To spread positive ideology and reflect a good social atmosphere, all walks of life launch some benchmark selection activities, such as “the most beautiful doctor”, “the most beautiful teacher” and so on. These role models (or benchmarks) tend to be a minority among numerous candidates. However, traditional methods are difficult to evaluate the performance of tens of thousands of candidates in a large sample, and the learning relationship and specific advantages between candidates are not considered in these methods. Thus, it is difficult to realize benchmark identification by traditional methods. To reduce the workload of benchmark identification, this study proposes a social network data envelopment analysis method for identifying benchmarks from a large-scale sample, by combining the data envelopment analysis (DEA) and social network analysis (SNA). First, a pairwise evaluation process is constructed through the DEA method to explore the efficiency state of each decision making unit (DMU) relative to another DMU, and a pairwise evaluation matrix that can reflect the reference relationship between any two DMUs is obtained. Then, a social network is built based on this matrix, and role-model DMUs in a large sample are identified by comparing the in-degree centrality values of all DMUs. Next, analyze the specific advantages of the benchmarks selected according to the meaning of weight in DEA model. Finally, an experiment is carried out on the online diagnosis data of 10,418 doctors in 55 departments on the Chunyu doctor platform to verify the social network data envelopment analysis method.