How to use variables as factors in collaborative filter

Question

I can find many resources on item-item or user-user similarity or collaborative filtering, but am having a hard time finding or knowing which terms to search for when combining them.

For example, a movie database that has a database of user profiles with features of the user such as location and age. Then a database of user rankings of items such as items purchased or returned. Finally a database of item features such as length of movie time and box office open profit.

So the matrices are: Users x User Features Items x Item Features Users x Item Purchases

First, I am having a hard time understanding how these can be combined. If I do a similarity matrix of the users and items, can these be used as a scaling factor to a collaborative filter based on userxitem purchases?

The second issue, I am not sure what it is called, so can not find it, but essentially I do not want latent factor modeling because I would like to manually weight user or item features. In one output, maybe the user feature of location should be weighted 2x a users age, and an item length of time should be weighted 1.5x the box office profits.

Answer 1

It is not straightforward to combine user and item features in collaborative filtering. You can perhaps draw inspiration from this work. The authors have used collaborative filtering results as one of the features in the next stage which combines collaborative filtering features with other features. This paper was awarded the best paper in ACM RecSys.