In group decision-making, ignoring the existence of uncertain factors causes the decision-making problem to lose its practical significance. Based on the maximum expert consensus model (MECM), we considered the uncertainty of the opinions of the three participating roles by introducing noncooperators. Additionally, three different opinion uncertainty sets were constructed to describe the characteristics of opinion uncertainty more accurately. Furthermore, by applying a robust optimization (RO) method to process uncertain sets, we propose mixed-integer robust MECMs, which reduce the risk of uncertain opinions to decision-makers (DMs). Moreover, numerical experiments used in the passenger satisfaction survey of the Shanghai Metro verified the validity of the models proposed in this article. The characteristics of the models were revealed through sensitivity analysis. Finally, to overcome the relatively highly conservative results of the classic RO method, we construct data-driven opinion uncertainty sets and propose data-driven RO models. Hence, DMs with different risk preferences can choose RO models with different risk levels according to the situation.