Accurate measurement of regional efficiency is a prerequisite for effective management. Prior studies have expanded on the overall "black box" evaluation with two stages of research and development (R&D) and commercialization, opening up the internal structure of the
regional innovation process, but ignoring the independent innovation activities of universities, research institutes, and firms in the R&D stage. We construct a mixed structure with two stages, three actors, and four subsystems, and conduct an empirical analysis of China's provincial samples
from 2017 to 2019 by using the network data envelopment analysis (DEA) model. Results show that the efficiency of the R&D stage at the provincial level is generally higher than that of the commercialization stage. However, the three subsystems of the R&D stage perform poorly. Spearman’s
rank correlation coefficient suggests that there is a significantly positive correlation between total regional efficiency and commercialization. In addition, we use the k-means method to divide 27 provinces into three clusters, setting a more appropriate improvement benchmark for inefficient
provinces. Based on enlightenment of regional tobacco industry, we put forward some proposals for specific stage and specific subsystem.