The multiple input multiple output non-orthogonal multiple access (MIMO-NOMA) is considered as a promising technology to improve the performance of the next-generation mobile network. In this paper, we propose a hybrid precoding massive MIMO-NOMA system, where the spectrum efficiency is improved with the proposed user grouping scheme and power allocation scheme. In particular, user grouping guarantees that the cluster headers have relatively high channel gains and low channel correlations among each other. With user grouping, power allocation for the studied system is based on an adaptive continuous Cross-Entropy (CE) approach. To achieve a higher sum data rate of entire network, the number of elite samples is chosen depending on the optimal ratio to the total number of samples used in each iteration. The optimal ratio is numerically validated in the evaluation. Moreover, a decreasing total sample number along with standard deviation is proposed to further increase the convergence rate of the CE approach. The simulation results demonstrate that our proposed power allocation scheme can provide significantly higher spectrum efficiency than traditional MIMO systems, and outperform existing MIMO-NOMA systems.