Supervised MF : Using user and item metadata

Modified MF formulation following BCS framework and user and item metadata incorporated

% This DEMO is for modified Matrix factorization (MF) formulation for Recommender System
% following blind compressive sensing (BCS) framework with user and item metadata
% OPTIMIZATION PROBLEM
% minimize_(U,V, A, C) ||Y-M(UV)||_F + lambda_u||U||_F + lambda_v||V||_1 + mu_u||W-UC||_F + mu_v||Q-AV||_F

We incorporate user and item metadata into the base BCS MF formulation as class label matrices. Users and items are grouped into multiple classes based on available secondary information. This classification is used to generate class label matrices Q and W. Latent factor matrices are recovered consistent with the class label information.