Say we want to build a video recommender, how do we ensure diversity of videos being recommended ? What sort of algorithm or loss function should we use to do it ?
Hi Titas. It is a good question. It would seem like the first step is to define diversity. While you might want to include gender diversity (male vs female creators) you might not want to include veracity diversity (that is, you don't want to mix things you know to be true with things you know to be untrue). So, defining which elements need diversity is key. The HR literature on diversity includes things like this:
Physical / identity
* Age, race, disability, HIV status, skin color, pregnancy status, gender, marital status
Cognitive / intellectual
* Concrete vs abstract thinking, risk aversion, long vs short time horizon
Related attributes
* Geography, industry, social class, years of education
Some span multiple categories
* Religion / faith, political orientation, sexual orientation
However, this is for HR and not for a video recommender. So, to really answer your question, we would need to know more about what kind of videos are these.
- Badges
- Science topic
- Similar topics
- Computer Communications (Networks)
- Simulators