Evaluating Large Language Models as Generative User Simulators for Conversational Recommendation

Large language models can be effective user simulators for conversational recommendation, but they may exhibit deviations from human behavior that can be reduced with model selection and prompting strategies.