Researchers propose a new paradigm, FacetCRS, to mitigate filter bubbles in conversational recommender systems. This approach models user preferences into multiple facets, capturing diverse and dynamic user preferences, and demonstrates state-of-the-art performance in benchmark datasets.