Non-Orthogonal Multiple Access(NOMA) is introduced into massive Multiple Input Multiple Output(MIMO) system in order to further improve the spectrum efficiency. Nevertheless, when the channel between the base station and the user is poor, it is difficult to guarantee the spectrum efficiency of the user. A massive MIMO-NOMA system based on Intelligent Reflecting Surface(IRS) is discussed to deal with this problem. An IRS-assisted massive MIMO-NOMA system model is considered firstly. Then according to the cluster head selection algorithm, a cluster head is selected for each beam to design analog precoding. Users are grouped according to the correlation of equivalent channels, and the user with the strongest equivalent channel gain in each beam is selected to design digital precoding. Finally, the power allocation of the base station and the phase shift matrix of the IRS are jointly optimized to ensure the fairness of the user’s spectral efficiency, that is, to maximize the minimum spectral efficiency. The simulation results show that compared with the massive MIMO-NOMA system without IRS, the IRS-assisted massive MIMO-NOMA system can significantly improve the spectrum efficiency of the weakest user.