Polarization analysis in Raman spectroscopy provides a powerful tool in chemical physics to identify the symmetry of the vibration modes of molecules, macromolecules and crystals, according to their crystallographic structure and local spatial orientation. In this paper we propose two new approaches in polarized Raman spectroscopy for data with rotational and spatial diversity, respectively. We show that the joint use of parallel and crossed polarization data yields more accurate source separation results and improves the uniqueness properties of the solution in blind nonnegative source separation algorithms. The proposed approaches are validated on two real polarized Raman data sets.