based recommendation is a powerful method to make compelling product recommendations when and to whom they matter most. Affinity-based algorithms gather behavioral data on consumers which are used to develop rich, multi-dimensional affinity profiles for every individual site user. The site then recommends based on the user’s unique preferences and meaningful actions taken on the site.
Affinity profiling centers on creating weighted scores for users. These scores are calculated and assigned by measuring the correlation between consumer interactions and the item attributes with which they interact with most meaningfully. Meaningful interactions for eCommerce may include purchases, views or adds-to-cart; For publishers, these may include categories, tags, and keywords. The more interactions a consumer has with products or content that share a particular attribute, the better the recommendations engine can assign a level of interest.
To illustrate, for publishers, an affinity profile can reveal a user’s most preferred content categories (basketball) and tags (NBA).
To illustrate for eCommerce, Jane’s affinity profile reveals her most preferred product categories (women, shoes) and colors (red, black, blue).
At first glance, it seems that red is Jane’s preferred color attribute value, and “women” her most preferred category attribute value. Put another way, Jane likes to look at red clothes for women. However, once we rank Jane’s attribute values with a hierarchy of her most meaningful interactions– viewed, added to cart, purchased – a different story emerges. We see that Jane has never actually purchased any items in red and black is her preferred color.
Beyond that, the recommendations engine can look at brand, size, and other attributes Jane has interacted with to present Jane with the products most likely to prompt an engagement.