Index partitioning through a bipartite graph model for faster similarity search in recommendation systems