I have a database where each row is a user, and it has two types of columns: features of the user and genre grading from 0 to 6. For example:
| User | Age | Sex | ... | Genre 1 | Genre 2 | ... |
|---|---|---|---|---|---|---|
| 1 | 18 | M | 4 | 5 | ||
| 2 | 75 | F | 3 | 2 | ||
| 3 | 45 | M | 0 | 1 |
Let us say then that we have features F_1,F_2,...,F_m and Genres grading G_1, G_2,..., G_d where each G_i = 0,1,...,5. The basic idea is to build a function of the sorts F(F_1,F_2,...,F_m) = (G_1,G_2,...,G_d). This is, given the characteristics of the user, I want to estimate the genre scores in order to recommend the highest ranked genre, then the next one, and so on.
For example, given the fit of F, we could have only age and gender as features, and genres romance, action and historical. Thus, the idea is to train the function and, given a new user who is a 35 year old male, give the estimate F(35, male) = (1,3,2), where 1 is the rating for romantic, 3 for action, and 2 for historical. Thus, we would recommend action movies first.
There are no missing values, so no worries about that.
I thought of using multivariate multinomial regression, but I couldn't make it work in Python. I would like any pointers to tackle this problem, specially if it can be implemented in Python. I am not sure where to start, nor what algorithms I should use. I have seen collaborative filtering, but that seems to be related to how the user interacts with the system, not the user characteristics themselves. I believe my problem is much simplier.