This dataset accompanies the paper "Who's Asking? Simulating Role-Based Questions for Conversational AI Evaluation" (ACL Findings 2026).
CoRUS (Community-driven Roles for User-centric Question Simulation) is a framework that embeds role-based context implicitly into a question for user-centric evaluation of conversational AI. Most evaluations capture a model's capability to respond but ignore who is asking, yet… See the full description on the dataset page:
https://huggingface.co/datasets/navreeetkaur/CoRUS.
