What is the correct strategy to deal with multiple (implicit or explicit) feedback values in a retrieval and ranking model?

If you have multiple implicit or explicit inputs to a recommender what is the correct strategy to deal with these. Do you:

A. Implement different recommender systems that predict ratings for each user input that you iterate and improve upon; or,
B. Create a composite value made up of your different inputs and have a single recommender that you iterate and improve upon.
C. Something else?

Any guidance appreciated.