Testing recommendation algorithm results
What is the logic behind the “affinity” algo and the “dy automatic” for recommendations?
How can I test that different users are indeed receiving different experiences?
By default, the system lookup for the user viewed categories and scoring them accordingly. Let’s say my browsing behavior in a given session was
‘Men’ --> ‘Men’|‘T-shirts’
‘Men’ --> ‘Men’|‘underwear’
Men will get a 2X score and t-shirt & underwear will get a X score. Affinity widget will have a men flavor and will show some products, among others, from ‘underwear’ & ‘t-shirt’. In the long run, the user gender will be captured and the widgets will show the categories with the highest affinity.
You can add more columns of your feed and user interactions to be looked into by DY for affinity calculation.
Depends on the context you are in. Will show a mixture of best sellers, affinity & similar products.
For more info read https://support.dynamicyield.com/hc/en-us/articles/360022554694-Recommendation-Strategies#h_211dc1c3-abee-4b51-b3a0-dc5ba4298583Comment actions
Please sign in to leave a comment.