Collaborative filtering is the workhorse of modern recommendation systems and relies on user interaction in the ecosystem rather than profiles to make recommendations.
This technique relies on past user behavior and product ratings and does not assume any pre-existing knowledge. In short, users rate the inventory items and the assumption is that customer taste will remain relatively constant over time, which can be exploited to provide recommendations. Having said that, an intelligent system will augment and reorder recommendations with any available context (for example, the user is a female who has logged in from China).
The main issue with this class of techniques is cold start, but its advantages of being domain ...
