whong92 / recommender

Various implementations of recommender systems
0 stars 0 forks source link

Make recommender predict for new user #7

Closed whong92 closed 4 years ago

whong92 commented 5 years ago

Assuming the set of movies stay the same, embedding vector for a new user can be provided by converting it into a linear regression problem

whong92 commented 5 years ago

Given a new user u_j and ratings X_j, we can do: u_j = argmin u_j Loss(u_j * I - X_j)

Where I is the item matrix embedding