The .csv contains important attributes of products, that when you upload to the system, essentially become part of the catalog of products in the system that recommendations will pull from based on the criteria you specify in the activity.
2. There is recommendation -> activities we create the activities add a criteria and select the collection.
3. Add the recommendation metrics.
Here are some definitions that might help your understanding-
"Criteria are rules that determine which products to recommend based on a predetermined set of visitor behaviors. Criteria determine which action will result in which recommendation. You can test multiple recommendation types against each other by adding multiple criteria." (from: Criteria)
Hi tc82, The feed determines what products are available for the recommendations to return. The criteria then determines what algorithm the recommendations needs to use in order to recommend the right products . Collections help to keep your products uploaded via the feed organized.
For example,you could upload a csv feed with 100 products of clothing apparel and then have separate collections as one for women and another for men.
So when you set up the recommendation activity, you can choose to run it just on the men's collection and the criteria will only recommend products from the men's collection. And the criteria can be any of the following as mentioned in Criteria .