Question 7
7. [Recommender Systems] Consider a dataset containing information about movies: genre, director and release decade. We also have information about which users have seen each movie. The rating for a user on a movie is either 0 or 1.
Here is a summary of the database:
Consider user U1 is interested in the time period 2000s, the director D2 and the genre Humor. We have some existing recommender system R that recommended the movie B to user U1. The recommender system R could be one or more of the following options:
- User-user collaborative filtering.
- Item-item collaborative filtering.
- Content-based recommender system.
(a) Given the above dataset, which one(s) do you think R could be? (If more than one option is possible, you need to state them all.) Explain your answer. (6 Marks)
(b) If some user U2 wants to watch a movie, under what conditions can our recommender system R recommend U2 a movie? If R recommends a movie, how to do it? If R cannot recommend a movie, please explain
why it cannot be recommended. State any additional information R might want from U2 for predicting a movie for this user, if required. (10 Marks)
(c) Item-item collaborative filtering is seen to work better than user-user because users have multiple tastes. But this also means that users like to be recommended a variety of movies. Given the genre of each movie (there are 2 different genres in the dataset) and an item-item collaborative filtering recommender system that predicts k top-movies to a user (k can be an input to the recommender), suggest at least three ways to find top 5 movies to a user such that the recommender will try to incorporate movies from different genres as well. (Note: Explain in 3–5 lines maximum, no rigorous proof is required.) (12 Marks)
