Build a Recommender System for a List of Items

Question

So, I am having troubles picturing how to do this. I want to build a recommender system, but rather than doing it on a user/item based recommendations I want to do a list_of_items/item based recommendation. What do I mean with this?

Suppose I have my users make a list of movies they would like to watch. Any user can do the amount of lists they want, and basically what I want to do is to recommend new movies to add to the specific list, based on movies they've already watched and the accept/rejection of the movies I've already suggested them: so if user A has a list 1 of romance movies and a list 2 of action movies I want to recommend new romance movies for list 1 and new action movies for the list 2. Moreover, I have the features of the movies (both the ones in the lists and the ones they already accepted/rejected), like the genre, subgenre, length, year, etc. and I want to use them to have a better insight of the movies I am working with. Besides I want to use the information of the lists of user B, that are similar to the list of user A, to also recommend movies to user A based on the accept/rejection of the user B over some movies I've already gave to him.

The recommender system in this case is a classification: you either add (accept) or not (reject) the movie.

What would you suggest me to read in order to work on this problem?

Answer 1

I believe you can model this problem as a typical recommender system with users and items with one small difference in perspective.

I believe you can treat a list of movies as a user. What I mean is that you can create a matrix of lists of movies and movies in those lists and then apply the usual Recommendation systems algorithms, such as Matrix factorization (SVD) or CF neighborhood methods.

In this setting you can calculate similarities between lists of movies, try to understand what are the latent factors that describe each list, etc.