Unconventional emergencies usually have the characteristics of complexity, dynamic, and unpredictability, which greatly enhances the difficulty of emergency decision- making. Aiming at the multi-stage large group emergency decision-making problem featuring unknown stage weight and preference information expressed as interval numbers, we propose a new decision-making method. First, we present a similarity measurement formula for interval numbers. Each stage’s preference information is clustered using this simi- larity. To minimize the conflict of preferences, we derived two relative entropy optimization models to calculate the aggregation and stage weights. Next, we rank the alternatives based on the comprehensive group preference information. Finally, we present an illustrative example to verify the valid- ity and practicability of this approach, and discuss several advantages of this method for managing emergency decision- making problems.