For the freeze list, as KAJ226 mentioned, it's the core orbital of LiH molecule. In this case it would be the 1$s$ orbital of Li and we only care about the 2$s$, 2$p_x$ orbital of Li, 1$s$ orbital of H, total 6 qubits considering the spin-degeneracy. It is revealed that frozen core approximation does not make any significant error in a quantitative manner.
I must emphasize that the remove list must be chosen with great care. So your question is why remove list is chosen to be [-3, -2]. Well, it's just the tutorial of qiskit and I don't know if you tried it or not but using such remove list on the VQE calculation of BeH2 molecule cause undesirable kink on the energy versus interatomic distance plot.
So the answer is clear. You could test several remove list on your system to choose the best remove list for qubit reduction. It's my hypothesis(not tested) but I think such orbital contained in remove_list would be Li 2$p_x$ orbital, since it is usually unoccupied in a ground state of LiH molecule.(So it does not contribute any bonding of LiH one)
Thankfully, there already is a code in qiskit chemistry tutorial that clearly shows how the choice of remove list affects on the BeH2 molecule VQE calculation. See link:
https://github.com/qiskit-community/qiskit-community-tutorials/blob/master/chemistry/beh2_reductions.ipynb You could easily find that the choice of remove list dramatically contributes to the accuracy of result energy.